JP7077387B1 - Information processing equipment, information processing methods, and information processing programs - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable efficient search processing. An information processing device according to the present application has an acquisition unit and a search processing unit. The acquisition unit acquires a search query for a plurality of objects to be searched for data. The search processing unit uses a graph in which nodes corresponding to each of a plurality of objects are connected by edges to search for nodes in the vicinity of the search query, and the search processing unit includes a plurality of nodes selected by a predetermined criterion. The distance to the search query is calculated using the vector information of a plurality of nodes that have undergone vector quantization. [Selection diagram] FIG. 5

Description

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

Conventionally, a technique for searching (searching) various information has been provided. For example, there is provided a technique of generating a graph in which nodes corresponding to a search target are connected by edges and performing a search using the generated graph. Further, such a technique is used, for example, for image retrieval and the like.

Japanese Patent No. 6293335 Japanese Patent No. 630982 Japanese Patent No. 6311000

Masajiro Iwasaki "Approximate k-nearest neighbor search using tree-structured index", IPSJ Journal, 2011/2, Vol. 52, No. 2. pp.817-828.

However, in the above-mentioned conventional technique, there is room for improvement in terms of improving the efficiency of the search process. For example, in the above-mentioned conventional technique, the number of edges of the graph is adjusted to change the graph so that the graph can be searched efficiently, thereby improving the efficiency such as reducing the time of the search process. There is a limit to improving the efficiency of the search process by making changes. Therefore, it is desired to improve the efficiency of the search process itself.

The present application has been made in view of the above, and an object of the present application is to provide an information processing apparatus, an information processing method, and an information processing program that enable efficient search processing.

The information processing apparatus according to the present application uses an acquisition unit for acquiring search queries for a plurality of objects to be searched for data, and a graph in which nodes corresponding to each of the plurality of objects are connected by edges. In the search process for searching for a node in the vicinity of the search query, the distance between the plurality of nodes selected by a predetermined criterion and the search query is calculated using the vector information of the plurality of nodes subjected to vector quantization. It is characterized by having a search processing unit.

According to one aspect of the embodiment, there is an effect that efficient search processing can be enabled.

FIG. 1 is a diagram showing 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 showing a configuration example of the information processing system according to the first embodiment. FIG. 5 is a diagram showing a configuration example of the information processing apparatus according to the first embodiment. FIG. 6 is a diagram showing an example of an object information storage unit according to the first embodiment. FIG. 7 is a diagram showing an example of a graph information storage unit according to the first embodiment. FIG. 8 is a diagram showing an example of a quantized information storage unit according to the first embodiment. FIG. 9 is a diagram showing an example of a codebook information storage unit according to the first embodiment. FIG. 10 is a flowchart showing an example of information processing according to the first embodiment. FIG. 11 is a flowchart showing an example of the search process according to the first embodiment. FIG. 12 is a diagram showing an example of information processing according to the second embodiment. FIG. 13 is a diagram showing a configuration example of the information processing apparatus according to the second embodiment. FIG. 14 is a diagram showing an example of a quantized information storage unit according to the second embodiment. FIG. 15 is a diagram showing an example of a blob information storage unit according to the second embodiment. FIG. 16 is a flowchart showing an example of the search process according to the second embodiment. FIG. 17 is a diagram showing an example of information processing according to a modified example. FIG. 18 is a diagram showing an example of a graph information storage unit according to a modified example. FIG. 19 is a flowchart showing an example of the search process according to the modified example. FIG. 20 is a hardware configuration diagram showing an example of a computer that realizes the functions of the information processing device.

Hereinafter, the information processing apparatus, the information processing method, and the embodiment for implementing the information processing program (hereinafter referred to as “the embodiment”) according to the present application will be described in detail with reference to the drawings. It should be noted that this embodiment does not limit the information processing apparatus, information processing method, and information processing program according to the present application. Further, in each of the following embodiments, the same parts are designated by the same reference numerals, and duplicate explanations are omitted.

(Embodiment)
[1. First Embodiment]
[1-1. Information processing]
An example of information processing according to the first embodiment will be described with reference to FIG. 1. FIG. 1 is a diagram showing 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”) in which a plurality of objects to be searched for data are graph-structured. FIG. 1 shows a part of a search process in which the information processing apparatus 100 performs a neighborhood search using a graph in which the nodes corresponding to the vectorized objects of the object to be searched for data are connected by edges. .. The information processing apparatus 100 searches for a node in the vicinity of a given search query (vector) by tracing the graph, with each node of the graph as an object to be searched. 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 by the search process as nodes in the vicinity of the search query. Hereinafter, the case where the image information is the target of the data search will be described as an example, but the target of the data search may be various targets such as moving image information and audio information.

The information processing apparatus 100 performs a search process on a node corresponding to a huge amount of image information in the unit of millions to hundreds of millions, etc., but in the drawing, a part of it (in FIG. 1, dozens of nodes N1 etc.) ) Only shown. For example, as shown in the spatial information SP1 in FIG. 1, the information processing apparatus 100 acquires information about a plurality of nodes (vectors) as shown in the nodes N1, N7, N9, and the like. When described as "node N * (* is an arbitrary numerical value)" in this way, it means that the node is a node identified by the node ID "N *". For example, when described as "node N1", the node is a node identified by the node ID "N1". Each of the white circles (◯) in the spatial information SP1 indicates each node.

In the spatial information SP1 in FIG. 1, the nodes mainly related to the description are coded, but each of the unsigned white circles (○) is also a node, and includes a large number of nodes other than the node shown in the figure. Is done. In addition, 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 the distance between objects is defined may be an object.

For example, the spatial information SP1 in FIG. 1 may be an Euclidean space. For example, the spatial information SP1 corresponds to the number of dimensions of the vector of the object, and is assumed to be 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 the point of direct product quantization is Will be described later.

The dotted line connecting the white circles (◯), which are the nodes in the spatial information SP1, indicates the edge connecting the nodes. In the example of FIG. 1, for the sake of simplicity, a case where nodes are connected by undirected (bidirectional) edges (hereinafter, also simply referred to as “edges”) will be shown. The undirected edge here means an edge that can trace data in both directions between connected nodes. For example, the white circle (〇) indicating the node N1 and the white circle (〇) indicating the node N4 are connected by a dotted line, and the node N1 and the node N4 are connected by an edge. That is, in the graph shown in the spatial information SP1, it is possible to trace between the node N1 and the node N4 in both directions. Specifically, in the spatial information SP1, the node N1 can be traced to the node N4, and the node N4 can be traced to the node N1.

In the example of FIG. 1, for the sake of illustration, only the edges connecting the illustrated nodes are shown, but a large number of edges are included in addition to the illustrated edges. As described above, in the spatial information SP1 in FIG. 1, only a part of the edge is shown, but it is assumed that it is, for example, a k-nearest neighbor graph. The spatial information SP1 may be various graphs.

Further, the edge of the graph is not limited to the undirected edge, and may be a directed edge. 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 node that is the reference destination. For example, when two nodes are connected by two directed edges, a first edge with one as a reference source and the other as a reference destination, and a second edge with one as a reference destination and the other as a reference source. , The two nodes are connected by an undirected (bidirectional) edge and are in the same state as the state.

From here, the search process will be described with reference to FIG. In the example of FIG. 1, it is assumed that the information processing apparatus 100 has already acquired the graph GR1 as shown in the spatial information SP1. The information processing apparatus 100 may generate the graph GR1 by appropriately using various conventional techniques. First, prior to the explanation of the search process, the direct product quantization of the vector (space) will be described.

In FIG. 1, as shown in the spatial information SP1, the state in which the vector (space) is divided into four subspaces AR11 to AR14 by direct product quantization is conceptually illustrated. The subspaces AR11 to AR14 shown in FIG. 1 are shown in a conceptual diagram for explaining the distance between vectors of each node (object), and the subspaces AR11 to AR14 are multidimensional spaces. For example, the subspace AR11 shown in FIG. 1 is shown in a two-dimensional manner for being shown on a plane, but is assumed to be a multidimensional space such as 100 dimensions or 1000 dimensions.

Each of the subspaces AR11 to AR14 indicates a space corresponding to each division (partial vector) of the vector to be divided into four. For example, a 100-dimensional vector may be divided into 100. For example, the subspace AR11 is a dimensional space (“1st subvector”) corresponding to the first subvector (also referred to as “first subvector”) among four subvectors (also referred to as “subvectors”) in which the vector is divided into four. Also referred to as "first subspace"). The subspace AR11 indicates a subspace (first subspace) corresponding to the division position at the beginning of the vector. In the case of the search query QE1, the subspace AR11 is a space corresponding to the first subquery QE1-1 which is the first subvector.

