JP7353330B2 - Information processing device, information processing method, and information processing program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, and an information processing program.

Conventionally, techniques for searching (searching) various types of information have been provided. For example, there is a technology that generates a graph in which nodes corresponding to a search target are connected by edges, and performs a search using the generated graph. Further, such a technique is used for, for example, image search.

Patent No. 6293335 Patent No. 6300982 Patent No. 6311000

Masajiro Iwasaki "Neighborhood search using approximate k-nearest neighbor graph using tree-structured index", Information Processing Society of Japan Transactions, 2011/2, Vol. 52, No. 2. pp.817-828.

However, with the above-mentioned conventional technology, there is room for improvement in terms of improving the efficiency of search processing. For example, in the above-mentioned conventional technology, by adjusting the number of edges of the graph and changing the graph to one that allows efficient searching, efficiency is achieved by reducing search processing time. There is a limit to how much efficiency can be achieved in search processing through changes. Therefore, it is desired to generate information that can be used in search processing in order to perform efficient search processing, such as speeding up search processing, for example.

The present application has been made in view of the above, and aims to provide an information processing device, an information processing method, and an information processing program that generate information that can be used for search processing.

The information processing device according to the present application includes an acquisition unit that acquires first information indicating a plurality of objects to be data searched and second information used for classifying the plurality of objects; a third group indicating a plurality of groups used for classifying the plurality of objects indicated by the first information using the second information and performing batch processing in a search process targeting the plurality of objects; A generation unit that generates information.

According to one aspect of the embodiment, it is possible to generate information that can be used in search processing.

FIG. 1 is a diagram illustrating an example of information processing according to the first embodiment. FIG. 2 is a diagram showing an example of data according to the first embodiment. FIG. 3 is a diagram showing an example of data according to the first embodiment. FIG. 4 is a diagram illustrating a configuration example of an information processing system according to the first embodiment. FIG. 5 is a diagram illustrating a configuration example of an information processing apparatus according to the first embodiment. FIG. 6 is a diagram illustrating an example of an object information storage unit according to the first embodiment. FIG. 7 is a diagram illustrating an example of a graph information storage unit according to the first embodiment. FIG. 8 is a diagram illustrating an example of a quantization information storage unit according to the first embodiment. FIG. 9 is a diagram illustrating an example of a codebook information storage unit according to the first embodiment. FIG. 10 is a flowchart illustrating an example of information processing according to the first embodiment. FIG. 11 is a flowchart illustrating an example of search processing according to the first embodiment. FIG. 12 is a diagram illustrating an example of information processing according to the second embodiment. FIG. 13 is a diagram illustrating a configuration example of an information processing apparatus according to the second embodiment. FIG. 14 is a diagram illustrating an example of a quantization information storage unit according to the second embodiment. FIG. 15 is a diagram illustrating an example of a blob information storage unit according to the second embodiment. FIG. 16 is a flowchart illustrating an example of search processing according to the second embodiment. FIG. 17 is a diagram illustrating an example of information processing according to a modification. FIG. 18 is a diagram illustrating an example of a graph information storage unit according to a modification. FIG. 19 is a flowchart illustrating an example of a search process according to a modification. FIG. 20 is a diagram illustrating another example of information processing according to the modification. FIG. 21 is a diagram illustrating an example of blob information generation processing. FIG. 22 is a flowchart illustrating an example of information processing according to a modification. FIG. 23 is a diagram illustrating an example of information processing according to the third embodiment. FIG. 24 is a diagram illustrating a configuration example of an information processing apparatus according to the third embodiment. FIG. 25 is a diagram illustrating an example of a blob connection graph information storage unit according to the third embodiment. FIG. 26 is a flowchart illustrating an example of information processing according to the third embodiment. FIG. 27 is a flowchart illustrating an example of search processing according to the third embodiment. FIG. 28 is a flowchart illustrating an example of search processing according to the third embodiment. FIG. 29 is a diagram illustrating an example of blob connection graph information according to a modification. FIG. 30 is a diagram illustrating an example of a blob information storage unit according to a modification. FIG. 31 is a diagram illustrating an example of a blob connection graph information storage unit according to a modification. FIG. 32 is a hardware configuration diagram showing an example of a computer that implements the functions of the information processing device.

DESCRIPTION OF THE PREFERRED EMBODIMENTS An information processing apparatus, an information processing method, and an embodiment of an information processing program (hereinafter referred to as "embodiments") according to the present application will be described in detail below with reference to the drawings. Note that the information processing apparatus, information processing method, and information processing program according to the present application are not limited to this embodiment. Further, in each of the embodiments below, the same parts are given the same reference numerals, and redundant explanations will be omitted.

(Embodiment)
[1. First embodiment]
[1-1. Information processing〕
An example of information processing according to the first embodiment will be described using FIG. 1. FIG. 1 is a diagram illustrating an example of information processing according to the first embodiment. The information processing apparatus 100 executes a search process using a graph index (also simply referred to as a "graph") that is a graph structure of a plurality of objects that are data search targets. FIG. 1 shows part of a search process when the information processing device 100 performs a neighborhood search using a graph in which nodes corresponding to vectors of the object to be searched are connected by edges. . The information processing apparatus 100 uses each node of the graph as an object to be searched, and searches for nodes near a given search query (vector) by tracing the graph. Through the search process, the information processing apparatus 100 extracts a predetermined number of nodes (hereinafter also referred to as "search number") designated as the number of nearby nodes to be extracted as nodes in the vicinity of the search query. In the following, a case where image information is the object of data search will be described as an example, but the object of data search may be various objects such as video information and audio information.

The information processing device 100 performs search processing for nodes corresponding to a huge amount of image information in units of millions to hundreds of millions, but in the drawing, only a portion of them (several tens of nodes such as node N1 in FIG. 1) are shown. ) are shown. For example, the information processing apparatus 100 acquires information regarding a plurality of nodes (vectors) such as nodes N1, N7, N9, etc., as shown in spatial information SP1 in FIG. When "node N* (* is an arbitrary numerical value)" is written in this way, it indicates that the node is identified by the node ID "N*". For example, when "node N1" is written, the node is identified by the node ID "N1". Each white circle (◯) in the spatial information SP1 indicates each node.

In the spatial information SP1 in FIG. 1, codes are mainly attached to nodes related to the explanation, but each white circle (○) without a code is also a node, and many nodes are included in addition to the nodes shown. It will be done. Further, each node corresponds to each object (search target). For example, each of the plurality of local features extracted from the image may be an object. Further, for example, various data in which distances between objects are defined may be objects.

For example, the spatial information SP1 in FIG. 1 may be a Euclidean space. For example, it is assumed that the spatial information SP1 corresponds to the number of dimensions of the vector of the object, and is a multidimensional space such as 100 dimensions or 1000 dimensions. Note that FIG. 1 conceptually illustrates a state in which a vector (space) is divided into four subregions (subspaces) by direct product quantization as shown in spatial information SP1, but regarding the point of direct product quantization, will be described later.

A dotted line connecting white circles (◯) that are nodes in the spatial information SP1 indicates an edge connecting the nodes. In the example of FIG. 1, to simplify the explanation, a case is shown in which nodes are connected by undirected (bidirectional) edges (hereinafter also simply referred to as "edges"). Note that the undirected edge here means an edge that allows data to be traced in both directions between connected nodes. For example, the white circle (◯) indicating the node N1 and the white circle (◯) indicating the node N4 are connected by a dotted line, indicating that the nodes N1 and N4 are connected by an edge. That is, in the graph shown in the spatial information SP1, it is possible to trace the path between the node N1 and the node N4 in both directions. Specifically, the spatial information SP1 can be traced from the node N1 to the node N4, and from the node N4 to the node N1.

Note that in the example of FIG. 1, for convenience of illustration, only edges connecting the illustrated nodes are illustrated, but many edges other than the illustrated edges are included. In this way, the spatial information SP1 in FIG. 1 shows only a portion of the edges, but it is assumed that the spatial information SP1 in FIG. 1 is, for example, a k-nearest neighbor graph. Note that the spatial information SP1 may be various types of graphs.

Furthermore, the edges of the graph are not limited to undirected edges, but may be directed edges. In the case of a directed edge, it is possible to trace only from the node that is the reference source of the directed edge to the reference destination node. For example, if two nodes are connected by two directed edges: a first edge where one is the reference source and the other is the reference destination, and a second edge where one is the reference destination and the other is the reference source. , the two nodes are connected by an undirected (bidirectional) edge and have the same state.

The search process will now be described using FIG. 1. In the example of FIG. 1, it is assumed that the information processing apparatus 100 has already acquired a graph GR1 as shown in the spatial information SP1. Note that the information processing apparatus 100 may generate the graph GR1 using various conventional techniques as appropriate. First, prior to explaining the search process, vector (space) product quantization will be explained.

FIG. 1 conceptually illustrates a state in which a vector (space) is divided into four subspaces AR11 to AR14 by direct product quantization as shown in spatial information SP1. The subspaces AR11 to AR14 shown in FIG. 1 are shown as conceptual diagrams for explaining distances between vectors of nodes (objects), etc., but each of the subspaces AR11 to AR14 is a multidimensional space. For example, the subspace AR11 shown in FIG. 1 is illustrated in a two-dimensional manner because it is illustrated on a plane, but it is assumed that it is a multidimensional space, such as a 100-dimensional or 1000-dimensional space.

Each of the partial spaces AR11 to AR14 indicates a space corresponding to each division (partial vector) of the vector divided into four. Note that, for example, a 100-dimensional vector may be divided into 100 parts. For example, the subspace AR11 is a dimension space (" (also referred to as "first subspace"). Subspace AR11 indicates a subspace (first subspace) corresponding to the first division position of the vector. In the case of search query QE1, subspace AR11 is a space corresponding to first subquery QE1-1, which is the first subvector (first subvector).

Furthermore, the subspace AR12 is a dimensional space ("second subspace") corresponding to the second subvector from the beginning (also referred to as "second subvector") among the four subvectors obtained by dividing the vector into four. ). Subspace AR12 indicates a subspace (second subspace) corresponding to the second division position from the beginning of the vector. In the case of search query QE1, subspace AR12 is a space corresponding to second subquery QE1-2, which is the second subvector (second subvector) from the beginning.

Further, the subspace AR13 is a dimensional space ("third subspace") corresponding to the third subvector from the beginning (also referred to as "third subvector") among the four subvectors obtained by dividing the vector into four. ). Subspace AR13 indicates a subspace (third subspace) corresponding to the third division position from the beginning of the vector. In the case of search query QE1, subspace AR13 is a space corresponding to third partial query QE1-3, which is the third partial vector (third subvector) from the beginning.

Furthermore, the subspace AR14 is a dimension space (" 4th subspace). Subspace AR14 indicates a subspace (fourth subspace) corresponding to the fourth division position from the beginning of the vector. In the case of search query QE1, subspace AR14 is a space corresponding to fourth subquery QE1-4, which is the fourth subvector (fourth subvector) from the beginning.

In the example of FIG. 1, the subspaces AR11 to AR14 are shown with similar shapes, but the shapes of the subspaces AR11 to AR14 may be different, and the manner in which the regions in each of the subspaces AR11 to AR14 are divided may also be different. It's okay. Further, it is assumed that the connection relationship between nodes by edges is common to subspaces AR11 to AR14. That is, the graph GR1 is a graph corresponding to the spatial information SP1, and is common to the subspaces AR11 to AR14.

Furthermore, in the example of FIG. 1, a lookup table is generated for each divided partial vector (subvector). For example, clustering is performed for each of four subvectors: a first subvector, a second subvector, a third subvector, and a fourth subvector, and a lookup table (codebook information) is generated individually.

In FIG. 1, the first subvectors are clustered into nine groups corresponding to codebooks CD11 to CD19, as shown in subspace AR11, and the vectors corresponding to each of codebooks CD11 to CD19 are representative vectors ( centroid). For example, the first subvectors such as nodes N7 and N9 indicate that they are clustered into a group corresponding to codebook CD11. In this case, the first subvectors such as nodes N7 and N9 use the codebook information (also referred to as "first codebook information") corresponding to the first subvector in the distance calculation (calculation). is vector quantized into a vector.

Furthermore, the second sub-vectors are clustered into a plurality of groups corresponding to codebooks CD21 to CD24, etc. (see FIGS. 2 and 9), and the vectors corresponding to each of codebooks CD21 to CD24, etc. are representative vectors (centre Lloyd's). In this case, the second subvector of each node is calculated using the codebook information (also referred to as "second codebook information") corresponding to the second subvector in distance calculation. vector quantized.

The third sub-vectors are clustered into multiple groups corresponding to codebooks CD31 to CD34, etc. (see FIGS. 2 and 9), and the vectors corresponding to each of codebooks CD31 to CD34, etc. are representative vectors (centroids). It is calculated as In this case, the third subvector of each node is calculated using the codebook information (also referred to as "third codebook information") corresponding to the third subvector in distance calculation. vector quantized.

The fourth sub-vectors are clustered into multiple groups corresponding to codebooks CD41 to CD44, etc. (see FIGS. 2 and 9), and the vectors corresponding to each of codebooks CD24 to CD44, etc. are representative vectors (centroids). It is calculated as In this case, the fourth subvector of each node is calculated using the codebook information (also referred to as "fourth codebook information") corresponding to the fourth subvector in distance calculation. vector quantized.

Note that the codebook information, such as the process of determining the representative vector, is generated using various techniques as appropriate. Since generation of codebook information is a conventional technique, detailed explanation will be omitted. Further, the lookup table is not limited to the above, and for example, one lookup table (codebook information) may be used for all divided partial vectors (subvectors), but this point will be described later.

From here, a search process targeting the search query QE1 will be explained. First, the information processing device 100 obtains the search query QE1 (step S11). For example, the information processing device 100 acquires the search query QE1 from the terminal device 10 (see FIG. 4) used by the user.

Then, the information processing device 100 divides the vector that is the search query QE1 into four. That is, the information processing device 100 divides the search query QE1 into four partial queries. In FIG. 1, the information processing device 100 divides the search query QE1 into a first partial query QE1-1 which is a first subvector, a second partial query QE1-2 which is a second subvector, and a second partial query QE1-2 which is a third subvector. It is divided into four subvectors: 3 partial queries QE1-3 and a fourth partial query QE1-4, which is a fourth subvector. Specifically, the information processing device 100 sets "45, 23, 2..." as a first partial query QE1-1, "127, 34, 5..." as a second partial query QE1-2, and sets "20, 98, 110...'' is the third partial query QE1-3, and ``12, 45, 4...'' is the fourth partial query QE1-4.

Then, the information processing device 100 calculates the distance between each subvector of the search query QE1 and the vector of the codebook corresponding to that subvector. The information processing device 100 calculates the distance (difference) between each subvector of the search query QE1 and the vector of each codebook, as shown in codebook information TB1 to TB4 in FIG. 2. FIG. 2 is a diagram showing an example of data according to the first embodiment.

The codebook information TB1 indicates first codebook information corresponding to the first subvector, and the information processing apparatus 100 stores each vector of the codebooks CD11 to CD19 corresponding to the first subvector and the first partial query QE1. -1. In FIG. 2, the information processing device 100 calculates the distance between the codebook CD11 and the first partial query QE1-1 as a distance DS11, as shown in codebook information TB1. Similarly, the information processing device 100 calculates the distances between each of the codebooks CD12 to CD14, etc. and the first partial query QE1-1 as distances DS12 to DS14, etc.

Similarly, each of codebook information TB2 to TB4 indicates second to fourth codebook information corresponding to each of the second to fourth subvectors. The information processing device 100 calculates the distances between each of the codebooks CD21 to CD24, etc. and the second partial query QE1-2 as distances DS21 to DS24, etc., as shown in the codebook information TB2. The information processing device 100 calculates the distances between each of the codebooks CD31 to CD34, etc. and the third partial query QE1-3 as distances DS31 to DS34, etc., as shown in the codebook information TB3. The information processing device 100 calculates the distances between each of the codebooks CD41 to CD44, etc. and the fourth partial query QE1-4 as distances DS41 to DS44, etc., as shown in codebook information TB4.

Although the distances DS11 to DS44 and the like are shown abstractly in FIG. 2 for the sake of explanation, it is assumed that the distances DS11 to DS44, etc. are concrete values. The distance is a value expressed as a floating point number, but scale-offset-compression (https://www.unidata.ucar.edu/blogs/ developer/entry/compression_by_scaling_and_offfset" etc.) may be used to compress it into a 1-byte integer type. The information processing apparatus 100 may calculate the distance between each codebook and the query at the timing when the search query QE1 is acquired, or at the timing when the information becomes necessary. In the search process, the information processing device 100 uses the codebook information TB1 to TB4 as a lookup table to calculate the distance between each node and the search query QE1, that is, the approximate distance (also referred to as "first distance"). do. In this way, when searching a graph in the search process, the information processing device 100 uses approximate distances (also referred to as "second distances") rather than true distances (also referred to as "second distances") between each node and the search query. 1 distance) and perform the processing, it is possible to perform efficient search processing.

The information processing device 100 executes a search process targeting the search query QE1 (step S12). The information processing device 100 obtains a search result for the search query QE1 by performing a search process as shown in FIG. 11 using the graph GR1 for the search query QE1. Details of the search process shown in FIG. 11 will be described later. The information processing device 100 performs a search process on the search query QE1, thereby extracting the searched nodes as nodes in the vicinity of the search query QE1.

In the search process targeting the search query QE1, the information processing device 100 selects a predetermined node from among the nodes as a node (hereinafter also referred to as "starting point node") that becomes the starting point (starting point) of the search for the graph GR1. do. For example, the information processing apparatus 100 selects the starting node using a predetermined index such as a tree-structured index. In the example of FIG. 1, it is assumed that the information processing apparatus 100 selects node N7 as the starting point node. Note that although FIG. 1 shows a case where only node N7 is selected as the starting point node using a predetermined index to simplify the explanation, the information processing apparatus 100 can also select multiple nodes as the starting point node. Alternatively, the starting node may be selected by various methods such as random.

Here, in the search process targeting the search query QE1, the information processing device 100 parallelizes the distance between the search query and a node to which edges from one node are connected (hereinafter also referred to as "connection node"). Calculate (step S13). In FIG. 1, the information processing device 100 uses the search query QE1 for nodes N9, N12, N54, N85, etc., which are nodes (connection nodes) to which edges from the node N7 are connected, as shown in the batch processing information LT1. Calculate the distance between the two in parallel.