Further, 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 subspaces in which the vector is divided into four. Also called). The subspace AR12 indicates a subspace (second subspace) corresponding to the second division position from the beginning of the vector. In the case of the search query QE1, the subspace AR12 is a space corresponding to the second subquery QE1-2, which is the 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 in which the vector is divided into four. Also called). The subspace AR13 indicates a subspace (third subspace) corresponding to the third division position from the beginning of the vector. In the case of the search query QE1, the subspace AR13 is a space corresponding to the third subquery QE1-3, which is the third subvector from the beginning.

Further, the subspace AR14 is a space having a dimension corresponding to the fourth (that is, the last) subvector (also referred to as “fourth subvector”) from the four subvectors in which the vector is divided into four (“4th subvector”). Also referred to as "fourth subspace"). The subspace AR14 indicates a subspace (fourth subspace) corresponding to the fourth division position from the beginning of the vector. In the case of the search query QE1, the subspace AR14 is a space corresponding to the fourth subquery QE1-4, which is the fourth partial vector (fourth subvector) from the beginning.

In the example of FIG. 1, the subspaces AR11 to AR14 are shown in similar shapes, but the shapes of the subspaces AR11 to AR14 may be different, and the division mode of the regions in the subspaces AR11 to AR14 is also different. You may. Further, it is assumed that the connection relationship between the nodes by the edge is common in the 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.

Further, in the example of FIG. 1, a look-up table is generated for each divided partial vector (sub-vector). For example, the first subvector, the second subvector, the third subvector, and the fourth subvector are clustered for each of the four subvectors, and a look-up table (codebook information) is individually generated.

In FIG. 1, as shown in the subspace AR11, the first subvector is clustered into nine groups corresponding to the codebooks CD11 to CD19, and the vector corresponding to each of the codebooks CD11 to CD19 is a representative vector ( Calculated as a centroid). For example, it is shown that the first subvectors of the nodes N7, N9, etc. are clustered into the group corresponding to the codebook CD11. In this case, the first subvector of the nodes N7, N9, etc. uses the codebook information (also referred to as “first codebook information”) corresponding to the first subvector in the calculation (calculation) of the distance, and the codebook CD11. Is vector-quantized into a vector of.

The second subvector is clustered into a plurality of groups corresponding to the codebooks CD21 to CD24 and the like (see FIGS. 2 and 9), and the vector corresponding to each of the codebooks CD21 to CD24 and the like is a representative vector (cent). Lloyd) is calculated. In this case, the second subvector of each node uses the codebook information (also referred to as "second codebook information") corresponding to the second subvector in the calculation (calculation) of the distance, and the vector of the corresponding codebook. Is vector-quantized to.

The third subvector is clustered into a plurality of groups corresponding to the codebooks CD31 to CD34, etc. (see FIGS. 2 and 9), and the vector corresponding to each of the codebooks CD31 to CD34, etc. is a representative vector (centroid). Is calculated as. In this case, the third subvector of each node uses the codebook information (also referred to as "third codebook information") corresponding to the third subvector in the calculation (calculation) of the distance, and the vector of the corresponding codebook. Is vector-quantized to.

The fourth subvector is clustered into a plurality of groups corresponding to the codebooks CD41 to CD44 (see FIGS. 2 and 9), and the vector corresponding to each of the codebooks CD24 to CD44 is a representative vector (centroid). Is calculated as. In this case, the fourth subvector of each node uses the codebook information (also referred to as "fourth codebook information") corresponding to the fourth subvector in the calculation (calculation) of the distance, and the vector of the corresponding codebook. Is vector-quantized to.

The codebook information is generated by appropriately using various techniques such as the process of obtaining the representative vector. Since the generation of codebook information is a conventional technique, detailed description thereof will be omitted. Further, the look-up table is not limited to the above, and for example, one look-up table (codebook information) may be used for the entire divided partial vector (sub-vector), which will be described later.

From here, the search process targeting the search query QE1 will be described. First, the information processing apparatus 100 acquires 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 apparatus 100 divides the vector which is the search query QE1 into four. That is, the information processing apparatus 100 divides the search query QE1 into four partial queries. In FIG. 1, the information processing apparatus 100 uses the search query QE1 as a first subvector, a first partial query QE1-1, a second subvector, a second partial query QE1-2, and a third subvector. It is divided into four subvectors, a three-part query QE1-3 and a fourth subvector, the fourth subquery QE1-4. Specifically, in the information processing apparatus 100, "45, 23, 2 ..." is referred to as the first partial query QE1-1, "127, 34, 5 ..." is referred to as the second partial query QE1-2, and "20, "98,110 ..." is referred to as the third subquery QE1-3, and "12,45,4 ..." is referred to as the fourth subquery QE1-4.

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

The codebook information TB1 indicates the first codebook information corresponding to the first subvector, and the information processing apparatus 100 has the respective vectors of the codebooks CD11 to CD19 corresponding to the first subvector, and the first partial query QE1. Calculate the distance to -1. In FIG. 2, the information processing apparatus 100 calculates the distance between the codebook CD11 and the first partial query QE1-1 as the distance DS11, as shown in the codebook information TB1. Similarly, the information processing apparatus 100 calculates the distance between each of the codebooks CD12 to CD14 and the like and the first partial query QE1-1 as the distance DS12 to DS14 and the like.

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

Although shown abstractly in FIG. 2 for the sake of explanation, the distances DS11 to DS44 and the like are assumed to be specific values. The distance is a value expressed in floating point, but scale-offset-compression ("https://www.unidata.ucar.edu/blogs/") for speeding up by SIMD (Single Instruction, Multiple Data) described later. It may be compressed to a 1-byte integer type by (see developer / entry / compression_by_scaling_and_offfset "etc.). 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 is needed. The information processing apparatus 100 uses the codebook information TB1 to TB4 as a look-up table in the search process to calculate the distance between each node and the search query QE1, that is, an approximate distance (also referred to as “first distance”). do. As described above, when the information processing apparatus 100 searches the graph in the search process, the information processing apparatus 100 does not have a true distance between each node and the search query (also referred to as a “second distance”), but an approximate distance (also referred to as a “second distance”). By calculating (1 distance) and performing processing, efficient search processing can be enabled.

The information processing apparatus 100 executes a search process targeting the search query QE1 (step S12). The information processing apparatus 100 obtains the search result of the search query QE1 by performing the 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 apparatus 100 performs a search process on the search query QE1 to extract the nodes having the number of searches as the nodes in the vicinity of the search query QE1.

In the search process targeting the search query QE1, the information processing apparatus 100 selects a predetermined node as a node (hereinafter, also referred to as “starting point node”) that is the starting point (starting point) of the search of the graph GR1. do. For example, the information processing apparatus 100 selects a starting node using a predetermined index such as a tree-structured index. In the example of FIG. 1, the information processing apparatus 100 selects the node N7 as the starting node. Note that FIG. 1 shows a case where only the node N7 is selected as the starting node using a predetermined index for the sake of simplicity, but the information processing apparatus 100 may select a plurality of nodes as the starting node. Alternatively, the starting node may be selected by various methods such as random.

Here, the information processing apparatus 100 parallelizes the distance between the node (hereinafter, also referred to as “connection node”) to which the edges from one node are connected and the search query in the search process targeting the search query QE1. Calculate (step S13). In FIG. 1, as shown in the batch processing information LT1, the information processing apparatus 100 refers to the search query QE1 for the nodes N9, N12, N54, N85, etc., which are the nodes (connection nodes) to which the edges from the node N7 are connected. Calculate by parallelizing the distance with.

For example, in the node N9, as shown in the node information INF1, the first subvector is the codebook CD12, the second subvector is the codebook CD23, the third subvector is the codebook CD35, and the fourth subvector is the codebook CD47. Indicates that it is associated with. Since the size of the variable CD is determined by the size of the codebook, for example, if the codebook size is 16, the variable CD may be 4 bits, and a large amount of data can be compressed. Therefore, the information processing apparatus 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, and the distance between the node N9 and the search query QE1. Is calculated. For example, the information processing apparatus 100 uses the total 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 the codebook CD14, the second subvector is the codebook CD29, the third subvector is the codebook CD31, and the fourth subvector is the codebook CD45. Indicates that it is associated with. For example, the information processing apparatus 100 uses the total 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 apparatus 100 calculates the distance between the node N54 and the search query QE1 using the node information INF3 of the node N54, and uses the node information INF4 of the node N85 to set the node N85 and the search query QE1. Calculate the distance.