For example, in node N9, as shown in node information INF1, the first subvector is codebook CD12, the second subvector is codebook CD23, the third subvector is codebook CD35, and the fourth subvector is codebook CD47. Indicates that it is associated with. Note that the size of the variable CD is determined by the size of the codebook, so if the codebook size is 16, for example, the variable CD only needs to be 4 bits, making it possible to significantly compress data. Therefore, the information processing device 100 uses the distance DS12 of the codebook CD12, the distance DS23 of the codebook CD23, the distance DS35 of the codebook CD35, and the distance DS47 of the codebook CD47 to determine the distance between the node N9 and the search query QE1. Calculate. For example, the information processing device 100 calculates the sum of the distance DS12 of the codebook CD12, the distance DS23 of the codebook CD23, the distance DS35 of the codebook CD35, and the distance DS47 of the codebook CD47 as the distance between the node N9 and the search query QE1. calculate.

For example, in the node N12, as shown in the node information INF2, the first subvector is codebook CD14, the second subvector is codebook CD29, the third subvector is codebook CD31, and the fourth subvector is codebook CD45. Indicates that it is associated with. For example, the information processing device 100 calculates the sum of the distance DS14 of the codebook CD14, the distance DS29 of the codebook CD29, the distance DS31 of the codebook CD31, and the distance DS45 of the codebook CD45 as the distance between the node N12 and the search query QE1. calculate. Similarly, the information processing device 100 uses the node information INF3 of the node N54 to calculate the distance between the node N54 and the search query QE1, and uses the node information INF4 of the node N85 to calculate the distance between the node N85 and the search query QE1. Calculate distance.

For example, the information processing device 100 calculates the distance between each of the nodes N9, N12, N54, and N85 and the search query QE1 at once by parallelizing SIMD calculations. Thereby, the information processing device 100 can speed up the distance calculation by parallelizing the distance calculation, and can perform efficient search processing.

Note that, in order to simplify the explanation, FIG. 1 shows a case where the distance is computed in parallel for four nodes, but the number to be computed in parallel is determined based on the specifications of the information processing device 100. . For example, if the number that the information processing device 100 can process at once using SIMD (also referred to as “unit that can be processed at once”) is “4”, the processing will be similar to that shown in FIG. 1, but it will be possible to process at once using SIMD. When the number (unit that can be collectively processed) is "16", the information processing apparatus 100 calculates distances in parallel for 16 nodes. Further, when the batch processable unit is "32", the information processing apparatus 100 calculates distances in parallel for 32 nodes. In this way, the information processing device 100 enables efficient search processing by collectively calculating the distances of the number of nodes corresponding to the number that can be processed at once by SIMD (batch processable unit). be able to.

The information processing device 100 calculates the number of searches by performing a search process as shown in FIG. The node is extracted as a node in the vicinity of the search query QE1.

As described above, the information processing device 100 calculates the approximate distance (first distance) between the vector quantized vector of each node and the search query, and performs the search process using the first distance. This enables efficient search processing. Furthermore, the information processing apparatus 100 can perform efficient search processing by performing distance calculations for a number of nodes that can be parallelized at once. As described above, the information processing device 100 repeatedly uses a limited memory area (lookup table) (memory cache) by collectively calculating approximate distances between nodes of the graph and multiple connected nodes. speed).

In addition, in the example of FIG. 1, the information processing device 100 performs search processing using vectors divided by direct product quantization, thereby achieving more efficient search processing than when performing search processing without dividing vectors. Search processing can be enabled. For example, the information processing device 100 calculates distances between short vectors by direct product quantization, and can reduce the size of vectors that are targets of distance calculation. Further, the information processing apparatus 100 can limit the memory space accessed by the lookup table used during distance calculation to the lookup table of divided subvectors. The information processing apparatus 100 can reduce the vector size while suppressing a decrease in search accuracy by using Cartesian product quantization. In addition, in FIG. 1, the processing in the case where Cartesian product quantization is performed has been described, but the case where Cartesian product quantization is performed is only an example, and the information processing device 100 can process vectors that have not been subjected to Cartesian product quantization. The search process may be performed using .

[1-1-1. others〕
The above-described processing is merely an example, and the information processing apparatus 100 may perform search processing using various information and methods for efficient search processing. Regarding this point, each item will be explained in detail.

(Storage mode of edges (connection nodes))
Regarding the information of edges (connection nodes) connected to each node, there are two types of storage modes (storage methods) for edges (connection nodes) in nodes: the following first storage mode and second storage mode. Therefore, the storage mode may be selected depending on the usage mode.

For example, as a first storage mode, each node may be stored in association with the ID of the connected node (object), and information on objects subjected to product quantization may be stored in another table. For example, the data storage format shown in FIGS. 7 and 8 corresponds to the first storage format. In the case of the first storage mode, memory usage can be reduced.

Furthermore, for example, as a second storage mode, each node has a quantized object directly. In the case of the second storage mode, information obtained by quantizing connection nodes (objects) of a certain node is stored in association with each node. For example, a data storage mode in which quantized information (node information INF1 to INF4 in FIG. 2) of nodes N9, N12, N54, and N85, which are connected nodes of node N7, is stored in association with node N7. corresponds to the second storage mode. In the case of the second storage mode, since objects are accessed sequentially during a search, a decrease in speed can be suppressed.

(lookup table)
In the above example, each divided subvector is individually clustered and a lookup table (codebook information) is generated individually, but the lookup table is not limited to this case. There may be.

For example, all subvectors may be clustered and one lookup table may be generated. In the above example, for example, one lookup table may be generated by clustering all four subvectors: the first subvector, the second subvector, the third subvector, and the fourth subvector. .

Further, lookup tables may be generated for each of a plurality of classes by partially merging individual subvectors. For example, subvectors with similar variances may be merged to form multiple classes, and a lookup table may be generated for each of the multiple classes. In the above example, for example, if the variances of two subvectors, the first subvector and the third subvector, are similar, and the variances of the two subvectors, the second subvector and the fourth subvector, are similar, then the first The sub-vector and the third sub-vector may be merged to form the first class, and the second sub-vector and the fourth sub-vector may be merged to form the second class. In this case, two lookup tables, a first class lookup table (codebook information) and a second class lookup table (codebook information), may be generated.

(transposed)
The data retention described above is merely an example, and the information processing apparatus 100 may retain data in various ways so as to enable efficient data reference. For example, the information processing device 100 may hold transposed data. This point will be explained using FIG. 3. FIG. 3 is a diagram showing an example of data according to the first embodiment.

For example, if data is held as shown in node information INF1 to INF4 and node information INF1 to INF4 are sequentially processed, codebook information TB1 to TB4, which is a lookup table, is repeatedly processed as shown in step S14 in FIG. Please refer to it.

Therefore, as shown in step S15, transposed data TR1 to TR4 are generated by transposing the node information INF1 to INF4. For example, the information processing device 100 generates transposed data TR1 to TR4 by transposing node information INF1 to INF4.

In FIG. 3, first data TR1 is generated which is a list of codebooks corresponding to the first subvectors of nodes N9, N12, N54, and N85 among the node information INF1 to INF4. Specifically, codebook CD12 corresponding to the first subvector of node N9, codebook CD14 corresponding to the first subvector of node N12, codebook CD13 corresponding to the first subvector of node N54, and node N85. First data TR1, which is a list of codebook CD18 corresponding to the first subvector of , is generated.

Similarly, second data TR2 is generated, which is a list of codebooks corresponding to the second subvectors of nodes N9, N12, N54, and N85 among the node information INF1 to INF4. Also, third data TR3 is generated, which is a list of codebooks corresponding to the third subvectors of nodes N9, N12, N54, and N85 among the node information INF1 to INF4. Fourth data TR4 is generated, which is a list of codebooks corresponding to the fourth subvectors of nodes N9, N12, N54, and N85 among the node information INF1 to INF4.

Note that correspondence information is generated that indicates the correspondence between which data in the list of each transposed data TR1 to TR4 corresponds to which node. In the example of FIG. 3, among the list of transposed data TR1 to TR4, the first data corresponds to node N9, the second data corresponds to node N12, and the third data corresponds to node N54. Correspondingly, association information indicating that the fourth (last) data corresponds to node N85 is generated. By referring to the correspondence information, the information processing device 100 can specify which node each data in the list of transposed data TR1 to TR4 corresponds to. For example, the information processing device 100 generates association information.

The information processing device 100 performs processing using transposed data TR1 to TR4 obtained by transposing node information INF1 to INF4. For example, when the information processing device 100 refers to the lookup table using the transposed data TR1, it refers only to the codebook information TB1, and refers to only one codebook information TB1. Similarly, when the information processing device 100 refers to the lookup table using the transposed data TR2, it refers only to the codebook information TB2, and refers to only one codebook information TB2. Similarly, only one codebook information is referred to for the transposed data TR3 and TR4. As a result, the information processing apparatus 100 can efficiently refer to data, thereby enabling efficient search processing.

In this way, in order to refer to the lookup table for each subclass (subvector, etc.) during batch distance calculation, the data of the neighboring objects (connected nodes) of the node is transposed, and the data is sent to the node in the order in which it is compiled for each subclass. By having both, the locality of reference is further increased and speeding up is possible.

Note that, in order to simplify the explanation, FIG. 3 shows a case where transposed data is generated for four nodes, but the unit (number of nodes) in which transposed data is generated is based on the specifications of the information processing device 100. Determined by For example, if the batch processable unit of the information processing device 100 is "4", the process is similar to that shown in FIG. 3, but if the batch processable unit is "16", 16 nodes are treated as one unit. Transposed data is generated. Further, when the batch processable unit is "32", transposed data is generated with 32 nodes as one unit. In this way, by using the transposed data that is generated according to the number that the information processing apparatus 100 can process at once using SIMD (units that can be processed at once), the information processing apparatus 100 can refer to data efficiently. This makes it possible to perform efficient search processing.

For example, much of the time during the above-mentioned search process is spent on distance calculation, but distance calculation can be sped up by parallel calculation using SIMD as described above. When distance calculation is parallelized in this way, the time required to fetch object data occupies the search process. If this fetch time can be reduced, further speeding up can be achieved. Note that there is a method of prefetching to reduce the time required to fetch data, but there is a limit.

On the other hand, the information processing device 100 can suppress data fetching and achieve further speedup by increasing the locality of reference and making it possible to efficiently refer to data as described above. .

(search results)
Note that when returning search results based on the second distance (true distance) described above, the following two methods, the first method and the second method, can be considered.

For example, the first method is to search for more than the specified number of searches, and at the end of the search, calculate the true distance, sort by distance, and select the specified number of objects as search results. good. In this case, the information processing device 100 extracts nodes whose number (also referred to as "extended search number") is greater than the specified number of searches (also referred to as "first number"). By setting the number of searches (also referred to as ``number 2'') as the search number and performing the search process as shown in Figure 11, nodes with the extended search number, that is, nodes whose number is greater than the specified number of searches, are extracted as nearby candidate nodes. do. Then, the information processing device 100 calculates a second distance (true distance) for the neighboring candidate nodes, and selects the first number of nodes from among the neighboring candidate nodes, starting from the one with the shortest second distance, in the vicinity of the search query. Extract as a node.

For example, as a second method, only when a node (object) is within the search range or within the search range based on the first distance (approximate distance), the second distance (true The search result based on the true distance may be returned by calculating the approximate distance (distance) and replacing the approximate distance with the true distance.

The search range here is the range defined by "r" in FIG. 11, and the search range is defined by "r(1+ε)" using the search range coefficient "ε" in FIG. range. For example, when calculating the second distance (true distance) when the node enters the search range, the information processing device 100 calculates whether the node A second distance (true distance) is calculated. Further, when calculating the second distance (true distance) when the node enters the search range, the information processing device 100 calculates the second distance (true distance) when the first distance (approximate distance) of the node is equal to or less than “r(1+ε)”. , calculate the second distance (true distance) of that node.

For example, to improve accuracy, the condition may be set to be within the search range. Furthermore, when speeding up the processing is desired, the condition may be set to be within the search range. Note that the above is only an example, and which condition is to be used may be set as appropriate depending on the value of the search range coefficient "ε", the purpose of processing, etc.

(vector quantization)
Note that the information processing device 100 may perform two-stage vector quantization as shown in Patent Document 3. Then, the information processing device 100 may calculate the distance between the search query and each node using the following equation (1).

Here, the value on the left side of the above equation (1) indicates, for example, the square distance between the search query and the node. Further, for example, "x" in the above formula (1) corresponds to a query. Further, for example, "y" in the above formula (1) corresponds to a node. Further, for example, "q _c (y)" on the right side of the above equation (1) indicates a representative vector (centroid) of "y". For example, when the information processing apparatus 100 does not have node vector data for "y" in the above equation (1), the information processing apparatus 100 may use the numerical value of the centroid of the partial area to which each node belongs. Further, for example, "yq _c (y)" indicates a residual vector. Further, for example, “q _p ” on the right side of the above equation (1) indicates a predetermined quantizer (function).

Further, for example, "j" on the right side of the above equation (1) may be the number of divided spaces. For example, in the example of FIG. 1, "j" on the right side of the above equation (1) may be the number of divided spaces "4". Further, for example, "u _j ()" on the right side of the above equation (1) indicates a partial residual vector between the vectors in parentheses. For example, the information processing device 100 may calculate the distance between the query and the node by calculating the squared distance between the query and the node in each subspace using the above equation (1), and summing the squared distance between the query and the node. good.

[1-2. Information processing system configuration]
As shown in FIG. 4, the information processing system 1 includes a terminal device 10, an information providing device 50, and an information processing device 100. The terminal device 10, the information providing device 50, and the information processing device 100 are connected via a predetermined network N so that they can communicate by wire or wirelessly. FIG. 4 is a diagram illustrating a configuration example of an information processing system according to the first embodiment. Note that the information processing system 1 shown in FIG. 4 may include multiple terminal devices 10, multiple information providing devices 50, and multiple information processing devices 100.

The terminal device 10 is an information processing device used by a user. The terminal device 10 accepts various operations by the user. Note that below, the terminal device 10 may be referred to as a user. That is, in the following, the user can also be read as the terminal device 10. Note that the above-described terminal device 10 is realized by, for example, a smartphone, a tablet terminal, a notebook PC (Personal Computer), a desktop PC, a mobile phone, a PDA (Personal Digital Assistant), or the like.

The information providing device 50 is an information processing device that stores information for providing various information to users and the like. For example, the information providing device 50 stores object IDs based on character information etc. collected from various external devices such as web servers. For example, the information providing device 50 is an information processing device that provides image search services to users and the like. For example, the information providing device 50 stores various pieces of information for providing an image search service. For example, the information providing device 50 provides the information processing device 100 with vector information corresponding to an image that is a target of an image search service. Furthermore, by transmitting a query to the information processing device 100, the information providing device 50 receives an object ID and the like indicating an image corresponding to the query from the information processing device 100.

The information processing device 100 is a computer that provides a search service. The information processing apparatus 100 provides a specified number of search objects corresponding to the search query as search results. The information processing apparatus 100 uses a graph in which search target nodes (objects) are connected by edges to provide nodes (objects) near a search query as search results. The information processing device 100 uses a graph in which a group of nodes corresponding to each of a plurality of objects are connected by edges to search for nodes in the vicinity of a search query. The distance between the search query and the search query is calculated using vector information of a plurality of nodes subjected to vector quantization.

For example, upon receiving a query (search query) from the terminal device 10, the information processing device 100 searches for a target (object) similar to the search query, and provides the search result to the terminal device. Further, for example, the data that the information processing device 100 provides to the terminal device may be data itself such as image information, or information for referring to corresponding data such as a URL (Uniform Resource Locator). Good too. Further, the search query or search target (object) may be any type of data such as image, audio, text data, etc.

[1-3. Configuration of information processing device]
Next, the configuration of the information processing apparatus 100 according to the first embodiment will be described using FIG. 5. FIG. 5 is a diagram illustrating a configuration example of the information processing device 100 according to the first embodiment. As shown in FIG. 5, the information processing device 100 includes a communication section 110, a storage section 120, and a control section 130. Note that the information processing device 100 includes an input unit (for example, a keyboard, a mouse, etc.) that accepts various operations from an administrator of the information processing device 100, and a display unit (for example, a liquid crystal display, etc.) for displaying various information. May have.

(Communication Department 110)
The communication unit 110 is realized by, for example, a NIC (Network Interface Card). The communication unit 110 is connected to a network (for example, network N in FIG. 4) by wire or wirelessly, and transmits and receives information to and from the terminal device 10 and the information providing device 50.

(Storage unit 120)
The storage unit 120 is realized by, for example, a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. As shown in FIG. 5, the storage unit 120 according to the first embodiment includes an object information storage unit 121, a graph information storage unit 122, a quantization information storage unit 123, and a codebook information storage unit 124. .

(Object information storage unit 121)
The object information storage unit 121 according to the first embodiment stores various information regarding objects. For example, the object information storage unit 121 stores object IDs and vector data. FIG. 6 is a diagram illustrating an example of an object information storage unit according to the first embodiment. The object information storage unit 121 shown in FIG. 6 includes items such as "object ID" and "vector information."

"Object ID" indicates identification information for identifying an object. Moreover, "vector information" indicates vector information corresponding to the object identified by the object ID. That is, in the example of FIG. 6, vector data (vector information) corresponding to the object is registered in association with an object ID that identifies the object.

For example, the example in FIG. 6 indicates that the object (target) identified by the object ID "OB1" is associated with multidimensional vector information of "10, 24, 51, 2...".

Note that the object information storage unit 121 is not limited to the above, and may store various types of information depending on the purpose.

(Graph information storage unit 122)
The graph information storage unit 122 according to the first embodiment stores various information regarding graphs. For example, the graph information storage unit 122 stores the generated graph. FIG. 7 is a diagram illustrating an example of a graph information storage unit according to the first embodiment. The graph information storage unit 122 shown in FIG. 7 has items such as "node ID", "object ID", and "connected node information".

"Node ID" indicates identification information for identifying each node (object) in the graph. Moreover, "object ID" indicates identification information for identifying an object. Note that if the node ID and object ID are the same, the object ID is stored in the "node ID" and the graph information storage unit 122 does not need to include the item "object ID." For example, when used as an object ID and a node ID, the object ID may be stored in the "node ID" and the graph information storage unit 122 may not include the "object ID" item.