For example, the information processing apparatus 100 collectively calculates the distance between each of the nodes N9, N12, N54, and N85 and the search query QE1 by parallelizing the SIMD operation. As a result, the information processing apparatus 100 can speed up the distance calculation by parallelizing the distance calculation, and can enable efficient search processing.

Note that FIG. 1 shows a case where the distances are calculated in parallel for four nodes for the sake of simplicity, but the number to be parallelized is determined based on the specifications of the information processing apparatus 100. .. For example, when the number of information processing apparatus 100 that can be collectively processed by SIMD (also referred to as “collective processable unit”) is “4”, the processing is the same as in FIG. 1, but can be collectively processed by SIMD. When the number (unit that can be collectively processed) is "16", the information processing apparatus 100 calculates by parallelizing the distances for 16 nodes. Further, when the batch processable unit is "32", the information processing apparatus 100 calculates by parallelizing the distances for 32 nodes. In this way, the information processing apparatus 100 enables efficient search processing by collectively calculating the distances of the number of nodes corresponding to the number that can be collectively processed by SIMD (collective processable unit). be able to.

The information processing apparatus 100 performs the search process as shown in FIG. 11 using the graph GR1 for the search query QE1 while performing the distance calculation of the plurality of nodes in parallel as described above. Node is extracted as a node in the vicinity of the search query QE1.

As described above, the information processing apparatus 100 calculates an approximate distance (first distance) between the vector of each vector quantized node and the search query, and performs the search process using the first distance. Therefore, efficient search processing can be enabled. Further, the information processing apparatus 100 can enable efficient search processing by collectively calculating the distances of a number of nodes that can be parallelized. As described above, the information processing apparatus 100 repeatedly uses the limited memory area (look-up table) by collectively calculating the approximate distances of the nodes of the graph to the plurality of connected nodes (in the memory cache). Therefore, it is possible to increase the speed.

Further, in the example of FIG. 1, the information processing apparatus 100 is more efficient than the case where the search process is performed without dividing the vector by performing the search process using the vector divided by the direct product quantization. Search processing can be enabled. For example, the information processing apparatus 100 calculates the distance between short vectors by direct product quantization, and the size of the vector to be calculated for the distance can be reduced. Further, the information processing apparatus 100 can limit the memory space accessed by the look-up table used at the time of distance calculation to the look-up table of the divided subvectors. The information processing apparatus 100 can reduce the vector size while suppressing a decrease in search accuracy by direct product quantization. Note that FIG. 1 describes the process when the direct product quantization is performed, but the case where the direct product quantization is performed is only an example, and the information processing apparatus 100 is a vector in which the direct product quantization is not performed. You may perform the search process using.

[1-1-1. others〕
The above-mentioned processing is only an example, and the information processing apparatus 100 may perform the search processing by using various information and methods for the efficient search processing. In this regard, each item will be described in detail.

(Storage mode of edge (connection node))
Regarding the information of the edge (connection node) connected to each node, for example, the following two types of storage mode (storage method) of the edge (connection node) at the node can be considered. Therefore, the storage mode may be selected depending on the usage mode.

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

Further, for example, as the second storage mode, each node has an object directly quantized. In the case of the second storage mode, the connection node (object) of a certain node stores the quantized information in association with each node. For example, a storage mode in which the quantized information (node information INF1 to INF4 in FIG. 2) of each of the nodes N9, N12, N54, and N85, which are the connection nodes of the node N7, is stored in association with the node N7. Corresponds to the second storage mode. In the case of the second storage mode, since the objects are accessed sequentially at the time of searching, it is possible to suppress the occurrence of slowdown.

(Look-up table)
In the above example, the case where clustering is performed individually for each divided subvector and the look-up table (codebook information) is generated individually is shown, but the case is not limited to this case, and the look-up table can be used in any manner. There may be.

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

In addition, a lookup table 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 a plurality of classes, and a look-up table may be generated for each of the plurality of classes. In the above example, for example, when the variances of the two vectors of the first subvector and the third subvector are similar, and the variances of the two vectors of the second subvector and the fourth subvector are similar, the first The subvector and the third subvector may be merged into the first class, and the second subvector and the fourth subvector may be merged into the second class. In this case, two look-up tables, a first-class look-up table (codebook information) and a second-class look-up table (codebook information), may be generated.

(Transposition)
The above-mentioned data retention is only an example, and the information processing apparatus 100 may retain the data in various manners so that efficient data reference and the like can be performed. For example, the information processing apparatus 100 may hold the transposed data. This point will be described with reference to FIG. FIG. 3 is a diagram showing an example of data according to the first embodiment.

For example, when the data is retained as shown in the node information INF1 to INF4 and the node information INF1 to INF4 are sequentially processed, the codebook information TB1 to TB4 which is a look-up table is repeated as shown in step S14 of FIG. It will be referred to.

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

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

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

It should be noted that, in the list of each transposed data TR1 to TR4, correspondence information indicating the correspondence of which node corresponds to which node is generated. In the example of FIG. 3, in the list of each transposed data TR1 to TR4, the first (first) data corresponds to the node N9, the second data corresponds to the node N12, and the third data corresponds to the node N54. Corresponding, correspondence information is generated indicating that the fourth (last) data corresponds to node N85. The information processing apparatus 100 can specify which node the data in the list of the transposed data TR1 to TR4 corresponds to by referring to the correspondence information. For example, the information processing apparatus 100 generates correspondence information.

The information processing apparatus 100 performs processing using the transposed data TR1 to TR4 in which the node information INF1 to INF4 are transposed. For example, when the information processing apparatus 100 refers to the look-up 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 apparatus 100 refers to the look-up table using the transposed data TR2, it refers only to the codebook information TB2 and refers to only one codebook information TB2. Similarly, for the transposed data TR3 and TR4, only one codebook information is referred to. As a result, the information processing apparatus 100 can efficiently refer to the data, so that efficient search processing can be enabled.

In this way, the data of the object near the node (connection node) is transposed so that the lookup table for each subclass (subvector, etc.) is referenced when calculating the batch distance, and the data is noded in the order in which they are organized for each subclass. By having it, the locality of reference is further enhanced and the speed can be increased.

Note that FIG. 3 shows a case where transposed data is generated for four nodes for the sake of simplicity, but the unit (number of nodes) for generating transposed data is based on the specifications of the information processing apparatus 100. Will be decided. For example, when the batch processable unit of the information processing apparatus 100 is "4", the processing is the same as in FIG. 3, but when the batch processable unit is "16", 16 nodes are regarded as one unit. Transposed data is generated. Further, when the batch processable unit is "32", the transposed data is generated with 32 nodes as one unit. In this way, by using the translocation data generated according to the number (collective processable unit) that the information processing apparatus 100 can collectively process by SIMD, the information processing apparatus 100 can efficiently refer to the data. Therefore, efficient search processing can be enabled.

For example, most of the time during the search process as described above is distance calculation, but the distance calculation can be speeded up by parallel calculation by SIMD as described above. When the distance calculation is parallelized in this way, the time to fetch the object data occupies the search process. If this fetch time can be reduced, further speedup will be realized. There is a method of performing prefetch to reduce the time for fetching data, but there is a limit.

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

(search results)
When returning the search result by the above-mentioned second distance (true distance), the following two methods, the first method and the second method, can be considered.

For example, as the first method, the number of searches may be searched more than the specified number of searches, and at the end of the search, the true distance calculation may be performed to sort by distance, and the objects with the specified number of searches may be used as the search result. good. In this case, the information processing apparatus 100 extracts the number of extended searches (also referred to as "extended search number") more than the specified number of searches (also referred to as "first number") (also referred to as "extended search number"). By setting "2 numbers") as the number of searches and performing the search process as shown in FIG. 11, nodes with an extended search number, that is, nodes with a number larger than the specified number of searches, are extracted as neighborhood candidate nodes. do. Then, the information processing apparatus 100 calculates a second distance (true distance) for the neighborhood candidate node, and among the neighborhood candidate nodes, the first node from the one with the shortest second distance is searched for in the vicinity of the search query. Extract as a node of.

Further, for example, as a second method, the second distance (true) is obtained only when the node (object) is within the search range or within the search range based on the first distance (approximate distance) during the search. Distance) may be calculated and the approximate distance may be replaced with the true distance to return the search result by the true distance.