Further, "connected node information" indicates information regarding a node (reference destination node) that can be traced from the corresponding node. For example, "connected node information" includes information such as "reference destination." “Reference destination” indicates information for identifying a reference destination (node) that is connected by an edge and can be traced from that node. That is, in the example of FIG. 7, a reference destination (node) that can be traced from the node by an edge is registered in association with a node ID (object ID) that identifies the node. Note that the "connected node information" may include information (edge ID) for identifying an edge connected to a reference destination.

The example of FIG. 7 indicates that the node (node N1) identified by the node ID "N1" corresponds to the object (target) identified by the object ID "OB1". Further, an edge is connected from the node N1 to the node (node N4) identified by the node ID "N4", indicating that it is possible to trace from the node N1 to the node N4.

Further, the node (node N2) identified by the node ID "N2" corresponds to the object (target) identified by the object ID "OB2". Further, an edge is connected from the node N2 to the node (node N6) identified by the node ID "N6", indicating that it is possible to trace from the node N2 to the node N6.

Further, the node (node N7) identified by the node ID "N7" corresponds to the object (target) identified by the object ID "OB7". Furthermore, edges are connected from node N7 to nodes N9, N12, N54, and N85, indicating that each of nodes N9, N12, N54, and N85 can be traced from node N2.

Note that the graph information storage unit 122 is not limited to the above, and may store various information depending on the purpose. For example, the graph information storage unit 122 may store the length of an edge connecting each node (vector). That is, the graph information storage unit 122 may store information indicating the distance between each node (vector). Note that the graph information storage unit 122 is not limited to the above, and may store graph information using various data structures.

The graph may also include a program module that receives a query as input, searches for nodes by tracing edges in the graph, and extracts and outputs nodes similar to the query. That is, the graph may be assumed to be used as a program module that performs search processing using the graph. For example, the graph GR1 may be a program that, when vector data is input as a query, extracts nodes corresponding to vector data similar to the vector data from the graph and outputs them. For example, the graph GR1 may be data used as a program module that searches for similar images corresponding to a query image. For example, the graph GR1 causes a computer to function based on an input query to extract and output nodes similar to the query in the graph.

(Quantization information storage unit 123)
The quantization information storage unit 123 according to the first embodiment stores various information regarding allocation processing. FIG. 8 is a diagram illustrating an example of a quantization information storage unit according to the first embodiment. In the example of FIG. 8, the quantization information storage unit 123 has items such as "node ID", "object ID", and "quantization information".

"Node ID" indicates identification information for identifying each node (object) in the graph. Moreover, "object ID" indicates identification information for identifying an object. Note that when the node ID and the object ID are common, the object ID is stored in the "node ID", and the quantization information storage unit 123 does not need to include the item "object ID".

Moreover, "quantization information" indicates information on quantized vectors of each node (object). For example, "quantization information" includes information such as "element" and "codebook ID." "Element" indicates the arrangement of the corresponding object in the vector. In the example of FIG. 8, the "element" includes "#1", "#2", "#3", and "#4". In this case, the vector of each node (object) is divided into four parts, and each divided partial vector is quantized by the codebook. Note that the number of divisions is not limited to 4. For example, if the number of divisions is 6, "element" may include "#1", "#2", "#3", "#4", "#5", " #6” is included. "Codebook ID" indicates information for identifying the codebook corresponding to each element (partial vector).

In the example of FIG. 8, the vector of node N9 (object OB9) is quantized by the codebook (codebook CD12) whose first partial vector among the four divided partial vectors is identified by the codebook ID "CD12". Indicates that the In addition, in the vector of node N9 (object OB9), among the four divided partial vectors, the second partial vector from the beginning is quantized by the codebook (codebook CD23) identified by the codebook ID "CD23". to show that

In addition, in the vector of node N9 (object OB9), among the four divided partial vectors, the third partial vector from the beginning is quantized by the codebook (codebook CD35) identified by the codebook ID "CD35". to show that In addition, the vector of node N9 (object OB9) has a codebook (codebook CD47) in which the fourth (i.e., the last) partial vector from the beginning among the four divided partial vectors is identified by the codebook ID "CD47". ) indicates that it is quantized.

Note that the quantization information storage unit 123 is not limited to the above, and may store various information depending on the purpose.

(Codebook information storage unit 124)
The codebook information storage unit 124 according to the first embodiment stores various information regarding the codebook. For example, the codebook information storage unit 124 stores codebook IDs and vector information of each codebook. FIG. 9 is a diagram illustrating an example of a codebook information storage unit according to the first embodiment. In the example of FIG. 9, the codebook information storage unit 124 stores a lookup table that indicates the association between each codebook and a vector.

The codebook information storage unit 124 stores codebook information TB1 used for quantizing the first partial vector among the four divided partial vectors, and the second partial vector from the beginning among the four divided partial vectors. Codebook information TB2 used for quantization is stored. The codebook information storage unit 124 also stores codebook information TB3 used for quantizing the third partial vector from the beginning among the partial vectors divided into four; Codebook information TB4 and the like used for quantizing the th (ie, last) partial vector is stored.

In the example of FIG. 9, the codebook information TB1 includes the codebook identified by the codebook ID "CD11" (codebook CD11), the codebook identified by the codebook ID "CD12" (codebook CD12), etc. Store codebook information. For example, codebook CD11 indicates that multidimensional vector information of "5, 13..." is associated. Further, codebook CD12 indicates that multidimensional vector information of "27, 51..." is associated.

Note that the codebook information storage unit 124 is not limited to the above, and may store various information depending on the purpose. For example, the codebook information storage unit 124 may store information indicating the difference (distance) between each codebook and the search query.

(Control unit 130)
Returning to the explanation of FIG. 5, the control unit 130 is a controller, and for example, controls the internal processing of the information processing apparatus 100 by a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), etc. This is realized by executing various programs (corresponding to an example of an information processing program) stored in the storage device of , using the RAM as a work area. Further, the control unit 130 is a controller, and is realized by, for example, an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

As shown in FIG. 5, the control unit 130 includes an acquisition unit 131, a generation unit 132, a search processing unit 133, and a provision unit 134, and realizes or executes information processing functions and operations described below. do. Note that the internal configuration of the control unit 130 is not limited to the configuration shown in FIG. 5, and may be any other configuration as long as it performs information processing to be described later.

(Acquisition unit 131)
The acquisition unit 131 acquires various information. For example, the acquisition unit 131 acquires various information from the storage unit 120. For example, the acquisition unit 131 acquires various information from the object information storage unit 121, the graph information storage unit 122, the quantization information storage unit 123, the codebook information storage unit 124, and the like. Further, the acquisition unit 131 acquires various information from an external information processing device. The acquisition unit 131 acquires various information from the terminal device 10 and the information providing device 50.

The acquisition unit 131 acquires a graph. For example, the information processing device 100 may acquire spatial information SP1 in FIG. For example, the information processing device 100 may acquire a graph from an external device such as the information providing device 50.

The acquisition unit 131 acquires search queries for a plurality of objects that are data search targets. For example, the acquisition unit 131 acquires information regarding the search query QE1. For example, the acquisition unit 131 acquires a search query related to image search. For example, the acquisition unit 131 acquires a query from the terminal device 10 to be used. For example, the acquisition unit 131 acquires a query from the information providing device 50 that has received the query from the terminal device 10 to be used.

(Generation unit 132)
The generation unit 132 generates various information. For example, the generation unit 132 generates various information (data) from the information (data) stored in the storage unit 120. For example, the generation unit 132 generates various information from information (data) stored in the object information storage unit 121, the graph information storage unit 122, the quantization information storage unit 123, the codebook information storage unit 124, etc. .

The generation unit 132 may generate a graph as shown in the graph information storage unit 122. For example, the generation unit 132 generates spatial information SP1. Further, the generation unit 132 may generate information related to vector quantization as shown in the quantization information storage unit 123. For example, the generation unit 132 generates information in which each object such as the node N1 (object OB1) is vector quantized. Further, the generation unit 132 may generate information regarding the codebook as shown in the codebook information storage unit 124. The generation unit 132 may generate a codebook lookup table. For example, the generation unit 132 generates information regarding codebooks, such as codebook information TB1 to TB4. Note that when the information processing device 100 acquires the information shown in the graph information storage unit 122, quantization information storage unit 123, and codebook information storage unit 124 from an external device such as the information providing device 50, the information processing device 100 The generation unit 132 may not be included.

(Search processing unit 133)
The search processing unit 133 provides search services regarding objects. The search processing unit 133 searches for various information. The search processing unit 133 searches various information. For example, the search processing unit 133 searches for objects by searching a graph. For example, when the query acquired by the acquisition unit 131 is acquired, the search processing unit 133 searches for an object similar to the query by searching a graph. For example, the search processing unit 133 extracts objects similar to the query by searching the graph. For example, the search processing unit 133 extracts objects similar to the query by searching the graph based on the processing procedure shown in FIG. Note that the information processing device 100 does not need to include the search processing unit 133 when not providing a search service.

The search processing unit 133 selects various types of information in the search process. The search processing unit 133 extracts various information in the search process. The search processing unit 133 determines various information in the search process. The search processing unit 133 determines various information in the search process. The search processing unit 133 changes various information in the search process. The search processing unit 133 updates various information during search processing.

The search processing unit 133 uses a graph in which a group of nodes corresponding to each of a plurality of objects are connected by edges to search for nodes in the vicinity of a search query. The distance between the search query and the search query is calculated using vector information of a plurality of nodes subjected to vector quantization. The search processing unit 133 parallelizes and collectively calculates distances between a plurality of nodes and a search query. The search processing unit 133 performs parallel processing to calculate distances between a search query and a plurality of nodes whose number is determined based on the specifications of the information processing device 100.

The search processing unit 133 calculates distances between the plurality of nodes and the search query using a codebook in which each of the plurality of nodes is associated with a corresponding representative vector. The search processing unit 133 calculates the distance between the search query and a plurality of nodes in which edges from one node are connected. The search processing unit 133 searches the plurality of nodes and the search query using node information stored in association with one node and reference information indicating a representative vector with which each of the plurality of nodes is associated. Calculate the distance to.

The search processing unit 133 calculates distances between a plurality of nodes and a search query using vector information of a plurality of nodes, each of which is divided into a plurality of partial vectors by Cartesian product quantization. The search processing unit 133 uses a plurality of codebooks that are associated with representative vectors corresponding to vectors for each division position of a plurality of partial vectors into which each of the plurality of nodes is divided, to combine the plurality of nodes with the search query. Calculate the distance. The search processing unit 133 uses the node information stored in such a manner that the plurality of nodes and reference information indicating representative vectors associated with each of the plurality of partial vectors of each of the plurality of nodes are associated with one node. , calculate the distance between multiple nodes and the search query.

The search processing unit 133 uses reference information in which each of the plurality of nodes is associated with a list of representative vectors to which each of the plurality of partial vectors of each node is associated, and the search processing unit 133 associates the plurality of nodes with the search query. Calculate the distance. The search processing unit 133 includes transposed information in which representative vectors to which each of a plurality of partial vectors of each of a plurality of nodes corresponds are arranged in a list for each division position, and a correspondence indicating a position to which each of a plurality of nodes corresponds in a list. The distance between the plurality of nodes and the search query is calculated using the reference information including the attached information. The search processing unit 133 calculates distances between the plurality of nodes and the search query using one codebook in which each of the plurality of partial vectors of each of the plurality of nodes is associated with a corresponding representative vector.

The search processing unit 133 targets a predetermined node among the nodes to be processed in the search process. Calculate distance. The search processing unit 133 extracts a second number of nodes, which is a larger number than the first number of nodes to be extracted as neighboring nodes of the search query, as neighboring candidate nodes in the search process, and extracts a second number of nodes as neighboring candidate nodes. Calculate distance. The search processing unit 133 extracts a first number of nodes from among the nearby candidate nodes having the shorter second distance as nodes in the vicinity of the search query.

The search processing unit 133 calculates a second distance for nodes whose first distance is within a predetermined threshold. The search processing unit 133 calculates a second distance for nodes within the search range indicating the target range to be extracted as nearby nodes. The search processing unit 133 calculates a second distance for nodes within a search range indicating a target range of the search process.

(Providing unit 134)
The providing unit 134 provides various information. For example, the providing unit 134 transmits various information to the terminal device 10 and the information providing device 50. For example, the providing unit 134 provides an object ID corresponding to a search query as a search result. The providing unit 134 transmits the search results to the terminal device 10. The providing unit 134 provides the object ID searched by the search processing unit 133 to the terminal device 10 as a search result corresponding to the search query.

Further, the providing unit 134 may provide the object ID searched by the search processing unit 133 to the information providing device 50. For example, the providing unit 134 provides the information providing device 50 with the object ID extracted by the search processing unit 133 through the search. The providing unit 134 provides the object ID extracted by the search processing unit 133 to the information providing device 50 as information indicating a vector corresponding to the query.

[1-4. Information processing flow]
Next, the procedure of information processing by the information processing system 1 according to the first embodiment will be described using FIG. 10. FIG. 10 is a flowchart illustrating an example of information processing according to the first embodiment.

As shown in FIG. 10, the information processing apparatus 100 obtains search queries for a plurality of objects that are data search targets (step S101). In the example of FIG. 1, the information processing device 100 obtains the search query QE1.

The information processing apparatus 100 uses a graph in which a group of nodes corresponding to each of a plurality of objects are connected by edges to search for nodes near a search query. The distance between the node and the search query is calculated using vector information of a plurality of nodes subjected to vector quantization (step S102). In the example of FIG. 1, the information processing device 100 calculates the distances between the plurality of nodes N9, N12, N54, N85 and the search query QE1 by vector quantizing the distances between the plurality of nodes N9, N12, N54, N85. Calculate using information.

[1-5. Search processing example]
Here, an example of the search process according to the first embodiment will be described using FIG. 11 as an example. FIG. 11 is a flowchart illustrating an example of search processing according to the first embodiment. The search processing described below is performed by the search processing unit 133 of the information processing device 100. Further, the object referred to below may be read as a node. Note that in the following, the information processing device 100 (search processing unit 133) performs search processing. Note that if the search service is not provided, the information processing device 100 does not need to include the search processing unit 133. The search query for the process described below may be an additional node, a target node, an object specified by the user, or the like.

Here, the neighboring object set N(G, y) is a set of neighboring objects that are related by edges attached to node y. "G" may be predetermined graph data (for example, graph GR1 shown in spatial information SP1). For example, the information processing device 100 executes k-nearest neighbor search processing.

For example, the information processing device 100 sets the radius r of the hypersphere to ∞ (infinity) (step S300), and extracts a subset S from an existing object set (step S301). For example, the information processing apparatus 100 may extract the object (node) selected as the root node as the subset S. Further, for example, a hypersphere is a virtual sphere that indicates a search range. Note that the objects included in the object set S extracted in step S301 are also included in the initial set of the object set R as a search result.

Next, among the objects included in the object set S, the information processing apparatus 100 extracts the object having the shortest distance from the object y, assuming that the search query object is y, and sets it as the object s (step S302). For example, when the only object (node) selected as the root node is an element of S, the information processing apparatus 100 extracts the root node as the object s. Next, the information processing device 100 excludes the object s from the object set S (step S303).

Next, the information processing apparatus 100 determines whether the distance d(s, y) between the object s and the object y exceeds r(1+ε) (step S304). Here, ε is an expansion element, and r(1+ε) is a value indicating the radius of the search range (only nodes within this range are searched. Accuracy can be improved by making it larger than the search range). be. If the distance d(s, y) between object s and object y exceeds r(1+ε) (step S304: Yes), the information processing device 100 outputs the object set R as a neighboring object set of object y (step S305), the process ends.

If the distance d(s, y) between the object s and the search query object y does not exceed r(1+ε) (step S304: No), the information processing device 100 determines the neighboring object set N(G, s) of the object s. One object that is not included in object set C is selected from among the objects that are elements of , and the selected object u is stored in object set C (step S306). The object set C is provided for convenience in order to avoid duplicate searches, and is set to be an empty set at the start of processing.

Next, the information processing device 100 determines whether the distance d(u, y) between the object u and the object y is less than or equal to r(1+ε) (step S307). If the distance d(u,y) between the object u and the object y is less than or equal to r(1+ε) (step S307: Yes), the information processing device 100 adds the object u to the object set S (step S308). Further, if the distance d(u,y) between the object u and the object y is not less than or equal to r(1+ε) (step S307: No), the information processing apparatus 100 performs the determination (process) of step S309.

Next, the information processing device 100 determines whether the distance d(u, y) between the object u and the object y is less than or equal to r (step S309). If the distance d(u, y) between object u and object y exceeds r (step S309: No), the information processing apparatus 100 performs the determination (process) of step S315. That is, if the distance d(u, y) between object u and object y is not less than or equal to r, the information processing apparatus 100 performs the determination (process) of step S315.

If the distance d(u,y) between the object u and the object y is less than or equal to r (step S309: Yes), the information processing device 100 adds the object u to the object set R (step S310). Then, the information processing device 100 determines whether the number of objects included in the object set R exceeds ks (step S311). The predetermined number ks is an arbitrarily determined natural number. For example, ks may be the number of searches or the number of extraction targets. Further, for example, when there is no upper limit on the number of objects to be extracted in a range search or the like, ks may be set to infinity. For example, ks may be 4, etc. If the number of objects included in the object set R does not exceed ks (step S311: No), the information processing device 100 performs the determination (process) of step S313.

If the number of objects included in the object set R exceeds ks (step S311: Yes), the information processing device 100 selects the object that is the longest (farthest) from the object y among the objects included in the object set R. Exclude from object set R (step S312).

Next, the information processing device 100 determines whether the number of objects included in the object set R matches ks (step S313). If the number of objects included in the object set R does not match ks (step S313: No), the information processing device 100 performs the determination (process) of step S315. Further, when the number of objects included in the object set R matches ks (step S313: Yes), the information processing device 100 determines that the distance to the object y is the longest (farthest) among the objects included in the object set R. The distance between the object and object y is set to a new r (step S314).

Then, the information processing device 100 determines whether all objects that are elements of the neighboring object set N(G, s) of the object s have been selected and stored in the object set C (step S315). . If all objects that are elements of the neighboring object set N(G, s) of the object s have not been selected and stored in the object set C (step S315: No), the information processing device 100 selects the objects that are elements of the neighboring object set N(G, s) of the object s and stores them in the object set C (step S315: No). Return to and repeat the process.

When all objects that are elements of the object set N(G, s) in the vicinity of the object s have been selected and stored in the object set C (step S315: Yes), the information processing device 100 It is determined whether or not is an empty set (step S316). If the object set S is not an empty set (step S316: No), the information processing device 100 returns to step S302 and repeats the process. Furthermore, if the object set S is an empty set (step S316: Yes), the information processing device 100 outputs the object set R and ends the process (step S317). For example, the information processing apparatus 100 may select an object (node) included in the object set R as a neighboring node corresponding to the additional node (input object y). For example, the information processing apparatus 100 may extract (select) objects (nodes) included in the object set R as neighboring nodes corresponding to the target node (input object y). Further, for example, the information processing device 100 may provide the objects (nodes) included in the object set R as the search results corresponding to the search query (input object y) to the terminal device or the like that performed the search.

[2. Second embodiment]
The second embodiment will now be described. In the second embodiment, neighboring objects are grouped by clustering or the like, and each group (hereinafter also referred to as a "blob") is subjected to distance calculation at once. That is, in the second embodiment, batch distance calculation is performed in units of blobs. Note that descriptions of points similar to those in the first embodiment will be omitted as appropriate. In the second embodiment, the information processing system 1 includes an information processing device 100A instead of the information processing device 100.

[2-1. Information processing〕
First, an overview of information processing according to the second embodiment will be explained using FIG. 12. FIG. 12 is a diagram illustrating an example of information processing according to the second embodiment.

In the example of FIG. 12, it is assumed that the information processing device 100A has already acquired a graph GR21 as shown in the spatial information SP21. For example, the spatial information SP21 in FIG. 1 may be a Euclidean space. For example, it is assumed that the spatial information SP21 corresponds to the number of dimensions of the vector of the object, and is a multidimensional space such as 100 dimensions or 1000 dimensions. Note that the example of FIG. 12 will be explained based on a diagram in which direct product quantization is not performed to simplify the explanation, but vector (space) may be subjected to direct product quantization as in FIG. 1. .

First, each piece of information shown in FIG. 12 will be explained. Dotted lines connecting white circles (◯) that are nodes in the spatial information SP21 in FIG. 12 indicate edges connecting the nodes. In FIG. 12, an undirected edge is shown as an example as in FIG. 1, but the edges of the graph GR21 are not limited to undirected edges and may be directed edges. Note that notes and edges are the same as those in FIG. 1, so detailed explanations will be omitted.

In the spatial information SP21 of FIG. 12, an area surrounded by straight lines is a region in which nearby objects are grouped together (grouped) by clustering or the like, and the area is called a blob. In the example shown below, blobs are the processing unit for batch distance calculation. FIG. 12 shows a case where clustering is performed into ten regions (blobs) of blobs BL1 to BL10. For example, blob BL1 is a blob to which multiple nodes such as nodes N7, N9, N85, and N126 belong. Note that the black dots in the blobs BL1 to BL10 indicate the centroid (representative vector) of each blob.

Note that the blobs BL1 to BL10 are generated using various methods related to clustering and the like as appropriate. For example, clustering for classifying into blobs may be k-means clustering. Also, for example, clustering for classification into blobs does not use the centroid obtained at the center coordinates (average) of each cluster in one iteration (assignment) with k-means for the next iteration. Alternatively, the centroids may be replaced with the object closest to each centroid before the next iteration. In this case, the centroid of the final cluster obtained by k-means is an existing object, that is, a node. This increases the tendency for clusters (blobs) and edges (neighboring nodes) of each node to match, making it possible to further improve search performance.

In the example of FIG. 12, the information processing device 100A performs a search process using information on a plurality of classified blobs BL1 to BL10 by clustering a plurality of nodes (objects), as shown in the spatial information SP21. .

From here, a search process targeting the search query QE2 will be explained. First, the information processing device 100A obtains the search query QE2 (step S21). For example, the information processing device 100A obtains the search query QE2 from the terminal device 10 (see FIG. 4) used by the user.

Then, the information processing device 100A calculates the distance between the vector of the search query QE2 and the vector of the codebook. Although illustration is omitted, when cross product quantization is not performed, the information processing device 100A calculates the distance (difference) between the vector of the search query QE2 and the vector of the codebook for quantizing the entire vector. . The information processing device 100A then stores the distance (difference) between the vector of the search query QE2 and the vector of each codebook as codebook information TB21 in association with information for identifying each codebook. Note that when performing Cartesian product quantization, the information processing device 100A stores each subvector of the search query QE2 and the vector of each codebook, as shown in codebook information TB1 to TB4 of FIG. 2, as in FIG. Calculate the distance (difference) between Note that the calculation of the distance between the vector of the search query and the vector of the codebook is the same as in FIGS. 1, 2, etc., and therefore detailed explanation will be omitted.

The information processing device 100A executes a search process targeting the search query QE2 (step S22). The information processing device 100A obtains a search result for the search query QE2 by performing a search process as shown in FIG. 16 using the graph GR21 for the search query QE2. Details of the search process shown in FIG. 16 will be described later. The information processing device 100A performs a search process on the search query QE2, thereby extracting the searched nodes as nodes in the vicinity of the search query QE2.

In the search process targeting the search query QE2, the information processing device 100A selects a predetermined node from among the nodes as the node (starting point node) that becomes the starting point for the search of the graph GR21. In the example of FIG. 12, it is assumed that the information processing apparatus 100A selects node N9 as the starting point node. In the example of FIG. 12, the information processing device 100A starts the search process with node N9 as the processing target, for example.

Here, in the search processing targeting the search query QE2, the information processing device 100A parallelizes and calculates distances between the search query and a plurality of nodes belonging to the blob to which the processing target node belongs (step S23). In FIG. 12, as shown in the batch processing information LT2, the information processing device 100A uses the search query QE2 for nodes N7, N9, N85, N126, etc. that belong to the blob BL1 to which the processing target node N9 belongs. Calculate the distance in parallel.

The information processing device 100A uses the distance between each codebook indicated by the codebook information TB21 and the search query QE2 to calculate the distance between each of the nodes N7, N9, N85, and N126 and the search query QE2. For example, the information processing device 100A refers to the codebook information TB21 and calculates the distance of the codebook corresponding to the vector of the node N7 as the distance between the node N7 and the search query QE2. Note that distance calculation when cross-product quantization is performed is the same as in FIGS. 1, 2, etc., and detailed description thereof will be omitted.

Note that, as in the case of the information processing apparatus 100, the units that can be collectively processed are determined based on the specifications of the information processing apparatus 100A. For example, if the number that the information processing device 100A can process at once using SIMD (unit that can be processed at once) is "4", the distances are computed in parallel for four nodes, as in FIG. 1. For example, the information processing device 100A collectively calculates the distance between each of the nodes N7, N9, N85, and N126 and the search query QE2 by parallelizing SIMD calculations. Thereby, the information processing device 100A can speed up the distance calculation by parallelizing the distance calculation, and can perform efficient search processing. In addition, the information processing device 100A uses the graph GR21 to trace nodes (for example, nodes N12, N64) to which edges from the node N7 are connected, and target blobs (for example, blobs BL2, BL3) to which these nodes belong. The search process is performed by performing batch distance calculation. Note that in the search process, the information processing device 100A suppresses the same blob from being repeatedly subjected to batch distance calculation, but details will be described later.

Note that, in order to simplify the explanation, FIG. 12 shows a case where distances are computed in parallel for four nodes, but the number of parallelized nodes is determined based on the specifications of the information processing device 100A. . This point is also similar to FIGS. 1, 2, etc., so detailed explanation will be omitted. Further, the information processing device 100A may use transposed data similarly to the example shown in FIG.

As described above, the information processing device 100A calculates the approximate distance (second distance) between the vector quantized vector of each node and the search query, and performs the search process using the second distance. This enables efficient search processing. Further, the information processing device 100A can perform efficient search processing by collectively performing distance calculations for a number of nodes that can be parallelized. As described above, the information processing apparatus 100A can perform efficient search processing by collectively calculating blobs as a unit of batch processing.

In the search process, the information processing device 100A traces the graph GR21 and performs the above-described process on the nodes to be processed, thereby extracting the searched nodes as nodes in the vicinity of the search query QE2. For example, like the information processing device 100, the information processing device 100A obtains search results using either the first method or the second method described above.

The above-described processing is merely an example, and the information processing device 100A may perform search processing using various information and methods for efficient search processing. Regarding this point, each item will be explained in detail. For example, the information processing device 100A performs clustering and generates blobs so that the number of nodes belonging to each blob becomes the number that the information processing device 100 can process at once using SIMD (batch processable unit). Good too.

Further, the information processing device 100A may use information regarding blobs to suppress redundant processing. The information processing device 100A may have a blob distance calculation flag (hereinafter also simply referred to as a "flag") that indicates whether distance calculation has been performed for each blob, and a table that indicates which blob each object belongs to. good. In this case, the information processing device 100A may manage whether each blob has been processed using a flag. For example, immediately before processing neighboring nodes one by one, the information processing device 100A identifies the blob to which the node belongs based on a table, determines whether the blob has already been subjected to batch distance calculation using a flag, and if the distance has not yet been calculated, performs batch calculation. You may also perform processing on individual nodes. This point will also be explained with reference to FIGS. 15 and 16. Thereby, the information processing device 100A can prevent duplicate processing from being performed.

[2-2. Configuration of information processing device]
Next, the configuration of the information processing apparatus 100A according to the second embodiment will be described using FIG. 13. FIG. 13 is a diagram illustrating a configuration example of an information processing apparatus according to the second embodiment. As shown in FIG. 13, the information processing device 100A includes a communication section 110, a storage section 120A, and a control section 130A. Note that in the information processing device 100A, descriptions of the same points as the information processing device 100 will be omitted as appropriate.

(Storage unit 120A)
The storage unit 120A is realized by, for example, a semiconductor memory device such as a RAM or a flash memory, or a storage device such as a hard disk or an optical disk. As shown in FIG. 13, the storage unit 120A according to the second embodiment includes an object information storage unit 121, a graph information storage unit 122, a quantization information storage unit 123A, a codebook information storage unit 124, and a blob information storage unit 121. It has an information storage section 125.

(Quantization information storage unit 123A)
The quantization information storage unit 123A according to the second embodiment stores various information regarding allocation processing. FIG. 14 is a diagram illustrating an example of a quantization information storage unit according to the second embodiment. In the example of FIG. 14, the quantization information storage unit 123A has items such as "node ID", "object ID", "blob ID", and "quantization information".

"Node ID" indicates identification information for identifying each node (object) in the graph. Moreover, "object ID" indicates identification information for identifying an object. Note that when the node ID and object ID are common, the object ID is stored in the "node ID", and the quantization information storage unit 123A does not need to include the item "object ID".

“Blob ID” indicates information for identifying a blob to which a node (object) belongs.

Moreover, "quantization information" indicates information on quantized vectors of each node (object). For example, "quantization information" includes information such as "element" and "codebook ID." "Element" indicates the arrangement of the corresponding object in the vector. In the example of FIG. 14, the "element" includes "#1", "#2", "#3", and "#4". In this case, the vector of each node (object) is divided into four parts, and each divided partial vector is quantized by the codebook.

The example in FIG. 14 indicates that the node N1 (object OB1) belongs to the blob (blob BL9) identified by the blob ID "BL9". For example, when a vector is divided into four by Cartesian product quantization, the quantization information of node N1 stores the same information as the four codebooks indicating the quantization information of node N1 shown in FIG. Further, for example, when cross-product quantization is not performed, the quantization information of the node N1 stores information indicating one codebook for quantizing the vector of the node N1.

Note that the quantization information storage section 123A is not limited to the above, and may store various information depending on the purpose.

(Blob information storage unit 125)
The blob information storage unit 125 according to the second embodiment stores information regarding blobs. For example, the blob information storage unit 125 stores various information that identifies objects associated with each blob. FIG. 15 is a diagram illustrating an example of a blob information storage unit according to the second embodiment. In the example of FIG. 15, the blob information storage unit 125 includes items such as "blob ID", "node ID", and "vector information".

“Blob ID” indicates identification information for identifying a blob. Further, "node ID" indicates a node (object) associated with the blob identified by the blob ID. "Vector information" indicates vector information of a blob. For example, "vector information" indicates the vector corresponding to the centroid of the blob.

The example shown in FIG. 15 indicates that the nodes (objects) associated with the blob (blob BL1) identified by the blob ID "BL1" are nodes N7, N9, N85, N126, etc. Blob BL1 indicates that multidimensional vector information of "51, 4, 102, 33..." is associated with it.

Further, the nodes (objects) associated with the blob (blob BL9) identified by the blob ID "BL9" are nodes N1, N4, N5, etc. Blob BL9 indicates that multidimensional vector information of "12, 55, 12, 6..." is associated with it.

Note that the blob information storage unit 125 is not limited to the above, and may store various types of information depending on the purpose. The blob information storage unit 125 may store a flag indicating whether each blob is a target of distance calculation processing. For example, the blob information storage unit 125 stores a value indicating that the blob has not been processed as a target for distance calculation (also referred to as "unprocessed flag value") and a value indicating that it has been processed as a target for distance calculation ("processed flag value"). (also referred to as "flag value") as the value of each blob's flag. For example, at the start of the search process, the blob information storage unit 125 sets the value of the flag of each blob to a value (for example, 0) indicating that the distance calculation for that blob has not been processed, and The value of the flag of the blob that has become , is changed to a value (for example, 1) that indicates that distance calculation has been completed.

For example, the information processing device 100A refers to the value of the flag of the blob that is the processing target and determines whether to perform the batch distance calculation process for that blob. The information processing device 100A refers to the flag value of each blob, and if the blob's flag value is an unprocessed flag value, the information processing device 100A determines that the blob has not yet been processed in the batch distance calculation, and processes the blob. Perform bulk distance calculation of nodes belonging to . On the other hand, if the value of the flag of the blob is the processed flag value, the information processing device 100A determines that the blob has been subjected to the batch distance calculation, and performs the batch distance calculation of the nodes belonging to the blob. Not performed.

(Control unit 130A)
Returning to the explanation of FIG. 13, the control unit 130A is a controller, and includes various programs (information processing programs) stored in a storage device inside the information processing device 100A, for example, by a CPU, MPU, GPU, etc. (corresponding to one example) is realized by being executed using RAM as a work area. Further, the control unit 130A is a controller, and is realized by, for example, an integrated circuit such as an ASIC or an FPGA.

As shown in FIG. 13, the control unit 130A includes an acquisition unit 131, a generation unit 132A, a search processing unit 133A, and a provision unit 134, and realizes or executes information processing functions and operations described below. do. Note that the internal configuration of the control unit 130A is not limited to the configuration shown in FIG. 13, and may be any other configuration as long as it performs information processing to be described later.

(Generation unit 132A)
The generation unit 132A generates various information similarly to the generation unit 132.

The generation unit 132A may generate information related to vector quantization as shown in the quantization information storage unit 123. For example, the generation unit 132 generates information indicating that the node N1 (object OB1) belongs to the blob BL9. Further, the generation unit 132 may generate information regarding blobs as shown in the blob information storage unit 125. For example, the generation unit 132 generates information indicating that nodes N7, N9, N85, N126, etc. belong to blob BL1. Note that when the information processing device 100 acquires the information shown in the graph information storage unit 122, quantization information storage unit 123A, codebook information storage unit 124, and blob information storage unit 125 from an external device such as the information providing device 50, The information processing device 100 does not need to include the generation unit 132A.

(Search processing unit 133A)
Similar to the search processing unit 133, the search processing unit 133A performs various processes related to search processing.

The search processing unit 133A uses information on a plurality of blobs in which a plurality of objects are classified in a search process for searching for nodes near a search query in a group of nodes corresponding to each of a plurality of objects. Distances between a plurality of nodes belonging to a blob and a search query are calculated using vector quantized vector information of a plurality of nodes. The search processing unit 133A parallelizes and collectively calculates distances between a plurality of nodes and a search query. The search processing unit 133A performs parallel processing to calculate the distance between the search query and a plurality of nodes whose number is determined based on the specifications of the information processing device 100A.

The search processing unit 133A calculates the distance between the search query and a plurality of nodes belonging to one blob adjacent to the blob to which the search query applies. The search processing unit 133A calculates distances between a plurality of nodes and a search query in a search process that searches for nodes in the vicinity of a search query using a graph in which a plurality of nodes are connected by edges. The search processing unit 133A calculates the distance between the search query and a plurality of nodes belonging to one blob to which a connection node to which edges from the target node to be processed belong in the search process.

The search processing unit 133A calculates the distance between the search query and a plurality of nodes belonging to a blob other than the blob to which the target node belongs. In a search process in which the search processing unit 133A searches for nodes near the search query using a blob graph connected from each of the plurality of nodes to other blobs other than the blob to which each of the plurality of nodes belongs, Calculate the distance between multiple nodes and a search query. In the graph before conversion in which a plurality of nodes are connected by edges, the search processing unit 133A searches for a blob to which a node to which an edge is connected from one of the plurality of nodes belongs, other than the blob to which the one node belongs. In a search process that searches for nodes near a search query using a blob graph, which is a converted graph generated by connecting edges from one node to a blob, multiple nodes and a search query are Calculate the distance.

The search processing unit 133A calculates the distance between the search query and a plurality of nodes belonging to one blob, which is a blob in which edges from the target node to be processed are connected in the search process. The search processing unit 133A uses information indicating a processed blob, which is a blob that has already been a processing target, to determine whether one blob is to be processed. If the one blob is a processed blob, the search processing unit 133A determines that the one blob is not to be processed.

The search processing unit 133A targets a predetermined node among the nodes to be processed in the search process. Calculate distance. In the search process, the search processing unit 133A extracts a second number of nodes, which is a larger number than the first number of nodes to be extracted as nodes in the vicinity of the search query, as neighboring candidate nodes, and selects a second number of nodes as neighboring candidate nodes. Calculate distance. The search processing unit 133A extracts a first number of nodes from among the nearby candidate nodes having the shorter second distance as nodes in the vicinity of the search query.

The search processing unit 133A calculates a second distance for nodes whose first distance is within a predetermined threshold. The search processing unit 133A calculates a second distance for nodes within the search range indicating the target range to be extracted as nearby nodes. The search processing unit 133A calculates a second distance for nodes within a search range indicating a target range of the search process.