The search range referred to here is a 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. The range. For example, when calculating the second distance (true distance) when entering the search range, the information processing apparatus 100 determines that the node when the first distance (approximate distance) of the node is "r" or less. The second distance (true distance) of is calculated. Further, when calculating the second distance (true distance) when entering the search range, the information processing apparatus 100 determines that the first distance (approximate distance) of the node is "r (1 + ε)" or less. , Calculate the second distance (true distance) of that node.

For example, in order to improve the accuracy, the condition may be that the search range is entered. Further, when speeding up the processing is required, it may be a condition that the process is within the search range. It should be noted that the above is only an example, and which condition is used may be appropriately set according to the value of the search range coefficient “ε”, the purpose of processing, and the like.

(Vector quantization)
As shown in Patent Document 3, the information processing apparatus 100 may perform two-step vector quantization. Then, the information processing apparatus 100 may calculate the distance between the search query and each node by the following equation (1).

Here, the value on the left side in the above equation (1) indicates, for example, the squared distance between the search query and the node. Further, for example, "x" in the above equation (1) corresponds to a query. Further, for example, "y" in the above equation (1) corresponds to a node. Further, for example, "q _c (y)" in 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 the vector data of the nodes for "y" in the above equation (1), the numerical value of the centroid of the partial region to which each node belongs may be used. Further, for example, "yq _c (y)" indicates a residual vector. Further, for example, "q _p " in the right side of the above equation (1) indicates a predetermined quantizer (function).

Further, for example, "j" in the right side of the above equation (1) may be the number of divided spaces. For example, in the example of FIG. 1, "j" in the right side of the above equation (1) may be the number of divided spaces "4". Further, for example, "u _j ()" in the right side of the above equation (1) indicates a partial residual vector between the vectors in parentheses. For example, the information processing apparatus 100 may calculate the squared distance between the query and the node in each subspace using the above equation (1) and calculate the distance between the query and the node by adding them together. 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 to each other via a predetermined network N so as to be communicable by wire or wirelessly. FIG. 4 is a diagram showing a configuration example of the information processing system according to the first embodiment. The information processing system 1 shown in FIG. 4 may include a plurality of terminal devices 10, a plurality of information providing devices 50, and a plurality of information processing devices 100.

The terminal device 10 is an information processing device used by the user. The terminal device 10 accepts various operations by the user. In the following, the terminal device 10 may be referred to as a user. That is, in the following, the user can be read as the terminal device 10. The terminal device 10 described above 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 in which information for providing various information to a user or the like is stored. For example, the information providing device 50 stores an object ID based on character information or the like collected from various external devices such as a web server. For example, the information providing device 50 is an information processing device that provides an image search service to users and the like. For example, the information providing device 50 stores each 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 targeted by an image search service. Further, the information providing device 50 receives an object ID or the like indicating an image corresponding to the query from the information processing device 100 by transmitting the query to the information processing device 100.

The information processing device 100 is a computer that provides a search service. The information processing apparatus 100 provides as a search result a specified number of objects corresponding to a search query. The information processing apparatus 100 provides a node (object) in the vicinity of a search query as a search result by using a graph in which nodes (objects) to be searched are connected by edges. The information processing apparatus 100 uses a graph in which nodes corresponding to each of a plurality of objects are connected by edges to search for nodes in the vicinity of a search query, and the information processing apparatus 100 has a plurality of nodes selected according to a predetermined criterion. The distance between and the search query is calculated using the vector information of multiple nodes that have undergone vector quantization.

For example, when the information processing apparatus 100 receives a query (search query) from the terminal apparatus 10, it searches for an object (object) similar to the search query and provides the search result to the terminal apparatus. Further, for example, the data provided by the information processing device 100 to the terminal device may be the data itself such as image information, or the information for referring to the corresponding data such as a URL (Uniform Resource Locator). May be good. Further, the search query and the search target (object) may be any kind of data such as image, voice, and text data.

[1-3. Information processing device configuration]
Next, the configuration of the information processing apparatus 100 according to the first embodiment will be described with reference to FIG. FIG. 5 is a diagram showing a configuration example of the information processing apparatus 100 according to the first embodiment. As shown in FIG. 5, the information processing apparatus 100 includes a communication unit 110, a storage unit 120, and a control unit 130. The information processing device 100 includes an input unit (for example, a keyboard, a mouse, etc.) that receives various operations from the administrator of the information processing device 100, and a display unit (for example, a liquid crystal display, etc.) for displaying various information. You may have.

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

(Memory 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 (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 about the object. For example, the object information storage unit 121 stores object IDs and vector data. FIG. 6 is a diagram showing 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”.

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

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

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

(Graph information storage unit 122)
The graph information storage unit 122 according to the first embodiment stores various information related to the graph. For example, the graph information storage unit 122 stores the generated graph. FIG. 7 is a diagram showing 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 "connection node information".

The "node ID" indicates identification information for identifying each node (target) in the graph. Further, the "object ID" indicates identification information for identifying an object. When the node ID and the object ID are common, the object ID is stored in the "node ID", and the item of the "object ID" may not be included in the graph information storage unit 122. For example, when used as an object ID and a node ID, the object ID may be stored in the "node ID", and the item of the "object ID" may not be included in the graph information storage unit 122.

Further, "connection node information" indicates information about a node (referenced node) that can be traced from the corresponding node. For example, "connection 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 by an edge from the node is associated with and registered with respect to the node ID (object ID) that identifies the node. The "connection node information" may include information (edge ID) for identifying an edge connected to the reference destination.

In the example of FIG. 7, it is shown that the node (node N1) identified by the node ID “N1” corresponds to the object (target) identified by the object ID “OB1”. Further, from the node N1, it is shown that the edge is connected to the node (node N4) identified by the node ID "N4", and the node N1 can be traced to the node N4.

Further, it is shown that the node (node N2) identified by the node ID "N2" corresponds to the object (target) identified by the object ID "OB2". Further, it is shown that the edge is connected to the node (node N6) identified by the node ID "N6" from the node N2, and the node N2 can be traced to the node N6.

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

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). The graph information storage unit 122 is not limited to the above, and may store graph information according to various data structures.

Further, the graph may include a program module that takes a query as an input, searches for a node by tracing an edge in the graph, and extracts and outputs a node similar to the query. That is, the graph may be expected to be used as a program module that performs search processing using the graph. For example, the graph GR1 may be a program that extracts and outputs a node corresponding to the vector data similar to the vector data from the graph when the vector data is input as a query. For example, the graph GR1 may be data used as a program module for searching a similar image corresponding to a query image. For example, graph GR1 causes a computer to extract and output nodes similar to the query in the graph based on the input query.

(Quantization information storage unit 123)
The quantized information storage unit 123 according to the first embodiment stores various information related to the allocation process. FIG. 8 is a diagram showing an example of a quantized information storage unit according to the first embodiment. In the example of FIG. 8, the quantized information storage unit 123 has items such as “node ID”, “object ID”, and “quantized information”.

The "node ID" indicates identification information for identifying each node (target) in the graph. Further, the "object ID" indicates identification information for identifying an object. When the node ID and the object ID are common, the object ID is stored in the "node ID", and the item of the "object ID" may not be included in the quantization information storage unit 123.

Further, "quantization information" indicates information on the quantized vector 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 case where the "element" includes "# 1", "# 2", "# 3", and "# 4" is shown. In this case, the vector of each node (object) is divided into four, and it is shown that each divided partial vector is quantized by the codebook. The number of divisions is not limited to 4, for example, when the number of divisions is 6, the "elements" include "# 1", "# 2", "# 3", "# 4", "# 5", and "# 5". # 6 "is included. The "codebook ID" indicates information for identifying the codebook corresponding to each element (partial vector).

In the example of FIG. 8, the vector of the node N9 (object OB9) is quantized by the codebook (codebook CD12) in which the first partial vector of the four-divided partial vectors is identified by the codebook ID “CD12”. Indicates that it will be done. Further, the vector of the node N9 (object OB9) is quantized by the codebook (codebook CD23) in which the second partial vector from the beginning is identified by the codebook ID "CD23" among the four-divided partial vectors. Indicates that.

Further, the vector of the node N9 (object OB9) is quantized by the codebook (codebook CD35) in which the third partial vector from the beginning is identified by the codebook ID "CD35" among the four-divided partial vectors. Indicates that. Further, the vector of the node N9 (object OB9) is a codebook (codebook CD47) in which the fourth (that is, the last) partial vector from the beginning of the four-divided partial vectors is identified by the codebook ID "CD47". ) Indicates that it is quantized.

The quantized 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 related to the codebook. For example, the codebook information storage unit 124 stores the codebook ID and the vector information of each codebook. FIG. 9 is a diagram showing 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 shows as an example a case where a lookup table showing the correspondence between each codebook and a vector is stored.

The codebook information storage unit 124 uses the codebook information TB1 used to quantize the first partial vector of the four-divided subvectors, and the second subvector from the beginning of the four-divided subvectors. The codebook information TB2 used for quantization is stored. Further, the codebook information storage unit 124 uses codebook information TB3 for quantizing the third partial vector from the beginning among the four divided partial vectors, and four from the beginning among the four divided partial vectors. The codebook information TB4 and the like used for quantizing the second (that is, the last) partial vector are stored.

In the example of FIG. 9, the codebook information TB1 includes a codebook (codebook CD11) identified by the codebook ID “CD11”, a codebook (codebook CD12) identified by the codebook ID “CD12”, and the like. Memorize codebook information. For example, the codebook CD11 shows that the multidimensional vector information of "5, 13 ..." Is associated with each other. Further, the codebook CD12 shows that the multidimensional vector information of "27, 51 ..." Is associated with the codebook CD12.

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 description of FIG. 5, the control unit 130 is a controller, and is inside the information processing device 100 by, for example, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), or the like. Various programs (corresponding to an example of an information processing program) stored in the storage device of the above are realized by executing 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 an information processing function or operation described below. do. 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 is configured to perform information processing described later.

(Acquisition unit 131)
The acquisition unit 131 acquires various types of 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 the graph. For example, the information processing apparatus 100 may acquire the 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 a search query for a plurality of objects to be searched for data. For example, the acquisition unit 131 acquires information about the search query QE1. For example, the acquisition unit 131 acquires a search query related to an 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 the 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, and the like. ..

The generation unit 132 may generate a graph as shown in the graph information storage unit 122. For example, the generation unit 132 generates the 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 about the codebook as shown in the codebook information storage unit 124. The generation unit 132 may generate a codebook look-up table. For example, the generation unit 132 generates information about the codebook, such as codebook information TB1 to TB4. When the information processing device 100 acquires the information shown in the graph information storage unit 122, the quantization information storage unit 123, and the codebook information storage unit 124 from an external device such as the information providing device 50, the information processing device 100 may obtain the information. It is not necessary to have the generation unit 132.

(Search processing unit 133)
The search processing unit 133 provides a search service for objects. The search processing unit 133 searches for various types of information. The search processing unit 133 searches for various types of information. For example, the search processing unit 133 searches for an object by searching the 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 the graph. For example, the search processing unit 133 extracts an object similar to a query by searching the graph. For example, the search processing unit 133 extracts an object similar to a query by searching the graph based on the processing procedure as shown in FIG. If the information processing apparatus 100 does not provide the search service, the information processing apparatus 100 does not have to have the search processing unit 133.

The search processing unit 133 selects various information in the search processing. The search processing unit 133 extracts various information in the search processing. The search processing unit 133 determines various information in the search processing. The search processing unit 133 determines various information in the search processing. The search processing unit 133 changes various information in the search processing. The search processing unit 133 updates various information in the search processing.

The search processing unit 133 uses a graph in which nodes corresponding to each of a plurality of objects are connected by edges to search for nodes in the vicinity of the search query, and the search processing unit 133 selects a plurality of nodes according to a predetermined criterion. The distance between and the search query is calculated using the vector information of multiple nodes that have undergone vector quantization. The search processing unit 133 performs the calculation of the distance between the plurality of nodes and the search query in parallel and collectively. The search processing unit 133 processes in parallel the calculation of the distance between the plurality of nodes of the batch processing number determined based on the specifications of the information processing apparatus 100 and the search query.

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

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

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

The search processing unit 133 does not perform vector quantization on a predetermined node among the nodes processed in the search processing, unlike the first distance, which is the distance obtained by vector quantization. Calculate the 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 a node in the vicinity of the search query in the search process, as a neighborhood candidate node, and targets the neighborhood candidate node as a second node. Calculate the distance. The search processing unit 133 extracts the first number of nodes from the one with the shortest second distance among the neighborhood candidate nodes as the nodes in the vicinity of the search query.

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

(Providing section 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 the object ID corresponding to the search query as a search result. The providing unit 134 transmits the search result 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 object ID extracted by the search processing unit 133 to the information providing device 50. The providing unit 134 provides the information providing device 50 with the object ID extracted by the search processing unit 133 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 with reference to FIG. FIG. 10 is a flowchart showing an example of information processing according to the first embodiment.

As shown in FIG. 10, the information processing apparatus 100 acquires a search query for a plurality of objects to be searched for data (step S101). In the example of FIG. 1, the information processing apparatus 100 acquires the search query QE1.

Then, the information processing apparatus 100 uses a graph in which nodes corresponding to each of the plurality of objects are connected by edges to search for nodes in the vicinity of the search query, and the information processing apparatus 100 is selected by a predetermined criterion. The distance between the node and the search query is calculated using the vector information of a plurality of nodes that have undergone vector quantization (step S102). In the example of FIG. 1, the information processing apparatus 100 uses a vector of a plurality of nodes N9, N12, N54, N85 in which the distance between the plurality of nodes N9, N12, N54, N85 and the search query QE1 is vector-quantized. Calculate using information.

[1-5. Search processing example]
Here, an example of the search process according to the first embodiment will be described with reference to FIG. 11 as an example. FIG. 11 is a flowchart showing an example of the search process according to the first embodiment. The search process described below is performed by the search process unit 133 of the information processing apparatus 100. In addition, the objects referred to below may be read as nodes. In the following, the information processing apparatus 100 (search processing unit 133) performs search processing. If the search service is not provided, the information processing apparatus 100 does not have to have the search processing unit 133. The search query of the process described below may be an additional node, a target node, an object specified by the user, or the like.

Here, the neighborhood object set N (G, y) is a set of neighborhood objects associated with the edge assigned to the node y. “G” may be predetermined graph data (for example, graph GR1 shown in spatial information SP1). For example, the information processing apparatus 100 executes the k-nearest neighbor search process.

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

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

Next, the information processing apparatus 100 determines whether or not the distance d (s, y) between the object s and the object y exceeds r (1 + ε) (step S304). Here, ε is an extension element, and r (1 + ε) is a value indicating the radius of the search range (searching only the nodes within this range. The accuracy can be improved by making it larger than the search range). be. When the distance d (s, y) between the object s and the object y exceeds r (1 + ε) (step S304: Yes), the information processing apparatus 100 outputs the object set R as a neighborhood object set of the object y (step). S305), the process is terminated.

When 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 apparatus 100 determines the neighborhood object set N (G, s) of the objects s. One object not included in the object set C is selected from the objects that are the elements of the above, and the selected object u is stored in the object set C (step S306). The object set C is provided for convenience in order to avoid duplicate search, and is set to an empty set at the start of processing.

Next, the information processing apparatus 100 determines whether or not the distance d (u, y) between the object u and the object y is r (1 + ε) or less (step S307). When the distance d (u, y) between the object u and the object y is r (1 + ε) or less (step S307: Yes), the information processing apparatus 100 adds the object u to the object set S (step S308). Further, when the distance d (u, y) between the object u and the object y is not r (1 + ε) or less (step S307: No), the information processing apparatus 100 determines (processes) step S309.

Next, the information processing apparatus 100 determines whether or not the distance d (u, y) between the object u and the object y is r or less (step S309). When the distance d (u, y) between the object u and the object y exceeds r (step S309: No), the information processing apparatus 100 determines (processes) step S315. That is, when the distance d (u, y) between the object u and the object y is not r or less, the information processing apparatus 100 performs the determination (processing) in step S315.

When the distance d (u, y) between the object u and the object y is r or less (step S309: Yes), the information processing apparatus 100 adds the object u to the object set R (step S310). Then, the information processing apparatus 100 determines whether or not 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 the upper limit of the number of objects to be extracted in the range search or the like is not set, ks may be set to infinity. For example, ks = 4 or the like may be used. When the number of objects included in the object set R does not exceed ks (step S311: No), the information processing apparatus 100 determines (processes) step S313.

When the number of objects included in the object set R exceeds ks (step S311: Yes), the information processing apparatus 100 selects the object having the longest distance (far) from the object y among the objects included in the object set R. Exclude from the object set R (step S312).