[2-3. Search processing example]
Here, an example of the search process according to the second embodiment will be described using FIG. 16 as an example. FIG. 16 is a flowchart illustrating an example of search processing according to the second embodiment. The search processing described below is performed by the search processing unit 133A of the information processing device 100A. Note that descriptions of points similar to those of the first embodiment, such as FIG. 11, will be omitted as appropriate. For example, in FIG. 16, the same steps as those in FIG. 11 are given the same step numbers, and the description thereof will be omitted as appropriate.

Here, the neighboring object set N(G, y) is a set of neighboring objects that are related by edges attached to node y. "G" may be predetermined graph data (for example, graph GR21 shown in spatial information SP21, etc.). For example, the information processing device 100A executes k-nearest neighbor search processing.

For example, the search process shown in FIG. 16 differs from the search process shown in FIG. 11 in that the process shown in step S304a is performed after step S304. In this way, the search process shown in FIG. 16 is a normal graph search, and when a blob that has not been accessed yet is reached, the distance between the object (node) in that blob and the query is calculated all at once. , performs processing on each node.

Specifically, in the search process of FIG. 16, if the distance of the blob to which the object s belongs has not yet been calculated, the information processing device 100A performs all of the following (“-” in step S304a in FIG. 16). The following four processes (hereinafter referred to as "first process" to "fourth process") are executed (step S304a). In step S304a, the information processing device 100A executes a process (first process) of calculating the distances of objects within the blob at once. Furthermore, in step S304a, the information processing device 100A executes a process (second process) of setting a blob distance calculation flag. Furthermore, in step S304a, the information processing apparatus 100A executes a process (third process) of storing all objects in the blob for which the batch distance calculation has been performed in C. Furthermore, in step S304a, the information processing apparatus 100A executes a process (fourth process) in which all objects of the blob are set as u and processes from step S307 to step S314 are performed one by one. After that, the information processing device 100A performs the processing from step S306 onwards.

[2-4. Modified example]
From here, modified examples will be explained. In a modification of the second embodiment, a graph including the concept of blob may be used. For example, in a modification of the second embodiment, a graph may be used in which a blob is referenced by an edge from a node. Note that descriptions of points similar to those in the first embodiment and the second embodiment will be omitted as appropriate. The information processing device 100A according to the modification includes a graph information storage section 122A instead of the graph information storage section 122.

[2-4-1. Information processing〕
First, an overview of information processing according to the modified example will be explained using FIG. 17. FIG. 17 is a diagram illustrating an example of information processing according to a modification.

The information processing device 100A converts the graph GR21 in which nodes are connected by edges into a graph GR31 in which a blob is referenced by an edge from a node (step S31). That is, as shown in FIG. 17, the information processing device 100A generates a graph GR31, which is a graph in which the concept of a blob is introduced (blob graph), from a graph GR21, which is a normal graph.

In the example of FIG. 17, the information processing device 100A generates a graph GR31 including directed edges from nodes to blobs, as shown in spatial information SP31. Note that in the graph GR31, edges between nodes are deleted, and the graph GR31 includes directed edges from nodes to blobs, but does not include edges between nodes. The arrow line shown in FIG. 17 indicates a directed edge from the node at the base of the arrow to the blob at the base of the arrow. That is, the arrow line shown in FIG. 17 indicates a directed edge with the node at the base of the arrow as the reference source and the blob at the tip of the arrow as the reference destination. For example, FIG. 17 shows that edges from node N1 to two blobs, blob BL8 and blob BL10, are connected.

For example, the information processing device 100A converts the edge of each node from an edge to a connection node (nearby node) to an edge to a blob to which the connection node belongs. Note that the information processing device 100A does not generate edges to the blob to which each node itself belongs. In the example of FIG. 17, since the node N9 and the node N126 belong to the same blob BL1, the information processing device 100A does not generate an edge from the node N9 to the blob BL1 and an edge from the node N126 to the blob BL1. On the other hand, in the example of FIG. 17, since the node N9 and the node N18 belong to different blobs BL1 and BL4, the information processing device 100A generates an edge from the node N9 to the blob BL4 and an edge from the node N18 to the blob BL1. .

In the example of FIG. 17, the information processing device 100A performs a search process using a graph GR31 including directed edges from nodes (objects) to blobs, as shown in spatial information SP31. For example, the information processing device 100A obtains the search result of the search query QE2 by performing a search process as shown in FIG. 19 using the graph GR21 for the search query QE2. Details of the search process shown in FIG. 19 will be described later. Note that the search process in FIG. 17 is similar to FIG. 12 in that the search process is performed using the concept of a blob, and is the same as the search process shown in FIG. 12 except for the process related to edges.

[2-4-2. graph〕
Next, an outline of a graph according to a modified example will be explained using FIG. 18. FIG. 18 is a diagram illustrating an example of a graph information storage unit according to a modification. For example, the graph information storage unit 122A shown in FIG. 18 stores a graph (blob graph) in which the concept of blob is introduced. The graph information storage unit 122A shown in FIG. 18 has items such as "node ID", "object ID", and "blob information". In this way, the graph information storage unit 122A according to the modified example differs from the graph information storage unit 122 of FIG. 7 in that it includes “blob information” instead of “connection node information”. Note that descriptions of the same points as those of the graph information storage unit 122 in FIG. 7 will be omitted as appropriate.

Moreover, "blob information" indicates information regarding a blob (a referenced blob) that can be traced from a corresponding node. For example, "blob information" includes information such as "reference destination." “Reference destination” indicates information for identifying a reference destination (blob) that is connected by an edge and can be traced from that node. That is, in the example of FIG. 18, a reference destination (blob) that can be traced from the node by an edge is registered in association with a node ID (object ID) that identifies the node. Note that the "blob information" may include information (edge ID) for identifying an edge connected to a reference destination.

The example in FIG. 18 indicates that the node (node N1) identified by the node ID "N1" corresponds to the object (target) identified by the object ID "OB1". Furthermore, an edge is connected from the node N1 to the blob (blob BL8) identified by the blob ID "BL8", indicating that it is possible to trace from the node N1 to the blob BL8. Further, an edge is connected from the node N1 to the blob (blob BL10) identified by the blob ID "BL10", indicating that the blob BL10 can be traced from the node N1.

Further, the node (node N2) identified by the node ID "N2" corresponds to the object (target) identified by the object ID "OB2". Furthermore, an edge is connected from the node N2 to the blob (blob BL5) identified by the blob ID "BL5", indicating that the blob BL5 can be traced from the node N1.

Note that the graph information storage unit 122A is not limited to the above, and may store various information depending on the purpose.

The graph may also include a program module that receives a query as input, searches for nodes by tracing edges in the graph, and extracts and outputs nodes similar to the query. That is, the graph may be assumed to be used as a program module that performs search processing using the graph. For example, the graph GR31 may be a program that, when vector data is input as a query, extracts nodes corresponding to vector data similar to the vector data from the graph and outputs them. For example, the graph GR31 may be data used as a program module that searches for similar images corresponding to a query image. For example, based on an input query, the graph GR31 causes the computer to extract and output nodes similar to the query in the graph.

[2-4-3. Search processing example]
An example of the search process according to the modified example will now be described using FIG. 19 as an example. FIG. 19 is a flowchart illustrating an example of a search process according to a modification. The search processing described below is performed by the search processing unit 133A of the information processing device 100A. Note that descriptions of points similar to those of the first embodiment or the second embodiment, such as FIGS. 11 and 16, will be omitted as appropriate. For example, in FIG. 19, the same steps as those in FIG. 11 are given the same step numbers, and the description thereof will be omitted as appropriate.

In FIG. 19, the set N(G, y) is different from the search processing in FIGS. 11 and 16 in that it is a set of blobs related by edges attached to node y. In FIG. 19, N(G, s) and C are blob sets. That is, "nearby object set N" in FIG. 11 is replaced with "blob set N" in FIG. 19, and "object set C" in FIG. 11 is replaced with "blob set C" in FIG. . Furthermore, the word "object" in the process related to the change to a blob can be read as "blob" as appropriate. "G" may be graph (blob graph) data (for example, graph GR31 shown in spatial information SP31) into which the concept of blob has been introduced. For example, the information processing device 100A executes k-nearest neighbor search processing.

For example, in the search process shown in FIG. 19, the process in step S306 from FIG. 11 is changed as follows. In the search process in FIG. 19, the information processing device 100A selects one blob from N(G, s) that is not included in the blob set C, stores the selected blob in the blob set C, and stores the selected blob in the blob set C. The objects within the blob are collectively calculated, and the set thereof is set as B (also referred to as "object set B within the target blob") (step S306).

Furthermore, the search process shown in FIG. 19 differs from the search process shown in FIG. 11 in that the process shown in step S306a is performed after step S306. Specifically, in the search process of FIG. 19, the information processing device 100A selects one object u from the target blob object set B (step S306a). After that, the information processing device 100A performs the processing from step S307 onwards.

The search process shown in FIG. 19 is different from the search process shown in FIG. 11 in that the process shown in step S314a is performed before step S315. Specifically, in the search process of FIG. 19, the information processing device 100A determines whether all objects have been selected from object set B in the target blob (step S314a).

If all objects have been selected from object set B within the target blob (step S314a: Yes), the information processing device 100A performs the process of step S315. On the other hand, if all objects have not been selected from object set B within the target blob (step S314a: No), the information processing device 100A returns to step S306a and repeats the process.

[2-4-4. Other examples of generation processing]
In the above example, a case was shown in which a blob graph is generated by converting the edge of each node from an edge to a connected node to an edge to a blob to which the connected node belongs, but the information processing device 100A converts various information. It may be used as appropriate to generate a blob graph. For example, the information processing device 100A may generate a blob graph using a neighborhood search index (also simply referred to as an "index") that searches for multiple objects. Note that descriptions of points similar to those described above will be omitted as appropriate.

An example of generating a blob graph will be described using FIG. 20. FIG. 20 is a diagram illustrating another example of information processing according to the modification. In addition, although the case where a graph is used as an example of an index is demonstrated below as an example, the index may be any type of index as long as it searches for a plurality of objects. For example, the index may be an index using a hash-related description such as a hash table (hash index), an index having a tree structure (tree index), or the like.

Further, information indicating an object or a node corresponding to the object (also referred to as an "object node") may be described as first information, and information used to classify a plurality of objects may be described as second information. Further, information indicating a blob may be described as third information, and a graph in which object nodes are connected by edges may be described as fourth information. For example, the graph GR21 in FIG. 20 has node N1, which is an object node, connected by edges, and corresponds to the fourth information. Also, unlike the fourth information, graph GR31 in FIG. 17 and graph GR32 in FIG. 20 are blob graphs (also referred to as "group graphs") in which object nodes and blobs are connected at edges. In other words, the group graph is a graph in which edges are connected from object nodes to blobs, the object nodes are reference sources, and the blobs (groups) are reference destinations.

The information processing device 100A uses the graph GR21 in which object nodes are connected by edges as an index (fourth information) to generate a graph GR32 that is a group graph in which blobs are referenced by edges from object nodes (step S41). In FIG. 20, the information processing device 100A generates a graph GR32 including directed edges from nodes to blobs, as shown in spatial information SP32.

For example, the information processing device 100A selects one object from a plurality of objects, searches for neighboring objects of the one object using the graph GR21, and determines which neighboring objects belong from one object node corresponding to the one object. A graph GR32 is generated by a connection process that connects edges to other groups. For example, the information processing device 100A randomly selects one object from a plurality of objects, performs a concatenation process on the selected object, and generates the graph GR32.

For example, the information processing device 100A performs the following processes (1-1) to (1-4) to generate the graph GR32. Note that the object may be read as a node corresponding to the object.

(1-1): Obtain object NX randomly (selection)
(1-2): Search for a constant number of k neighboring objects of object NX (1-3): Obtain the blob to which each neighboring object belongs (1-4): Generate an edge to the blob obtained from object NX

For example, in process (1-1), the information processing device 100A randomly acquires one object NX from a plurality of objects stored in the object information storage unit 121.

For example, in process (1-2), the information processing device 100A performs a search process as shown in FIG. is extracted as a neighboring object of object NX.

For example, in process (1-3), the information processing device 100A refers to blob information stored in the blob information storage unit 125 and obtains blobs associated with neighboring objects (nodes) of object NX.

For example, in process (1-4), the information processing device 100A generates an edge to the blob acquired from the object NX. For example, the information processing device 100A generates the graph GR32 by associating information indicating the acquired blob with the object NX in the quantization information storage unit 123A.

For example, the information processing device 100A repeatedly performs the above processes (1-1) to (1-4) until there are no more objects to be processed, and generates the graph GR32. Note that when the blob shown in the spatial information SP21 is generated by a search, the graph GR32 may be generated by using the search results when the blob is generated. Note that an example of generating blobs by search will be described later.

For example, the information processing device 100A performs a search process using a graph GR32 including directed edges from nodes (objects) to blobs, as shown in spatial information SP32. For example, the information processing device 100A obtains the search result of the search query QE2 by performing a search process as shown in FIG. 19 using the graph GR32 for the search query QE2.

Further, in the above example, a case was shown in which blobs were generated by clustering, but blobs are not limited to clustering, and may be generated by various methods. For example, the information processing device 100A may generate information indicating a blob (third information) using second information such as an index.

The information processing device 100A uses the graph GR21 as an index as shown in FIG. 21 to generate third information indicating a blob. FIG. 21 is a diagram illustrating an example of blob information generation processing. Spatial information SP20 in FIG. 21 shows a state before the blob information in spatial information SP21 is generated, and graph GR21 in FIG. 21 is a graph similar to graph GR21 in FIG. 20. Note that the process shown in FIG. 21 shows a part of the first method described later, and the details will be described later.

For example, the information processing device 100A generates third information that classifies a plurality of objects into a plurality of blobs by searching using the graph GR21. For example, the information processing device 100A selects one node (also referred to as a "processing target node") from a plurality of nodes in the graph GR21, and selects at least one of the neighboring nodes that are nodes connected to the processing target node by an edge. The third information is generated by a classification process of classifying a group of objects corresponding to each of the classification target node, which is a node of the section, and the processing target node into one group. For example, the information processing device 100A selects a processing target node from a plurality of nodes by a predetermined process, performs a classification process on the selected processing target node, and generates third information.

For example, the information processing device 100A may select a processing target node based on the number of neighboring nodes (connection nodes) connected to the node by edges. For example, the information processing apparatus 100A may select processing target nodes in descending order of the number of neighboring nodes. A method of selecting processing target nodes in order of increasing number of neighboring nodes and generating third information indicating a blob is also referred to as a "first method." Further, for example, the information processing apparatus 100A may select processing target nodes in descending order of the number of neighboring nodes (connected nodes). A method of selecting processing target nodes in order of decreasing number of neighboring nodes and generating blobs is also referred to as a "second method." Further, for example, the information processing apparatus 100A may randomly select a processing target node. A method of randomly selecting processing target nodes and generating third information indicating a blob is also referred to as a "third method." Below, the outline of each processing procedure of the first method, second method, and third method will be explained.

[2-4-4-1. 1st method]
First, the first method will be explained. For example, in the case of the first method, the information processing device 100A performs the following processes (2-1) to (2-3) to generate third information indicating a blob.

(2-1): Obtain the node ND from the one with the largest number of neighboring nodes (number of edges) for each node (2-2): Set the neighboring nodes of the node ND and the node ND as one blob (2-3): Delete blob nodes and edges of those nodes

For example, the information processing device 100A performs processes (2-1) to (2-3) on all nodes.

For example, in process (2-1), the information processing device 100A refers to the graph GR21 and sequentially acquires one node ND. The information processing device 100A may obtain one node ND using list information of nodes arranged in descending order of the number of neighboring nodes.

For example, in process (2-2), the information processing device 100A refers to the graph to be processed (graph GR21, etc.) and generates information regarding the neighboring nodes of the node ND and the node ND as one blob. Note that the information processing device 100A may set neighboring nodes that are less than or equal to a constant K as blobs instead of setting all neighboring nodes as blobs. For example, instead of making all the neighboring nodes of the node ND into a blob together with the node ND, the information processing device 100A may make K neighboring nodes and the node ND among the neighboring nodes into one blob.

For example, in process (2-3), the information processing device 100A deletes the node that was made into a blob in process (2-2) and the edges of that node so that nodes are not selected redundantly. For example, the information processing device 100A deletes the blob (edge and) node if the node is a neighboring node of another node. Note that the above method is only an example, and the process (2-3) is not limited to the method of deleting edges and the like, but may be any method as long as it is possible to generate blobs. For example, the information processing device 100A may generate a blob by maintaining a list of nodes that already belong to the blob and managing nodes to be excluded. For example, the information processing device 100A may manage nodes that already belong to a blob by setting a predetermined flag (such as a deletion flag) on the node that is a blob. In this way, when using flags, the information processing apparatus 100A refers to the flags of each node during processing and determines whether that node is to be processed.

A specific example of the first method described above will be explained using FIG. 21. In FIG. 21, the information processing device 100A acquires the node N7, which has the maximum number of neighboring nodes of four, from the graph GR21 (step S51). In FIG. 21, the node N1 in the graph GR21 also has four neighboring nodes, but the information processing device 100A acquires the node N7. Note that when there are multiple nodes with the same number of neighboring nodes, the information processing device 100A may randomly acquire nodes from the multiple nodes.

Then, the information processing device 100A sets four nodes N9, N12, N54, and N126, which are neighboring nodes of the node N7, and five nodes including the node N7, as one blob BL11 (step S52). The information processing device 100A generates third information BLT1 indicating that five nodes N7, N9, N12, N54, and N126 belong to one blob BL11. The information processing device 100A then deletes the five nodes N7, N9, N12, N54, and N126 and the edges of each node from the graph GR21. As a result, the graph GR21 is updated to the graph GR21-1, as shown in the spatial information SP20-1.

Then, the information processing device 100A obtains the node N1 having four neighboring nodes from the graph GR21 (step S53). Then, the information processing device 100A sets four nodes N4, N5, N88, and N99, which are neighboring nodes of the node N1, and five nodes including the node N1, as one blob BL12 (step S54). The information processing device 100A generates third information BLT2 indicating that five nodes N1, N4, N5, N88, and N99 belong to one blob BL12. Then, the information processing device 100A deletes the five nodes N1, N4, N5, N88, and N99 and the edges of each node from the graph GR21-1.

The information processing device 100A repeats the above-described processing until there are no more nodes to be processed, and generates third information indicating a plurality of blobs.

[2-4-4-2. Second method]
Next, the second method will be explained. For example, in the case of the second method, the information processing device 100A performs the following processes (3-1) to (3-3) to generate third information indicating a blob.

(3-1): Obtain nodes ND from the smallest number of neighboring nodes (number of edges) for each node (3-2): Set neighboring nodes of node ND and node ND as one blob (3-3): Delete blob nodes and edges of those nodes

For example, the information processing device 100A performs processes (3-1) to (3-3) on all nodes.

For example, in process (3-1), the information processing device 100A refers to the graph GR21 and sequentially acquires one node ND. The information processing device 100A may obtain one node ND using list information of nodes arranged in descending order of the number of neighboring nodes.

For example, in the process (3-2), the information processing device 100A refers to the graph to be processed (graph GR21, etc.) and generates information regarding the neighboring nodes of the node ND and the node ND as one blob. Note that the information processing device 100A may set neighboring nodes that are less than or equal to a constant K as blobs instead of setting all neighboring nodes as blobs. For example, instead of making all the neighboring nodes of the node ND into a blob together with the node ND, the information processing device 100A may make K neighboring nodes and the node ND among the neighboring nodes into one blob.

For example, in process (3-3), the information processing apparatus 100A deletes the node that was made into a blob in process (3-2) and the edges of that node so that nodes are not selected twice. For example, the information processing device 100A deletes the blob (edge and) node if the node is a neighboring node of another node. Note that the second method is the same as the first method except that the order of node acquisition is from the smallest number of neighboring nodes (number of edges), so a description of specific examples and the like will be omitted. For example, like the first method, the process (3-3) is not limited to the method of deleting edges and the like, but may be any method as long as it is possible to generate blobs. For example, similar to the first method, the information processing device 100A may generate a blob by maintaining a list of nodes that already belong to the blob and managing nodes to be excluded.

[2-4-4-3. Third method]
Next, a third method will be explained. For example, in the case of the third method, the information processing device 100A performs the following processes (4-1) to (4-3) to generate third information indicating a blob.

(4-1): Randomly acquire nodes ND in sequence (4-2): Make neighboring nodes of node ND and node ND one blob (4-3): Delete the blob node and the edges of that node

For example, the information processing device 100A performs processes (4-1) to (4-3) on all nodes. Note that the third method is similar to the first method and the second method, except that the order of node acquisition is random, and detailed description thereof will be omitted. For example, the information processing device 100A may generate the third information in random order (third method) if the number of edges is the same.

[2-4-4-4. Fourth method]
Note that the first to third methods described above are merely examples, and the information processing apparatus 100A may generate blobs by various processes. For example, the information processing device 100A may generate a blob through a search using a neighborhood search index that searches for multiple objects. The method of generating blobs by searching using an index is also referred to as a "fourth method."

The fourth method will be explained below. Note that descriptions of points similar to those described above will be omitted as appropriate. In the following, a case will be described in which a graph is used as an example of the index, but as described above, the index may be of any type as long as it searches for a plurality of objects. For example, the index may be any index such as a hash index or a tree index.

From here, a case will be described in which the information processing device 100A generates third information indicating a blob by searching using the graph GR21 as an index as shown in FIG. 21. For example, the information processing device 100A performs the following processes (5-1) to (5-4) to generate the graph GR32.

(5-1): Obtain object NX randomly (selection)
(5-2): Search for a constant number of k neighboring objects of object NX (5-3): Make neighboring objects of object NX and object NX one blob (5-4): Execute processing to avoid duplicate processing

For example, in process (5-1), the information processing device 100A randomly acquires one object NX from the graph GR21.

For example, in process (5-2), the information processing device 100A performs a search process as shown in FIG. 11 using the graph GR21 with the object NX as the target (query). is extracted as a neighboring object of object NX. Note that the information processing device 100A may scan (search) all the objects and extract neighboring objects of the object NX.

For example, in process (5-3), the information processing device 100A generates information that makes the neighboring nodes of the neighboring objects of the object NX and the object NX one blob.

For example, in process (5-4), the information processing device 100A executes a process for avoiding duplicate processing so that one object does not belong to multiple blobs due to duplicate searches or the like. For example, the information processing device 100A deletes the blob object from the index. For example, the information processing device 100A deletes the blob object from the graph GR21.

Note that the information processing apparatus 100A is not limited to the above method, and may perform processing for avoiding duplicate processing using any method as long as one object does not belong to multiple blobs. For example, the information processing device 100A may set a predetermined flag (such as a deletion flag) on an object that has been made into a blob, thereby preventing objects that already belong to the blob from becoming processing targets. In this way, when using flags, the information processing apparatus 100A refers to the flag of each object during processing and determines whether the object is to be processed. For example, when the information processing device 100A uses the search results to generate the group graph described above, the information processing device 100A may perform processing to avoid duplicate processing using a method using a flag.

[2-4-5. Information processing device according to modification]
In the information processing device 100A according to the modification of the second embodiment, the acquisition unit 131 also performs the following processing. The acquisition unit 131 acquires first information indicating a plurality of objects to be searched for data, and second information used to classify the plurality of objects. The acquisition unit 131 acquires second information that is an index that searches for a plurality of objects. The acquisition unit 131 acquires second information that is a graph in which a plurality of nodes corresponding to each of a plurality of objects are connected by edges. The acquisition unit 131 acquires fourth information that is an index that searches for multiple objects. The acquisition unit 131 acquires fourth information that is a graph in which a plurality of object nodes corresponding to each of a plurality of objects are connected by edges.

Furthermore, in the information processing apparatus 100A according to the modification of the second embodiment, the generation unit 132A also performs the following processing. The generation unit 132A uses the second information acquired by the acquisition unit 131 to classify the plurality of objects indicated by the first information, and classifies the plurality of objects used for batch processing in the search process targeting the plurality of objects. Third information indicating the group is generated. The generation unit 132A uses the second information to generate third information that classifies the plurality of objects into a plurality of groups. The generation unit 132A generates third information for classifying a plurality of objects into a plurality of groups by searching using the second information.

The generation unit 132A selects one node from a plurality of nodes in the graph, and generates classification target nodes that are at least some nodes among neighboring nodes that are nodes connected to the one node by edges, and the one node. The third information is generated by a classification process that classifies the object group corresponding to each of the objects into one group. The generation unit 132A generates the third information through a classification process of classifying into one group a group of objects corresponding to each of the classification target nodes, which are a predetermined number of nodes among the nodes in the vicinity of one node, and one node. generate. The generation unit 132A generates the third information through a classification process of classifying the classification target node, which is all of the neighboring nodes of one node, and the object group corresponding to each of the one node into one group.

The generation unit 132A selects one node from the plurality of nodes based on the number of neighboring nodes that are nodes connected to each of the plurality of nodes by edges, and selects one node from the plurality of nodes, and selects a classification target node of the one node and a node of the one node. The third information is generated by a classification process of classifying the corresponding object groups into one group. The generation unit 132A selects one node in descending order of the number of neighboring nodes, and performs a classification process of classifying the classification target node of one node and the object group corresponding to each of the one node into one group. Generate third information. The generation unit 132A selects one node in descending order of the number of neighboring nodes, and performs a classification process of classifying the classification target node of one node and the object group corresponding to each of the one node into one group. Generate third information.

The generation unit 132A randomly selects one node from the plurality of nodes, and generates third information by performing a classification process of classifying the classification target node of the one node and the object group corresponding to each of the one node into one group. generate. The generation unit 132A excludes the classification target node of the one node classified into the one group and the processed node group that is the one node from the graph, and performs classification using the graph after excluding the processed node group. Third information is generated by repeating the process. The generation unit 132A generates third information by repeating the classification process using the graph after excluding the processed node group until a plurality of objects are classified into any group. The generation unit 132A generates third information indicating a plurality of groups to which objects each belong to are mutually exclusive.

Using the fourth information, the generation unit 132A generates an edge from an object node, which is a node corresponding to at least some of the plurality of objects, to a group other than the group to which the object node belongs among the plurality of groups. Generates a connected group graph. The generation unit 132A selects one object from the plurality of objects, searches for neighboring objects of the one object among the plurality of objects using the fourth information, and selects one object from the one object node corresponding to the one object. , a group graph is generated by a connection process that connects edges to other groups to which neighboring objects belong. The generation unit 132A randomly selects one object from a plurality of objects, searches for neighboring objects of the one object, and creates an edge from one object node corresponding to the one object to another group to which the neighboring object belongs. A group graph is generated by the concatenation process. The generation unit 132A generates a group graph using the fourth information.

[2-4-6. Information processing flow]
Next, an information processing procedure according to a modified example will be described using FIG. 22. FIG. 22 is a flowchart illustrating an example of information processing according to a modification.

As shown in FIG. 22, the information processing device 100A acquires first information indicating a plurality of objects that are data search targets (step S401). For example, the information processing device 100A acquires first information indicating a plurality of objects from the object information storage unit 121.

The information processing device 100A also acquires second information used to classify a plurality of objects (step S402). For example, the information processing device 100A obtains a graph as the second information from the graph information storage unit 122A.

The information processing device 100A uses the second information to classify the plurality of objects indicated by the first information, and indicates a plurality of groups used for performing batch processing in a search process targeting a plurality of objects. Third information is generated (step S403). For example, the information processing device 100A generates third information indicating a plurality of blobs in which objects belonging to each blob are mutually exclusive.

[3. Third embodiment]
In the above example, there are two graphs: a graph (also called an "object graph") in which nodes corresponding to objects (object nodes) are connected by edges, and a graph (blob graph) in which object nodes and groups (blobs) are connected by edges. Although the description uses two types of graphs, the information processing system 1 may use various types of graphs. For example, unlike object graphs and blob graphs, the information processing system 1 generates a graph that connects groups (blobs) (hereinafter also referred to as a "group connected graph" or "blob connected graph"), and generates a blob connected graph. may be used for the search.

This point will be described below as a third embodiment. In the third embodiment, the information processing system 1 includes an information processing device 100B instead of the information processing device 100 or the information processing device 100A. Note that descriptions of points similar to those described above in the first embodiment, second embodiment, etc. will be omitted as appropriate.

The information processing device 100B uses the index (index information) to connect edges from a first group (first blob) to which one object belongs to a second group (second blob) to which neighboring objects of one object belong. generate a group connected graph. In the following, a case will be described in which a graph is used as an example of the index, but as described above, the index may be of any type as long as it searches for a plurality of objects. For example, the index may be any index such as a hash index or a tree index.

[3-1. Information processing〕
First, an overview of information processing according to the third embodiment will be explained using FIG. 23. FIG. 23 is a diagram illustrating an example of information processing according to the third embodiment.

In the example of FIG. 23, it is assumed that the information processing apparatus 100B has already acquired a graph GR21 as shown in the spatial information SP21. Note that the information processing device 100B may generate the graph GR21 using various techniques related to graph generation as appropriate. Since the graph GR21 is similar to the graph GR21 shown in FIG. 12 etc., the explanation will be omitted.

The information processing device 100B generates information indicating the blobs BL1 to BL10, etc. using an arbitrary method. The information processing device 100B may classify each node into one of the blobs BL1 to BL10 using arbitrary clustering such as k-means. Note that the information processing device 100B may use any method to generate blobs as long as it is capable of generating blobs. For example, the information processing device 100B may generate information indicating the blobs BL1 to BL10, etc. using any of the first to fourth methods described above.

The information processing device 100B generates a blob connection graph using the graph GR21 as shown in the spatial information SP21 and the blob information (step S61). In the example of FIG. 23, the information processing device 100B uses a graph GR41, which is a blob connection graph in which a blob (first blob) and another blob (second blob) are connected by edges, as shown in spatial information SP41. generate. The arrow line shown in FIG. 23 indicates a directed edge from the blob at the base of the arrow (first blob) to the blob at the tip (second blob). That is, the arrow line shown in FIG. 23 indicates a directed edge that uses the blob at the beginning of the arrow as a reference source and the blob at the tip of the arrow as a reference destination. For example, FIG. 23 shows that edges from blob BL1 to five blobs, blob BL2, blob BL3, blob BL4, blob BL5, and blob BL8, are connected.

For example, the information processing device 100B searches for graph GR21 and generates graph GR41 based on the search results. For example, the information processing device 100B performs the following processes (6-1) to (6-4) to generate a graph GR41 that is a blob connection graph. Note that the following objects may be replaced with nodes corresponding to the objects.

(6-1): Obtain object NX randomly (selection)
(6-2): Search constant k neighboring objects of object NX (6-3): Obtain the blob to which each neighboring object belongs (6-4): Generate an edge from the blob to which object NX belongs to the obtained blob

For example, in process (6-1), the information processing device 100B randomly acquires one object NX from a plurality of objects stored in the object information storage unit 121.

For example, in process (6-2), the information processing device 100B performs a search process as shown in FIG. is extracted as a neighboring object of object NX.

For example, in process (6-3), the information processing device 100B refers to the blob information stored in the blob information storage unit 125 and obtains a blob to which a neighboring object (node) of the object NX is associated.

For example, in process (6-4), the information processing device 100B generates an edge from the blob to which the object NX belongs to the acquired blob. The information processing device 100B generates edges from the blob to which the object NX belongs to the blobs to which all neighboring objects belong. For example, the information processing device 100B associates the information (blob ID) indicating the blob acquired as a reference destination of the blob to which the object NX belongs with the information (blob ID) indicating the blob to which the object NX belongs, and generates blob connection graph information. By registering in the storage unit 126, a graph GR41 is generated.

For example, the information processing device 100B repeatedly performs the above processes (6-1) to (6-4) until there are no more objects to be processed, and generates the graph GR41. Note that if blob Y has already been registered as a reference destination of blob The same blob may be prevented from being registered redundantly in the reference destination of each blob.

For example, the information processing device 100B performs a search process using a graph GR41 that connects blobs with edges, as shown in spatial information SP41. For example, the information processing device 100B obtains the search results for the search query QE2 by performing search processing as shown in FIGS. 27 and 28 using the graph GR41, which is a blob connection graph, for the search query QE2.

The above-described method for generating a blob connected graph is merely an example, and the information processing apparatus 100B may generate a blob connected graph using various methods. For example, the information processing device 100B may generate a blob connection graph without searching for a graph. When the blob shown in the spatial information SP21 is generated by a search, a blob connection graph such as the graph GR41 may be generated by using the search results at the time of blob generation.

For example, the information processing device 100B may generate a blob connection graph by converting the connection relationship of object nodes in the graph GR21 into the connection relationship of blobs to which the object nodes belong. For example, an edge is connected from a node N1 belonging to blob BL9 to a node N88 belonging to blob BL8 and a node belonging to blob BL10, and an edge is connected from node N4 belonging to blob BL9 to a node belonging to blob BL7. . Therefore, the information processing device 100B generates a blob connection graph in which edges from blob BL9 to three blobs, blob BL7, blob BL8, and blob BL10, are connected. In this way, the information processing apparatus 100B may generate a blob connection graph that connects blobs with edges, based on edges that connect object nodes.

[3-2. Configuration of information processing device]
Next, the configuration of the information processing apparatus 100B according to the third embodiment will be described using FIG. 24. FIG. 24 is a diagram illustrating a configuration example of an information processing apparatus according to the third embodiment. As shown in FIG. 24, the information processing device 100B includes a communication section 110, a storage section 120B, and a control section 130B. Note that in the information processing device 100B, descriptions of the same points as the information processing device 100 or the information processing device 100A will be omitted as appropriate.

(Storage unit 120B)
The storage unit 120B is realized by, for example, a semiconductor memory device such as a RAM or a flash memory, or a storage device such as a hard disk or an optical disk. As shown in FIG. 24, the storage unit 120B according to the third embodiment includes an object information storage unit 121, a graph information storage unit 122, a quantization information storage unit 123, a codebook information storage unit 124, and a blob information storage unit 121. It has an information storage section 125 and a blob connection graph information storage section 126.

(Blob connection graph information storage unit 126)
The blob connection graph information storage unit 126 according to the third embodiment stores various information regarding the blob connection graph. For example, the blob connection graph information storage unit 126 stores the generated blob connection graph. FIG. 25 is a diagram illustrating an example of a blob connection graph information storage unit according to the third embodiment. The blob connection graph information storage unit 126 shown in FIG. 25 has items such as "blob ID" and "connection blob information."

“Blob ID” indicates identification information for identifying each blob (group) in the graph. “Connection blob information” indicates information regarding a blob (reference destination blob) that can be traced from a corresponding blob. For example, "connection blob information" includes information such as "reference destination." “Reference destination” indicates information for identifying a reference destination (blob) that is connected by an edge and can be traced from the blob. That is, in the example of FIG. 25, a reference destination (blob) that can be traced from the blob by an edge is registered in association with a blob ID that identifies the blob. Note that the "connection blob information" may include information (edge ID) for identifying an edge connected to a reference destination.

In the example of FIG. 25, the blob (blob BL1) identified by the blob ID "BL1" is identified by each of the blob IDs "BL2", "BL3", "BL4", "BL5", and "BL8". Shows that edges are connected to five blobs. That is, it shows that each of the five blobs BL2, BL3, BL4, BL5, and BL8 can be traced from the blob BL1.

Note that the blob connection graph information storage unit 126 is not limited to the above, and may store various information depending on the purpose. The blob connection graph may also include a program module that receives a query as input, searches for objects (object nodes) by tracing edges in the blob connection graph, and extracts and outputs objects similar to the query. That is, the blob connection graph may be assumed to be used as a program module that performs search processing using the blob connection graph. For example, the graph GR41, which is a blob connection graph, may be a program that, when vector data is input as a query, extracts and outputs objects corresponding to vector data similar to the vector data using a blob connection graph. . For example, the graph GR41 may be data used as a program module that searches for similar images corresponding to a query image. For example, based on an input query, the graph GR41 causes the computer to extract and output objects similar to the query in the blob connection graph.

(Control unit 130B)
Returning to the explanation of FIG. 24, the control unit 130B is a controller, and includes various programs (information processing programs) stored in a storage device inside the information processing device 100B by, for example, a CPU, MPU, GPU, etc. (corresponding to one example) is realized by being executed using RAM as a work area. Further, the control unit 130B is a controller, and is realized by, for example, an integrated circuit such as an ASIC or an FPGA.