Next, the information processing apparatus 100 determines whether or not the number of objects included in the object set R matches ks (step S313). When the number of objects included in the object set R does not match ks (step S313: No), the information processing apparatus 100 determines (processes) step S315. Further, when the number of objects included in the object set R matches ks (step S313: Yes), the information processing apparatus 100 has the longest distance (far) from the object y among the objects included in the object set R. The distance between the object and the object y is set to a new r (step S314).

Then, the information processing apparatus 100 selects all the objects from the objects that are the elements of the object set N (G, s) in the vicinity of the objects s, and determines whether or not the objects have been stored in the object set C (step S315). .. When all the objects have been selected from the objects which are the elements of the object set N (G, s) in the vicinity of the objects and stored in the object set C (step S315: No), the information processing apparatus 100 has step S306. Return to and repeat the process.

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

[2. Second embodiment]
From here, the second embodiment will be described. In the second embodiment, neighboring objects are grouped by clustering or the like, and each group (hereinafter, also referred to as “blob”) is targeted for performing distance calculation collectively. That is, in the second embodiment, the batch distance calculation is performed for each blob. The same points as in the first embodiment will be omitted as appropriate. In the second embodiment, the information processing system 1 has an information processing device 100A instead of the information processing device 100.

[2-1. Information processing]
First, the outline of the information processing according to the second embodiment will be described with reference to FIG. FIG. 12 is a diagram showing an example of information processing according to the second embodiment.

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

First, each information shown in FIG. 12 will be described. The dotted line connecting the white circles (◯), which are the nodes in the spatial information SP21 in FIG. 12, indicates the edge connecting the nodes. In FIG. 12, an undirected edge is shown as an example as in FIG. 1, but the edge of the graph GR 21 is not limited to the undirected edge and may be a directed edge. Since the notes and edges are the same as those in FIG. 1, detailed description thereof will be omitted.

In the spatial information SP21 of FIG. 12, the area surrounded by a straight line is a group of nearby objects by clustering or the like, and the area is referred to as a blob. In the example shown below, the blob is the processing unit for batch distance calculation. FIG. 12 shows a case where clustering is performed in 10 regions (blobs) of blobs BL1 to BL10. For example, blob BL1 indicates that it is a blob to which a plurality of nodes such as nodes N7, N9, N85, and N126 belong. The black dots in the blobs BL1 to BL10 indicate the centroid (representative vector) of each blob.

The blobs BL1 to BL10 are generated by appropriately using various methods related to clustering and the like. For example, the clustering for classifying into blobs may be k-means clustering. Also, for example, in clustering for classifying into blobs, the centroid obtained at the center coordinates (average) of each cluster in one iteration (assignment) in k-means is not used for the next iteration. Alternatively, you may replace the centroid with the object closest to each centroid before performing the next iteration. In this case, the centroid of the final cluster obtained by k-means is an existing object, that is, a node. As a result, the tendency for the cluster (blob) and the edge (neighboring node) of each node to match increases, and the search performance can be further improved.

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

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

Then, the information processing apparatus 100A calculates the distance between the vector of the search query QE2 and the vector of the codebook. Although not shown, the information processing apparatus 100A calculates the distance (difference) between the vector of the search query QE2 and the vector of the codebook for quantizing the entire vector when the direct product quantization is not performed. .. Then, the information processing apparatus 100A holds the distance (difference) between the vector of the search query QE2 and the vector of each codebook as the codebook information TB21 in association with the information for identifying each codebook. When direct product quantization is performed, the information processing apparatus 100A has the same as FIG. 1, as shown in the codebook information TB1 to TB4 of FIG. 2, each subvector of the search query QE2 and the vector of each codebook. Calculate the distance (difference) between and. Since the distance calculation between the search query vector and the codebook vector is the same as in FIGS. 1, 2, etc., detailed description thereof will be omitted.

The information processing apparatus 100A executes a search process targeting the search query QE2 (step S22). The information processing apparatus 100A obtains the search result of the search query QE2 by performing the 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 apparatus 100A performs a search process on the search query QE2 to extract the nodes having the number of searches as the nodes in the vicinity of the search query QE2.

In the search process for the search query QE2, the information processing apparatus 100A selects a predetermined node among the nodes as a node (starting point node) to be the starting point of the search of the graph GR21. In the example of FIG. 12, the information processing apparatus 100A selects the node N9 as the starting node. In the example of FIG. 12, the information processing apparatus 100A starts the search process with the node N9 as the processing target, for example.

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

The information processing apparatus 100A calculates the distance between each of the nodes N7, N9, N85, and N126 and the search query QE2 by using the distance between each codebook indicated by the codebook information TB 21 and the search query QE2. For example, the information processing apparatus 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. Since the distance calculation in the case of direct product quantization is the same as in FIGS. 1, 2, etc., detailed description thereof will be omitted.

The batch processable unit is determined based on the specifications of the information processing apparatus 100A, as in the case of the information processing apparatus 100. For example, when the number (units that can be collectively processed) that the information processing apparatus 100A can process collectively by SIMD is "4", the distances are calculated in parallel for four nodes as in FIG. 1. For example, the information processing apparatus 100A collectively calculates the distance between each of the nodes N7, N9, N85, and N126 and the search query QE2 by parallelizing the SIMD operation. As a result, the information processing apparatus 100A can speed up the distance calculation by parallelizing the distance calculation, and can enable efficient search processing. Further, the information processing apparatus 100A traces the nodes (for example, nodes N12 and N64) to which the edges from the node N7 are connected by using the graph GR21, and targets the blobs (for example, blobs BL2 and BL3) to which those nodes belong. Perform the batch distance calculation and perform the search process. The information processing apparatus 100A suppresses that the same blob is repeatedly subject to batch distance calculation in the search process, but the details will be described later.

Note that FIG. 12 shows a case where the distances are calculated in parallel for four nodes for the sake of simplicity, but the number to be parallelized is determined based on the specifications of the information processing apparatus 100A. .. Since this point is the same as in FIGS. 1, 2, etc., detailed description thereof will be omitted. Further, the information processing apparatus 100A may use transposed data as in the example shown in FIG.

As described above, the information processing apparatus 100A calculates an approximate distance (second distance) between the vector of each vector quantized node and the search query, and performs the search process using the second distance. Therefore, efficient search processing can be enabled. Further, the information processing apparatus 100A can enable efficient search processing by collectively calculating the distances of a number of nodes that can be parallelized. As described above, the information processing apparatus 100A can enable efficient search processing by collectively calculating blobs as a unit of batch processing.

In the search process, the information processing apparatus 100A traces the graph GR21 and performs the above-mentioned process on the node to be processed, thereby extracting the node having the number of searches as a node in the vicinity of the search query QE2. For example, the information processing apparatus 100A obtains a search result by any of the above-mentioned first method and the second method, similarly to the information processing apparatus 100.

The above-mentioned processing is only an example, and the information processing apparatus 100A may perform the search processing by using various information and methods for the efficient search processing. In this regard, each item will be described in detail. For example, the information processing apparatus 100A performs clustering so that the number of nodes belonging to each blob is the number that the information processing apparatus 100 can collectively process by SIMD (collective processable unit), and generates blobs. May be good.

Further, the information processing apparatus 100A may suppress duplicate processing by using the information about the blob. Even if the information processing apparatus 100A has a blob distance calculation flag (hereinafter, also simply referred to as “flag”) indicating whether the blob distance calculation has been performed for each blob, and a table indicating which blob each object belongs to. good. In this case, the information processing apparatus 100A may manage whether or not each blob has been processed by the flag. For example, the information processing apparatus 100A identifies the blob to which the node belongs from the table immediately before processing the neighboring nodes one by one, determines whether the blob has been calculated for the batch distance with a flag, and if it has not been calculated, performs the batch calculation. It may be done and processing of individual nodes may be performed. This point will also be described with reference to FIGS. 15 and 16. As a result, the information processing apparatus 100A can prevent duplicate processing from being performed.

[2-2. Information processing device configuration]
Next, the configuration of the information processing apparatus 100A according to the second embodiment will be described with reference to FIG. 13. FIG. 13 is a diagram showing a configuration example of the information processing apparatus according to the second embodiment. As shown in FIG. 13, the information processing apparatus 100A includes a communication unit 110, a storage unit 120A, and a control unit 130A. In the information processing apparatus 100A, the same points as those of the information processing apparatus 100 will be omitted as appropriate.

(Memory unit 120A)
The storage unit 120A is realized by, for example, a semiconductor memory element 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. It has an information storage unit 125.