As shown in FIG. 24, the control unit 130B includes an acquisition unit 131, a generation unit 132B, a search processing unit 133B, and a provision unit 134, and realizes or executes information processing functions and operations described below. do. Note that the internal configuration of the control unit 130B is not limited to the configuration shown in FIG. 24, and may be any other configuration as long as it performs information processing to be described later.

The acquisition unit 131 according to the third embodiment acquires blob information indicating a plurality of blobs in which a plurality of objects to be searched are classified, and index information indicating an index for searching a plurality of objects. do. The acquisition unit 131 acquires index information indicating a graph in which a plurality of nodes corresponding to each of a plurality of objects are connected by edges. The acquisition unit 131 acquires blob information indicating a plurality of blobs in which a plurality of objects are classified by clustering processing. The acquisition unit 131 acquires blob information indicating a plurality of blobs, each of which has mutually exclusive objects. The acquisition unit 131 acquires a query.

(Generation unit 132B)
The generation unit 132B generates various types of information in the same way as the generation unit 132 or the generation unit 132A.

The generation unit 132B uses the index information acquired by the acquisition unit 131 to generate an edge from a first blob to which one of the plurality of objects belongs to a second blob to which a neighboring object of the first object belongs. Generate a connected blob connectivity graph. The generation unit 132B selects one object from the plurality of objects, searches for the previous neighboring objects of the one object among the plurality of objects using the index information, and selects the neighboring objects from the first blob to which the one object belongs. A blob connectivity graph is generated by connecting edges to the second blob to which the object belongs.

The generation unit 132B randomly selects one object from a plurality of objects, searches for neighboring objects of the one object, and connects edges from the first blob to which the one object belongs to the second blob to which the neighboring object belongs. By doing this, a blob connected graph is generated. The generation unit 132B uses the graph to generate a blob connection graph in which edges are connected from the first blob to the second blob.

The generation unit 132B uses the graph to search for neighboring objects of one object, and generates a blob connection graph by connecting edges from the first blob to which the one object belongs to the second blob to which the neighboring object belongs. do. The generation unit 132B connects an edge from the first blob to a second blob to which a neighboring object corresponding to a neighboring node, which is a node connected by an edge to one node corresponding to one object in the graph, belongs. Generate a blob connected graph.

The generation unit 132B generates a blob connection graph by connecting, from the first blob, a first representative point that is a representative point of the first blob and a second representative point that is a representative point of a second blob. generate. The generation unit 132B generates a blob connection graph by connecting the first representative point, which is the centroid of the first blob, and the second representative point, which is the centroid of the second blob, with an edge. The generation unit 132B generates an image between a first representative point, which is a center point calculated based on an object belonging to the first blob, and a second representative point, which is a center point calculated based on an object belonging to a second blob. A blob-connected graph is generated by connecting the blobs with edges.

(Search processing unit 133B)
The search processing unit 133B performs various processes related to search processing in the same way as the search processing unit 133 or the search processing unit 133A.

The search processing unit 133B performs a search process using the blob connection graph generated by the generation unit 132B. The search processing unit 133B performs a search process to search for nearby objects of the search query using the blob connection graph. The search processing unit 133B performs a search process to search for nearby objects of the search query by tracing edges connecting blobs in the blob connection graph.

In the search process, the search processing unit 133B uses blob information indicating a plurality of blobs to calculate the distance between the target object, which is an object belonging to one blob, and the search query, and the vector information of the target object that has been vector quantized. Calculate using The search processing unit 133B parallelizes and collectively calculates the distance between the target object and the search query. The search processing unit 133B performs parallel processing to calculate the distance between the search query and the target object of the batch processing number determined based on the specifications of the information processing device.

[3-3. Information processing flow]
Next, the information processing procedure according to the third embodiment will be described using FIG. 26. FIG. 26 is a flowchart illustrating an example of information processing according to the third embodiment.

As shown in FIG. 26, the information processing apparatus 100B acquires group information indicating a plurality of groups into which a plurality of objects to be data searched are classified (step S501). For example, the information processing device 100B acquires blob information indicating a plurality of blobs in which nodes (object nodes) corresponding to a plurality of objects to be data searched are classified.

The information processing apparatus 100B acquires index information indicating an index that searches for a plurality of objects (step S502). For example, the information processing device 100B obtains, as an index, a graph in which nodes (object nodes) corresponding to a plurality of objects are connected by edges.

The information processing device 100B uses index information to create a group connection graph in which edges are connected from a first group to which one of the plurality of objects belongs to a second group to which a neighboring object of the one object belongs. is generated (step S503). For example, the information processing device 100B uses index information to create a blob whose edges are connected from a first blob to which one of the plurality of objects belongs to a second blob which is a blob to which a neighboring object of the one object belongs. Generate a connected graph.

[3-4. Search processing example]
Here, an example of the search process according to the third embodiment will be described using FIGS. 27 and 28 as examples. 27 and 28 are flowcharts illustrating an example of search processing according to the third embodiment. The search processing described below is performed by the search processing unit 133B of the information processing device 100B. Note that descriptions of points similar to the search processing described in the first embodiment and the second embodiment will be omitted as appropriate.

In FIGS. 27 and 28, N(G, s) and C are a set of blobs (blob set). Further, “G” may be graph data (blob connected graph) in which blobs are connected by edges (for example, graph GR41 shown in spatial information SP41). For example, the information processing device 100B executes k-nearest neighbor search processing.

First, the process (main process) shown in FIG. 27 will be explained. For example, the information processing device 100B sets the radius r of the hypersphere to ∞ (infinity) and sets the blob set C to the empty set (Φ) (step S601). Then, the information processing device 100B extracts a partial blob set B from the existing blob set (all blobs) (step S602). For example, the information processing device 100B may extract, as the partial blob set B, the blob to which the object (node) selected as the root node belongs. Further, for example, the information processing device 100B may randomly extract blobs as the partial blob set B.

The information processing device 100B then executes determination processing (step S603). The information processing device 100B executes the determination processing of steps S701 to S717 shown in FIG. 28. Details of the determination process shown in FIG. 28 will be described later.

Then, the information processing device 100B sets partial blob set B in blob set S (step S604).

The information processing device 100B selects the blob s with the minimum query distance d among the blobs included in the blob set S (step S605). Note that the "query distance" here is the shortest distance between all objects in the blob and the query object (query). For example, when the object serving as a query (search query) is y, the information processing device 100B selects the blob s having the shortest query distance to the object y from the blob set S. Note that the query distance is not limited to the above, and may be, for example, the distance between the query and a representative point (centroid, etc.) of the blob. The information processing device 100B then excludes the blob s from the blob set S (step S606).

Then, the information processing device 100B determines whether the query distance d of the blob s exceeds r(1+ε) (step S607). If the query distance d of the blob s exceeds r(1+ε) (step S607: Yes), the information processing device 100B outputs the object set R as a neighboring object set of the object y (step S608), and ends the process.

If the query distance d of the blob s does not exceed r(1+ε) (step S607: No), the information processing device 100B determines whether the blob s is included in the blob set C (step S609).

If the blob s is included in the blob set C (step S609: Yes), the information processing device 100B returns to step S605 and repeats the process.

If the blob s is not included in the blob set C (step S609: No), the information processing device 100B adds the blob s to the blob set C (step S610).

Then, the information processing device 100B sets a neighboring blob set N(G, s) of the blob s in the partial blob set B (step S611). The neighboring blob set N(G, s) is, for example, a set of blobs (nearby blobs) associated with blob s. For example, the neighboring blob set N(G, s) is a set of blobs (nearby blobs) to which edges from blob s are connected. The information processing device 100B then executes determination processing (step S612). Although details will be described later, for example, the information processing device 100B executes the determination processing of steps S701 to S717 shown in FIG. 28, similar to step S603 described above.

Then, the information processing device 100B determines whether the blob set S is an empty set (Φ) (step S613). If the blob set S is not an empty set (step S613: No), the information processing device 100B returns to step S605 and repeats the process. Furthermore, if the blob set S is an empty set (step S613: Yes), the information processing device 100B outputs the object set R and ends the process (step S614). For example, the information processing device 100B may select an object (node) included in the object set R as a neighboring node corresponding to the additional node (input object y). For example, the information processing device 100B may extract (select) objects (nodes) included in the object set R as neighboring nodes corresponding to the target node (input object y). Further, for example, the information processing device 100B may provide the objects (nodes) included in the object set R as the search results corresponding to the search query (input object y) to the terminal device or the like that performed the search.

From here, the process (determination process) shown in FIG. 28 will be explained. First, the information processing device 100B sets a partial blob set B in the blob set T (step S701).

Then, the information processing device 100B obtains one blob b from the blob set T, and deletes blob b from the blob set T (step S702). Then, the information processing device 100B sets min, which is a variable used in the determination process, to ∞ (infinity) (step S703).

Then, the information processing device 100B acquires one object u from blob b, and deletes object u from blob b (step S704).

Then, the information processing device 100B determines whether the distance d(u, y) between the object u and the object y is less than min (step S705).

If the distance d(u,y) between the object u and the object y is less than min (step S705: Yes), the information processing device 100B sets (updates) the value of min to the distance d(u,y). (Step S706), and the process of Step S707 is performed.

If the distance d(u, y) between object u and object y is not less than min (step S705: No), the information processing device 100B does not perform the process of step S706, but performs the process of step S707.

The information processing device 100B determines whether the distance d(u, y) between the object u and the object y is less than or equal to r(1+ε) (step S707).

If the distance d(u,y) between the object u and the object y is less than or equal to r(1+ε) (step S707: Yes), the information processing device 100B adds the blob b to the blob set S (step S708), The process of step S709 is performed.

If the distance d(u, y) between object u and object y is not less than or equal to r(1+ε) (step S707: No), the information processing device 100B performs the process of step S709 without performing the process of step S708. .

The information processing device 100B determines whether the distance d(u, y) between the object u and the object y is less than or equal to r (step S709).

If the distance d(u,y) between the object u and the object y is less than or equal to r (step S709: Yes), the information processing device 100B adds the object u to the object set R (step S710) and performs the steps in step S711. Perform processing.

If the distance d(u, y) between the object u and the object y is not less than or equal to r (step S709: No), the information processing device 100B performs the process of step S711 without performing the process of step S710.

The information processing device 100B determines whether the number of objects included in the object set R exceeds ks (step S711). The predetermined number ks is an arbitrarily determined natural number. For example, ks may be the number of searches or the number of extraction targets. Further, for example, when there is no upper limit on the number of objects to be extracted in a range search or the like, ks may be set to infinity. For example, ks may be 4, etc.

If the number of objects included in the object set R exceeds ks (step S711: Yes), the information processing device 100B excludes from the object set R the object that is farthest from the object y among the objects included in the object set R. (Step S712), and performs the process of Step S713.

If the number of objects included in the object set R does not exceed ks (step S711: No), the information processing device 100B performs the process of step S713 without performing the process of step S712.

The information processing device 100B determines whether the number of objects included in the object set R matches ks (step S713).

If the number of objects included in the object set R matches ks (step S713: Yes), the information processing device 100B determines the distance between the object y and the object that is farthest from the object y among the objects included in the object set R. is set to r (step S714), and the process of step S715 is performed.

If the number of objects included in the object set R does not match ks (step S713: No), the information processing device 100B performs the process of step S715 without performing the process of step S714.

The information processing device 100B determines whether blob b is an empty set (Φ) (step S715). If blob b is not an empty set (step S715: No), the information processing device 100B returns to step S704 and repeats the process.

Further, when blob b is an empty set (step S715: Yes), the information processing device 100B sets min as the blob query distance of blob b in partial blob set B (step S716). For example, the information processing device 100B sets the shortest distance among the distances between all objects of blob b and the query to min as the query distance.

The information processing device 100B determines whether the blob set T is an empty set (Φ) (step S717). If the blob set T is not an empty set (step S717: No), the information processing device 100B returns to step S702 and repeats the process.

Further, if the blob set T is an empty set (step S717: Yes), the determination process is ended.

[3-5. Modified example]
The blob connection graph is not limited to the example described above, and may be configured in various ways. For example, the blob connected graph may be a graph in which representative points of each group (blob) are connected by edges. This point will be explained below as a modification of the third embodiment. Note that descriptions of points similar to the first embodiment, second embodiment, and third embodiment will be omitted as appropriate. The information processing device 100B according to the modification of the third embodiment includes a blob information storage section 125A and a blob connection graph information storage section 126A instead of the blob information storage section 125 and the blob connection graph information storage section 126.

[3-5-1. Information processing〕
First, an overview of information processing according to a modified example will be explained using FIG. 29. FIG. 29 is a diagram illustrating an example of blob connection graph information according to a modification. The example in FIG. 29 shows a case where blobs are generated by clustering.

In the example of FIG. 29, the information processing device 100B generates a blob connection graph that connects the centroids of each blob. The information processing device 100B generates a graph GR42, which is a blob connection graph in which centroids C1 to C10 corresponding to each of the blobs BL1 to BL10 are connected by edges, as shown in the spatial information SP41. The information processing device 100B generates the graph GR42 using various conventional techniques related to graph generation.

For example, in the information processing device 100B, the centroid C1 of the blob BL1 is the centroid C2 of the blob BL2, the centroid C3 of the blob BL3, the centroid C4 of the blob BL4, the centroid C7 of the blob BL7, and the centroid C8 of the blob BL8. A graph GR42 is generated which is connected to the five centroids of by edges. For example, the information processing device 100B connects the centroid C2 of the blob BL2 to three centroids, the centroid C1 of the blob BL1, the centroid C3 of the blob BL3, and the centroid C4 of the blob BL4, by edges. Generate graph GR42.

Note that the information processing device 100B may generate the graph GR42 using any method. The information processing device 100B may generate the graph GR42 using various indexes. For example, the information processing device 100B may generate the graph GR42 using the graph GR21 in which object nodes are connected. In this case, the information processing apparatus 100B can generate the graph GR42 by the same process as the graph GR41 in FIG. 23, so detailed explanation will be omitted. For example, the information processing device 100B may generate the graph GR42 by a process similar to the example shown in FIG. 23. Although the graph GR42 shows an undirected (bidirectional) edge as an example, it may be a directed edge.

In this way, the information processing device 100B according to the modification generates a centroid graph as a blob connection graph. Then, the information processing device 100B performs a normal search using the centroid using the generated blob connection graph, and performs distance calculation on a blob-by-blob basis. In this way, the information processing apparatus 100B can simplify processing by representing blobs with centroids. Note that if the blobs are not clustered, the information processing device 100B may calculate the center point based on the objects belonging to each blob, and use the calculated center point as the representative point.

In the example of FIG. 29, the information processing device 100B performs a search process using a graph GR42 in which centroids, which are representative points of blobs, are connected by edges, as shown in spatial information SP42. For example, the information processing device 100B obtains the search results for the search query QE2 by performing search processing as shown in FIGS. 27 and 28 using the graph GR21 for the search query QE2. This point is similar to the example described above, and detailed explanation will be omitted.

[3-5-2. information〕
Next, an outline of the information related to the modified example will be explained using FIGS. 30 and 31. FIG. 30 is a diagram illustrating an example of a blob information storage unit according to a modification. FIG. 31 is a diagram illustrating an example of a blob connection graph information storage unit according to a modification.

First, a blob information storage unit 125A according to a modification shown in FIG. 30 will be described. The blob information storage unit 125A according to the modification shown in FIG. 30 includes items such as "blob ID", "node ID", "vector information", and "centroid ID". In this way, the blob information storage unit 125A according to the modified example is different from the blob information storage unit 125 of FIG. 15 in that it has a "centroid ID." Note that descriptions of the same points as those of the blob information storage unit 125 in FIG. 15 will be omitted as appropriate.

"Centroid ID" indicates identification information for identifying the centroid of each blob. "Vector information" indicates vector information of a centroid, which is a representative point of a blob.

The example shown in FIG. 30 indicates that the centroid of blob BL1 identified by blob ID "BL1" is the centroid (centroid C1) identified by centroid ID "C1". The centroid of blob BL9 identified by blob ID "BL9" indicates that it is the centroid (centroid C9) identified by centroid ID "C9".

Note that the blob information storage unit 125A is not limited to the above, and may store various information depending on the purpose.

Next, a blob connection graph information storage unit 126A according to a modification shown in FIG. 31 will be described. The blob connection graph information storage unit 126A according to the modification shown in FIG. 30 has items such as "centroid ID" and "connection centroid information." Note that descriptions of points similar to those of the blob connected graph information storage unit 126 in FIG. 25 will be omitted as appropriate.

"Centroid ID" indicates identification information for identifying a centroid that is a representative point of each blob in the graph. “Connection centroid information” indicates information regarding a centroid (reference destination centroid) that can be traced from a corresponding centroid. For example, "connection centroid information" includes information such as "reference destination." “Reference destination” indicates information for identifying a reference destination (centroid) that is connected by an edge and can be traced from the centroid. That is, in the example of FIG. 31, a reference destination (centroid) that can be traced from the centroid by an edge is registered in association with a centroid ID that identifies the centroid. Note that the "connection centroid information" may include information (edge ID) for identifying an edge connected to a reference destination.

In the example of FIG. 31, from the centroid (centroid C1) identified by the centroid ID "C1", each of the centroid IDs "C2", "C3", "C4", "C7", and "C8" The edge is connected to the five centroids identified by . In other words, it is shown that each of the five centroids C2, C3, C4, C7, and C8 can be traced from the centroid C1.

Note that the blob connection graph information storage unit 126A is not limited to the above, and may store various information depending on the purpose. The blob connection graph may also include a program module that receives a query as input, searches for objects (object nodes) by tracing edges in the blob connection graph, and extracts and outputs objects similar to the query. That is, the blob connection graph may be assumed to be used as a program module that performs search processing using the blob connection graph. For example, the graph GR42, which is a blob connection graph, may be a program that, when vector data is input as a query, extracts and outputs objects corresponding to vector data similar to the vector data using a blob connection graph. . For example, the graph GR42 may be data used as a program module that searches for similar images corresponding to a query image. For example, based on an input query, the graph GR 42 causes the computer to extract and output objects similar to the query in the blob connection graph.

As described above, in each embodiment and modification, the information processing system 1 executes the following processing. For example, the information processing system 1 generates blobs by clustering vectors. For example, the information processing system 1 generates a graph index using the centroid of a blob as a node.

For example, the information processing system 1 acquires a certain number of arbitrary vectors and performs clustering using Cartesian product quantization. That is, the information processing system 1 divides the vector into partial vectors, performs clustering for each partial vector, and generates a codebook.

For example, the information processing system 1 performs product quantization on objects belonging to blob units and generates transposed indexes. For example, the information processing system 1 generates a graph (blob graph) including edges from objects to blobs. For example, the information processing system 1 searches for nearby objects using a general product quantization method. The general method of direct product quantization mentioned here is, for example, after calculating the distances between objects and all centroids of blobs and obtaining a certain number of top blobs, the distances of objects within the blobs are subjected to direct product quantization. It may also be possible to obtain nearby objects by calculating the distance. Note that, as described above, when the information processing system 1 generates a centroid graph, it is possible to search for and obtain a certain number of blobs using the graph. Alternatively, the information processing system 1 may generate a graph in which blobs are connected by edges (blob connection graph) instead of the above-described graph including edges from objects to blobs (blob graph).

[4. effect〕
As described above, the information processing device 100A according to the second embodiment or the modification includes the acquisition section 131 and the generation section 132A. The acquisition unit 131 acquires first information indicating a plurality of objects to be searched for data, and second information used to classify the plurality of objects. The generation unit 132A uses the second information acquired by the acquisition unit 131 to classify the plurality of objects indicated by the first information, and classifies the plurality of objects used for batch processing in the search process targeting the plurality of objects. Third information indicating the group is generated.

In this way, the information processing apparatus 100A according to the second embodiment or the modified example classifies the plurality of objects using the second information used to classify the plurality of objects, and targets the plurality of objects. By generating third information indicating a plurality of groups used for performing batch processing in the search process, it becomes possible to perform batch processing in the search process using the third information. Therefore, the information processing device 100A can generate information that can be used for search processing.

The acquisition unit 131 acquires second information that is an index that searches for a plurality of objects. The generation unit 132A uses the second information to generate third information that classifies the plurality of objects into a plurality of groups.

In this way, the information processing apparatus 100A generates the third information using an index that searches for a plurality of objects, thereby making it possible to perform batch processing in the search process using the third information. Therefore, the information processing device 100A can generate information that can be used for search processing.

Furthermore, the generation unit 132A generates third information for classifying a plurality of objects into a plurality of groups by searching using the second information.

In this way, the information processing apparatus 100A generates the third information through a search using an index, thereby making it possible to perform batch processing in the search process using the third information. Therefore, the information processing device 100A can generate information that can be used for search processing.

The acquisition unit 131 acquires second information that is a graph in which a plurality of nodes corresponding to each of a plurality of objects are connected by edges. The generation unit 132A uses the second information to generate third information that classifies the plurality of objects into a plurality of groups.

In this way, the information processing device 100A generates third information using a graph in which a plurality of nodes corresponding to each of a plurality of objects are connected by edges, thereby performing batch processing in a search process using the third information. It becomes possible to do this. Therefore, the information processing device 100A can generate information that can be used for search processing.

The generation unit 132A also selects one node from a plurality of nodes in the graph, and generates classification target nodes that are at least some of the neighboring nodes that are nodes connected to the one node by edges, and The third information is generated by a classification process of classifying objects corresponding to each of the nodes into one group.

In this way, the information processing device 100A generates the third information through the classification process using the graph, thereby making it possible to perform batch processing in the search process using the third information. Therefore, the information processing device 100A can generate information that can be used for search processing.

In addition, the generation unit 132A performs a classification process that classifies into one group a predetermined number of nodes to be classified among nodes in the vicinity of one node, and a group of objects corresponding to each of one node. Generate information.

In this way, the information processing device 100A performs a classification process of classifying into one group a group of objects corresponding to each of the classification target nodes, which are a predetermined number of nodes among the nodes in the vicinity of one node, and one node. By generating the third information, it becomes possible to perform batch processing in the search process using the third information. Therefore, the information processing device 100A can generate information that can be used for search processing.

Furthermore, the generation unit 132A generates third information through a classification process of classifying the classification target node, which is all of the neighboring nodes of one node, and the object group corresponding to each of the one node into one group.

In this way, the information processing device 100A generates the third information by the classification process of classifying all neighboring nodes of one node and the object group corresponding to each of the nodes into one group. 3 information makes it possible to perform batch processing in search processing. Therefore, the information processing device 100A can generate information that can be used for search processing.

Further, the generation unit 132A selects one node from the plurality of nodes based on the number of neighboring nodes that are nodes connected to each of the plurality of nodes by edges, and selects the classification target node of the one node and the one node. Third information is generated by a classification process that classifies objects corresponding to each node into one group.

In this way, the information processing device 100A generates the third information by repeating the classification process for one node selected based on the number of neighboring nodes, thereby performing batch processing in the search process using the third information. It becomes possible to do so. Therefore, the information processing device 100A can generate information that can be used for search processing.

Furthermore, the generation unit 132A performs a classification process of selecting one node in descending order of the number of neighboring nodes, and classifying the classification target node of the one node and the object group corresponding to each of the one node into one group. Accordingly, the third information is generated.

In this way, the information processing device 100A generates the third information by repeating the classification process for one node selected in descending order of the number of neighboring nodes, so that the information processing device 100A can collectively perform the search process using the third information. It becomes possible to perform processing. Therefore, the information processing device 100A can generate information that can be used for search processing.

Furthermore, the generation unit 132A performs a classification process of selecting one node in order from the smallest number of neighboring nodes, and classifying the classification target node of the one node and the object group corresponding to each of the one node into one group. Accordingly, the third information is generated.

In this way, the information processing device 100A generates the third information by repeating the classification process for one node selected in descending order of the number of neighboring nodes, so that the information processing device 100A can collectively perform the search process using the third information. It becomes possible to perform processing. Therefore, the information processing device 100A can generate information that can be used for search processing.

In addition, the generation unit 132A randomly selects one node from the plurality of nodes, and performs a classification process of classifying the classification target node of the one node and the object group corresponding to each of the one node into one group. 3 Generate information.

In this way, the information processing device 100A generates the third information by repeating the classification process for one randomly selected node, thereby making it possible to perform batch processing in the search process using the third information. Become. Therefore, the information processing device 100A can generate information that can be used for search processing.

In addition, the generation unit 132A excludes the classification target node of the one node classified into the one group and the processed node group that is the one node, and uses the graph after excluding the processed node group. , the third information is generated by repeating the classification process.

In this way, the information processing device 100A can perform batch processing in the search process using the third information by excluding classified nodes and repeating the classification process to generate the third information. Therefore, the information processing device 100A can generate information that can be used for search processing.

Furthermore, the generation unit 132A generates third information by repeating the classification process using the graph after excluding the processed node group until a plurality of objects are classified into one of the groups.

In this way, the information processing device 100A generates the third information by repeating the classification process until all objects are classified into one of the groups, thereby making it possible to perform batch processing in the search process using the third information. becomes. Therefore, the information processing device 100A can generate information that can be used for search processing.

Furthermore, the generation unit 132A generates third information indicating a plurality of groups to which objects belonging to each group are mutually exclusive.

In this way, the information processing device 100A generates the third information indicating a plurality of mutually exclusive groups, each of which belongs to a plurality of groups, thereby making it possible to perform batch processing in the search process using the third information. . Therefore, the information processing device 100A can generate information that can be used for search processing.

The acquisition unit 131 acquires fourth information that is an index that searches for multiple objects. Using the fourth information, the generation unit 132A generates an edge from an object node, which is a node corresponding to at least some of the plurality of objects, to a group other than the group to which the object node belongs among the plurality of groups. Generates a connected group graph.

In this way, the information processing apparatus 100A can perform search processing using the group graph by generating a group graph using an index that searches for a plurality of objects. Therefore, the information processing device 100A can generate information that can be used for search processing.

The generation unit 132A selects one object from the plurality of objects, searches for neighboring objects of the one object among the plurality of objects using the fourth information, and selects one object from the one object node corresponding to the one object. , a group graph is generated by a connection process that connects edges to other groups to which neighboring objects belong.

In this way, the information processing device 100A generates a group graph through a connection process that connects edges from one object node to other groups to which neighboring objects belong, and performs a search process using the group graph. becomes possible. Therefore, the information processing device 100A can generate information that can be used for search processing.

The generation unit 132A randomly selects one object from a plurality of objects, searches for neighboring objects of the one object, and creates an edge from one object node corresponding to the one object to another group to which the neighboring object belongs. A group graph is generated by the concatenation process.

In this way, the information processing device 100A generates a group graph through a connection process that connects edges from one randomly selected object node to another group to which neighboring objects belong, and performs a search using the group graph. It becomes possible to perform processing. Therefore, the information processing device 100A can generate information that can be used for search processing.

The acquisition unit 131 acquires fourth information that is a graph in which a plurality of object nodes corresponding to each of a plurality of objects are connected by edges. The generation unit 132A generates a group graph using the fourth information.

In this way, the information processing apparatus 100A can perform search processing using the group graph by generating a group graph using a graph in which a plurality of object nodes are connected by edges. Therefore, the information processing device 100A can generate information that can be used for search processing.

[5. Hardware configuration]
The information processing apparatuses 100, 100A, and 100B according to each of the embodiments described above are realized by, for example, a computer 1000 having a configuration as shown in FIG. 32. FIG. 32 is a hardware configuration diagram showing an example of a computer that implements the functions of the information processing device. The computer 1000 includes a CPU 1100, a RAM 1200, a ROM (Read Only Memory) 1300, an HDD (Hard Disk Drive) 1400, a communication interface (I/F) 1500, an input/output interface (I/F) 1600, and a media interface (I/F). ) has 1700.

CPU 1100 operates based on a program stored in ROM 1300 or HDD 1400, and controls each section. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 is started, programs depending on the hardware of the computer 1000, and the like.

The HDD 1400 stores programs executed by the CPU 1100, data used by the programs, and the like. Communication interface 1500 receives data from other devices via network N and sends it to CPU 1100, and sends data generated by CPU 1100 to other devices via network N.

The CPU 1100 controls output devices such as a display and a printer, and input devices such as a keyboard and mouse via an input/output interface 1600. CPU 1100 obtains data from an input device via input/output interface 1600. Further, CPU 1100 outputs the generated data to an output device via input/output interface 1600.

Media interface 1700 reads programs or data stored in recording medium 1800 and provides them to CPU 1100 via RAM 1200. CPU 1100 loads this program from recording medium 1800 onto RAM 1200 via media interface 1700, and executes the loaded program. The recording medium 1800 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or a PD (Phase change rewritable disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory. etc.

For example, when the computer 1000 functions as the information processing device 100 according to the first embodiment, the CPU 1100 of the computer 1000 realizes the functions of the control unit 130 by executing a program loaded onto the RAM 1200. The CPU 1100 of the computer 1000 reads these programs from the recording medium 1800 and executes them, but as another example, these programs may be acquired from another device via the network N.

As mentioned above, some embodiments of the present application have been described in detail based on the drawings, but these are merely examples, and various modifications and variations can be made based on the knowledge of those skilled in the art, including the embodiments described in the disclosure section of the invention. It is possible to carry out the invention in other forms with modifications.

[6. others〕
Furthermore, among the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or the processes described as being performed manually can be performed manually. All or part of the process can also be performed automatically using known methods. In addition, information including the processing procedures, specific names, and various data and parameters shown in the above documents and drawings may be changed arbitrarily, unless otherwise specified. For example, the various information shown in each figure is not limited to the illustrated information.

Furthermore, each component of each device shown in the drawings is functionally conceptual, and does not necessarily need to be physically configured as shown in the drawings. In other words, the specific form of distributing and integrating each device is not limited to what is shown in the diagram, and all or part of the devices can be functionally or physically distributed or integrated in arbitrary units depending on various loads and usage conditions. Can be integrated and configured.

Furthermore, the processes described in the embodiments described above can be combined as appropriate within a range that does not conflict with the process contents.

Further, the above-mentioned "section, module, unit" can be read as "means", "circuit", etc. For example, the acquisition unit can be read as an acquisition means or an acquisition circuit.

1 Information processing system 10 Terminal device 50 Information providing device 100, 100A, 100B Information processing device 120, 120A, 120B Storage unit 121 Object information storage unit 122 Graph information storage unit 123, 123A Quantization information storage unit 124 Codebook information storage unit 125, 125A Blob information storage unit 126, 126A Blob connected graph information storage unit 130, 130A, 130B Control unit 131 Acquisition unit 132, 132A, 132B Generation unit 133, 133A, 133B Search processing unit 134 Providing unit

Claims (17)

an acquisition unit that acquires first information indicating a plurality of objects to be data searched and second information used to classify the plurality of objects;
The second information acquired by the acquisition unit is used to classify the plurality of objects indicated by the first information, and the plurality of objects used for batch processing in the search process targeting the plurality of objects are classified. a generation unit that generates third information indicating a group;
Equipped with
The acquisition unit includes:
obtaining the second information that is a graph in which a plurality of nodes corresponding to each of the plurality of objects are connected by edges;
The generation unit is
One node is selected from the plurality of nodes in the graph, and the classification target nodes are at least some nodes among the neighboring nodes that are nodes connected to the one node by an edge, and the one node is Excluding the classification target node of the one node classified into the one group and the processed node group of the one node from the graph. The information processing apparatus is characterized in that the third information is generated by repeating the process using the graph after excluding the processed node group. The acquisition unit includes:
obtaining the second information that is an index for searching the plurality of objects;
The generation unit is
The information processing apparatus according to claim 1, wherein the third information for classifying the plurality of objects into the plurality of groups is generated using the second information. The generation unit is
The information processing apparatus according to claim 2, wherein the third information for classifying the plurality of objects into the plurality of groups is generated by a search using the second information. The generation unit is
The classification target node, which is a predetermined number of nodes among the neighboring nodes of the one node, and the object group corresponding to each of the one node are classified into the one group. The information processing device according to claim 1, wherein the information processing device generates information. The generation unit is
generating the third information by the classification process of classifying the classification target node, which is all of the neighboring nodes of the one node, and object groups corresponding to each of the one node into the one group; The information processing device according to any one of claims 1 to 4, characterized by: The generation unit is
The one node is selected from the plurality of nodes based on the number of neighboring nodes that are nodes connected to each of the plurality of nodes by edges, and the one node is classified as the classification target node of the one node and the one node. The information processing apparatus according to any one of claims 1 to 5 , wherein the third information is generated by the classification process of classifying a group of objects corresponding to each of the objects into the one group. The generation unit is
the classification process of selecting the one node in descending order of the number of neighboring nodes and classifying the object group corresponding to each of the classification target node of the one node and the one node into the one group; The information processing device according to claim 6 , wherein the third information is generated by: The generation unit is
the classification process of selecting the one node in descending order of the number of neighboring nodes, and classifying the object group corresponding to each of the classification target node of the one node and the one node into the one group; The information processing device according to claim 6 , wherein the third information is generated by: The generation unit is
The classification process randomly selects the one node from the plurality of nodes and classifies the classification target node of the one node and the object group corresponding to each of the one node into the one group. The information processing device according to any one of claims 1 to 5 , wherein the information processing device generates third information. The generation unit is
The third information is generated by repeating the classification process using the graph after excluding the processed node group until the plurality of objects are classified into one of the groups. The information processing device according to any one of claims 1 to 9 . The generation unit is
The information processing apparatus according to any one of claims 1 to 10 , wherein the third information is generated indicating the plurality of groups to which objects belonging to each group are mutually exclusive. The acquisition unit includes:
obtaining fourth information that is an index for searching the plurality of objects;
The generation unit is
Using the fourth information, an edge is sent from an object node that corresponds to at least some of the plurality of objects to a group other than the group to which the object node belongs among the plurality of groups. The information processing device according to any one of claims 1 to 11 , wherein the information processing device generates a connected group graph. The generation unit is
Select one object from the plurality of objects, use the fourth information to search for nearby objects of the one object among the plurality of objects, and select one object node corresponding to the one object. 13. The information processing apparatus according to claim 12 , wherein the group graph is generated by a connection process that connects an edge to the other group to which the neighboring object belongs. The generation unit is
The one object is randomly selected from the plurality of objects, the neighboring objects of the one object are searched, and the other group to which the neighboring object belongs is determined from the one object node corresponding to the one object. The information processing apparatus according to claim 13 , wherein the group graph is generated by the connection process of connecting edges to each other. The acquisition unit includes:
obtaining the fourth information, which is a graph in which a plurality of object nodes corresponding to each of the plurality of objects are connected by edges;
The generation unit is
The information processing apparatus according to any one of claims 12 to 14 , wherein the group graph is generated using the fourth information. An information processing method performed by a computer, the method comprising:
an acquisition step of acquiring first information indicating a plurality of objects to be data searched and second information used to classify the plurality of objects;
The second information acquired in the acquisition step is used to classify the plurality of objects indicated by the first information, and the plurality of objects used for batch processing in the search process targeting the plurality of objects are classified. a generation step of generating third information indicating the group;
including;
The acquisition step includes:
obtaining the second information that is a graph in which a plurality of nodes corresponding to each of the plurality of objects are connected by edges;
The production step includes:
One node is selected from the plurality of nodes in the graph, and the classification target nodes are at least some nodes among the neighboring nodes that are nodes connected to the one node by an edge, and the one node is Excluding the classification target node of the one node classified into the one group and the processed node group of the one node from the graph. and generate the third information by repeating the process using the graph after excluding the processed node group.
An information processing method characterized by: an acquisition procedure for acquiring first information indicating a plurality of objects that are the targets of a data search, and second information used to classify the plurality of objects;
The second information acquired by the acquisition procedure is used to classify the plurality of objects indicated by the first information, and the plurality of objects used for batch processing in the search process targeting the plurality of objects are classified. a generation procedure for generating third information indicating a group;
make the computer run
The acquisition procedure is as follows:
obtaining the second information that is a graph in which a plurality of nodes corresponding to each of the plurality of objects are connected by edges;
The generation procedure is
One node is selected from the plurality of nodes in the graph, and the classification target nodes are at least some nodes among the neighboring nodes that are nodes connected to the one node by an edge, and the one node is Excluding the classification target node of the one node classified into the one group and the processed node group of the one node from the graph. and generating the third information by repeating the process using the graph after excluding the processed node group.

Images