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

The "node ID" indicates identification information for identifying each node (target) in the graph. Further, the "object ID" indicates identification information for identifying an object. When the node ID and the object ID are common, the object ID is stored in the "node ID", and the item of the "object ID" may not be included in the quantized information storage unit 123A.

The "blob ID" indicates information for identifying the blob to which the node (object) belongs.

Further, "quantization information" indicates information on the quantized vector 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 case where the "element" includes "# 1", "# 2", "# 3", and "# 4" is shown. In this case, the vector of each node (object) is divided into four, and it is shown that each divided partial vector is quantized by the codebook.

In the example of FIG. 14, the node N1 (object OB1) is shown to belong to the blob (blob BL9) identified by the blob ID “BL9”. For example, when the vector is divided into four by the direct product quantization, the quantization information of the node N1 stores the same information as the four codebooks showing the quantization information of the node N1 shown in FIG. Further, for example, when the direct product quantization is not performed, the information indicating one codebook for quantizing the vector of the node N1 is stored in the quantization information of the node N1.

The quantized information storage unit 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 about the blob. For example, the blob information storage unit 125 stores various information that identifies an object associated with each blob. FIG. 15 is a diagram showing 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, the "node ID" indicates a node (object) associated with the blob identified by the blob ID. "Vector information" indicates the vector information of the blob. For example, "vector information" indicates a vector corresponding to the blob's centroid.

In the example shown in FIG. 15, it is shown that the node (object) associated with the blob (blob BL1) identified by the blob ID “BL1” is the node N7, N9, N85, N126 or the like. The blob BL1 indicates that the multidimensional vector information of "51, 4, 102, 33 ..." Is associated with each other.

Further, it is shown that the node (object) associated with the blob (blob BL9) identified by the blob ID "BL9" is a node N1, N4, N5 or the like. The blob BL9 indicates that the multidimensional vector information of "12, 55, 12, 6 ..." Is associated with the vector information.

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

For example, the information processing apparatus 100A refers to the value of the flag of the blob to be processed, and determines whether or not to execute the batch distance calculation process for the blob. The information processing apparatus 100A refers to the value of the flag of each blob, and if the value of the flag of the blob is the unprocessed flag value, determines that the blob is before the processing of the batch distance calculation, and determines that the blob. Performs batch distance calculation for nodes belonging to. On the other hand, when the value of the flag of the blob is the processed flag value, the information processing apparatus 100A determines that the blob has been processed for 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 description of FIG. 13, the control unit 130A is a controller, and various programs (information processing programs) stored in the storage device inside the information processing device 100A by, for example, a CPU, MPU, GPU, or the like. (Corresponding to one example) is realized by executing 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 FPGA.

As shown in FIG. 13, the control unit 130A has an acquisition unit 131, a generation unit 132A, a search processing unit 133A, and a provision unit 134, and realizes or executes an information processing function or operation described below. do. 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 is configured to perform information processing described later.

(Generator 132A)
The generation unit 132A generates various information in the same manner as the generation unit 132.

The generation unit 132A may generate information on 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 about the blob as shown in the blob information storage unit 125. For example, the generation unit 132 generates information indicating that the nodes N7, N9, N85, N126 and the like belong to the blob BL1. When the information processing device 100 acquires the information shown in the graph information storage unit 122, the quantization information storage unit 123A, the codebook information storage unit 124, and the blob information storage unit 125 from an external device such as the information providing device 50. The information processing device 100 does not have to have the generation unit 132A.

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

The search processing unit 133A uses the information of a plurality of blobs that classify a plurality of objects in a search process for searching a node in the vicinity of a search query for a node group corresponding to each of the plurality of objects. The distance between a plurality of nodes belonging to a blob and a search query is calculated using the vector information of the plurality of vector-quantified nodes. The search processing unit 133A performs the calculation of the distance between the plurality of nodes and the search query in parallel and collectively. The search processing unit 133A processes in parallel the calculation of the distance between the plurality of nodes of the batch processing number determined based on the specifications of the information processing apparatus 100A and the search query.

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

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

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

The search processing unit 133A is different from the first distance, which is the distance in which vector quantization is performed for a predetermined node among the nodes processed in the search processing, and the search processing unit 133A does not perform vector quantization. Calculate the distance. The search processing unit 133A extracts a second number of nodes as a neighborhood candidate node, which is a larger number than the first number of nodes to be extracted as a node in the vicinity of the search query in the search processing, and targets the neighborhood candidate node as the second node. Calculate the distance. The search processing unit 133A extracts the first number of nodes from the one with the shortest second distance among the neighborhood candidate nodes as the nodes in the vicinity of the search query.

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

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

Here, the neighborhood object set N (G, y) is a set of neighborhood objects associated with the edge assigned to the node y. “G” may be predetermined graph data (for example, graph GR21 shown in spatial information SP21). For example, the information processing apparatus 100A executes the k-nearest neighbor search process.

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. As described above, 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 the blob and the query is collectively calculated. , Process each node.

Specifically, in the search process of FIG. 16, when the blob to which the object s belongs has not yet calculated the distance, the information processing apparatus 100A has all of the following (“−” in step S304a in FIG. The four processes shown in succession to the above; hereinafter referred to as "first process" to "fourth process") are executed (step S304a). In step S304a, the information processing apparatus 100A executes a process (first process) of collectively calculating the distance of the objects in the blob. Further, in step S304a, the information processing apparatus 100A executes a process (second process) of setting the blob distance calculation flag. Further, in step S304a, the information processing apparatus 100A executes a process (third process) of storing all the objects in the blob for which the batch distance calculation has been performed in C. Further, in step S304a, the information processing apparatus 100A executes a process (fourth process) of performing the processes from step S307 to step S314 one by one with all the objects of the blob as u. After that, the information processing apparatus 100A performs the processing after step S306.

[2-4. Modification example]
From here, a modification will be described. In the modification according to the second embodiment, a graph including the concept of blob may be used. For example, in the modification according to the second embodiment, a graph in which the reference destination by the edge from the node is a blob may be used. The same points as those of the first embodiment and the second embodiment will be omitted as appropriate. The information processing apparatus 100A according to the modified example has a graph information storage unit 122A instead of the graph information storage unit 122.

[2-4-1. Information processing]
First, the outline of information processing according to the modified example will be described with reference to FIG. FIG. 17 is a diagram showing an example of information processing according to a modified example.

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

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

For example, the information processing apparatus 100A converts the edge of each node from the edge to the connecting node (neighboring node) to the edge to the blob to which the connecting node belongs. The information processing apparatus 100A does not generate an edge 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 apparatus 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 apparatus 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 apparatus 100A performs the search process using the graph GR31 including the directed edge from the node (object) to the blob, as shown in the spatial information SP31. For example, the information processing apparatus 100A obtains the search result of the search query QE2 by performing the 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. The search process in FIG. 17 is common to FIG. 12 in that the search process is performed using the concept of blob, and is the same as the search process shown in FIG. 12 except for the process related to the edge.

[2-4-2. graph〕
Next, the outline of the graph according to the modified example will be described with reference to FIG. FIG. 18 is a diagram showing an example of a graph information storage unit according to a modified example. For example, the graph information storage unit 122A shown in FIG. 18 stores a graph (graph for blobs) into which the concept of blobs is introduced. The graph information storage unit 122A shown in FIG. 18 has items such as “node ID”, “object ID”, and “blob information”. As described above, the graph information storage unit 122A according to the modified example is different from the graph information storage unit 122 in FIG. 7 in that it has "blob information" instead of "connection node information". The same points as the graph information storage unit 122 in FIG. 7 will be omitted as appropriate.

Further, "blob information" indicates information about blobs (referenced blobs) that can be traced from the corresponding node. For example, "blob information" includes information such as "reference destination". "Reference destination" indicates information for identifying a reference destination (brob) that is connected by an edge and can be traced from the node. That is, in the example of FIG. 18, a reference destination (blob) that can be traced by an edge from the node is associated with and registered with respect to the node ID (object ID) that identifies the node. The "blob information" may include information (edge ID) for identifying an edge connected to a reference destination.

In the example of FIG. 18, it is shown that the node (node N1) identified by the node ID “N1” corresponds to the object (target) identified by the object ID “OB1”. Further, it is shown that the edge is connected to the blob (blob BL8) identified by the blob ID "BL8" from the node N1 and the blob BL8 can be traced from the node N1. Further, it is shown that the edge is connected to the blob (blob BL10) identified by the blob ID "BL10" from the node N1 and the blob BL10 can be traced from the node N1.

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

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

Further, the graph may include a program module that takes a query as an input, searches for a node by tracing an edge in the graph, and extracts and outputs a node similar to the query. That is, the graph may be expected to be used as a program module that performs search processing using the graph. For example, the graph GR 31 may be a program that extracts and outputs a node corresponding to the vector data similar to the vector data from the graph when the vector data is input as a query. For example, the graph GR 31 may be data used as a program module for searching a similar image corresponding to the query image. For example, the graph GR 31 makes a computer function to extract and output a node similar to the query in the graph based on the input query.

[2-4-3. Search processing example]
From here, an example of the search process according to the modified example will be described with reference to FIG. 19 as an example. FIG. 19 is a flowchart showing an example of the search process according to the modified example. The search process described below is performed by the search process unit 133A of the information processing apparatus 100A. It should be noted that the same points as 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 points as 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) differs from the search process of FIGS. 11 and 16 in that the set N (G, y) is a set of blobs associated with the edges assigned to the node y. In FIG. 19, N (G, s) and C are blob sets. That is, the "neighborhood object set N" in FIG. 11 is read as "blob set N" in FIG. 19, and the "object set C" in FIG. 11 is read as "blob set C" in FIG. .. In addition, the wording of "object" in the process related to the above change to blob is appropriately read as "blob". “G” may be graph (blob graph) data (for example, graph GR31 shown in spatial information SP31) into which the concept of blob is introduced. For example, the information processing apparatus 100A executes the k-nearest neighbor search process.

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

Further, the search process shown in FIG. 19 is different 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 apparatus 100A selects one object u from the object set B in the target blob (step S306a). After that, the information processing apparatus 100A performs the processing after step S307.

Further, 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 apparatus 100A determines whether or not all the objects in the target blob have been selected (step S314a).

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

[3. effect〕
As described above, the information processing apparatus 100 according to the first embodiment has an acquisition unit 131 and a search processing unit 133. The acquisition unit 131 acquires a search query for a plurality of objects to be searched for data. The search processing unit 133 uses a graph in which nodes corresponding to each of a plurality of objects are connected by edges to search for nodes in the vicinity of the search query, and the search processing unit 133 selects a plurality of nodes according to a predetermined criterion. The distance between and the search query is calculated using the vector information of multiple nodes that have undergone vector quantization.

As described above, the information processing apparatus 100 according to the first embodiment is performed by using vector-quantified information for a plurality of nodes whose distance from the search query is selected according to a predetermined criterion. It is possible to reduce the calculation load for multiple nodes. Therefore, the information processing apparatus 100 can enable efficient search processing.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 calculates the distances between the plurality of nodes and the search query in parallel and collectively performs the calculation.

As described above, the information processing apparatus 100 according to the first embodiment can enable efficient search processing by performing the calculation of the distance from the search query in parallel for a plurality of nodes in a batch. can.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 calculates the distance between the plurality of nodes of the batch processing number determined based on the specifications of the information processing apparatus 100 and the search query in parallel. To process.

As described above, the information processing apparatus 100 according to the first embodiment performs the calculation of the distance from the search query in parallel for a plurality of nodes that can be collectively processed by the information processing apparatus 100. Therefore, efficient search processing can be enabled.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 uses a codebook associated with a representative vector to which each of the plurality of nodes corresponds to the plurality of nodes and the search query. Calculate the distance.

As described above, the information processing apparatus 100 according to the first embodiment can enable efficient search processing by using a codebook associated with a representative vector in which each of the plurality of nodes corresponds. can.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 calculates the distance between a plurality of nodes to which edges are connected from one node and a search query.

As described above, the information processing apparatus 100 according to the first embodiment performs efficient search processing by using vector-quantized information for a plurality of nodes that can be traced from a certain node. Can be made possible.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 associates a plurality of nodes with reference information indicating a representative vector associated with each of the plurality of nodes with one node. Using the stored node information, calculate the distance between multiple nodes and the search query.

As described above, in the information processing apparatus 100 according to the first embodiment, the reference information indicating the representative vector to which each of the plurality of nodes and the plurality of nodes is associated is stored in association with one node. By using information, efficient access to information becomes possible. Therefore, the information processing apparatus 100 can enable efficient search processing.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 uses the vector information of the plurality of nodes, each of which is divided into a plurality of partial vectors by direct product quantization, with the plurality of nodes. Calculate the distance to the search query.

As described above, the information processing apparatus 100 according to the first embodiment is efficient by calculating the distance between a plurality of nodes and the search query by using the information of the divided partial vectors by direct product quantization. Search processing can be enabled.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 is associated with a representative vector corresponding to a vector for each division position of a plurality of partial vectors in which each of the plurality of nodes is divided. Calculate the distance between multiple nodes and the search query using multiple codebooks.

As described above, the information processing apparatus 100 according to the first embodiment enables efficient search processing by using a codebook associated with the representative vector corresponding to the vector for each division position of the partial vector. can do.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 has reference information indicating a representative vector to which each of the plurality of nodes and the plurality of partial vectors of the plurality of nodes is associated with each other. Using the node information stored in association with one node, the distance between multiple nodes and the search query is calculated.

As described above, in the information processing apparatus 100 according to the first embodiment, the reference information indicating the representative vector to which each of the plurality of nodes and the plurality of partial vectors of each of the plurality of nodes is associated is combined with one node. Efficient access to information is possible by using the node information stored in association with each other. Therefore, the information processing apparatus 100 can enable efficient search processing.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 associates each of the plurality of nodes with a list of representative vectors to which each of the plurality of partial vectors of each node is associated. Calculate the distance between multiple nodes and the search query using the reference information.

As described above, in the information processing apparatus 100 according to the first embodiment, reference information is associated with a list of representative vectors to which each of the plurality of partial vectors of each node is associated with each of the plurality of nodes. By using, efficient access to information becomes possible. Therefore, the information processing apparatus 100 can enable efficient search processing.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 includes transposition information in which representative vectors corresponding to each of the plurality of partial vectors of the plurality of nodes are arranged in a list for each division position. , The distance between the plurality of nodes and the search query is calculated using the reference information including the corresponding information indicating the corresponding position in the list of each of the plurality of nodes.

As described above, in the information processing apparatus 100 according to the first embodiment, the transposition information in which the representative vectors corresponding to each of the plurality of partial vectors of the plurality of nodes are arranged in a list for each division position and the plurality of nodes Efficient access to the information is possible by using the reference information including the corresponding information indicating the corresponding position in the list. Therefore, the information processing apparatus 100 can enable efficient search processing.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 uses one codebook in which each of the plurality of partial vectors of each of the plurality of nodes is associated with the corresponding representative vector. , Calculate the distance between multiple nodes and the search query.

As described above, the information processing apparatus 100 according to the first embodiment can suppress an increase in the storage amount by using one codebook common to each division position of the partial vector.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 is a distance obtained by vector quantization for a predetermined node among the nodes processed in the search processing. Unlike the first distance, the second distance without vector quantization is calculated.

As described above, the information processing apparatus 100 according to the first embodiment performs search processing using vector-quantized information, and calculates a distance that is not vector-quantized for a predetermined node. It is possible to suppress an increase in search processing time while maintaining the accuracy of the search. Therefore, the information processing apparatus 100 can enable efficient search processing.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 uses 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 in the search processing. It is extracted as a neighborhood candidate node, and the second distance is calculated for the neighborhood candidate node.

As described above, the information processing apparatus 100 according to the first embodiment extracts nodes having a larger number of nodes as search results as candidates, and calculates the non-vector-quantized distances of those candidates. It is possible to suppress an increase in search processing time while maintaining the accuracy of the search. Therefore, the information processing apparatus 100 can enable efficient search processing.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 extracts the first number of nodes from the one with the shorter second distance as the nodes in the vicinity of the search query among the neighborhood candidate nodes.

As described above, the information processing apparatus 100 according to the first embodiment accurately sets as many nodes as the search results from the candidate having the shorter non-vector quantized distance as the neighboring nodes. A node whose distance has been calculated can be a neighboring node.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 calculates the second distance for the node whose first distance is within a predetermined threshold value.

As described above, the information processing apparatus 100 according to the first embodiment calculates the distance not vector-quantized only for the nodes whose vector-quantized distance is within the threshold value, thereby achieving the accuracy of the search. It is possible to suppress an increase in search processing time while maintaining the above. Therefore, the information processing apparatus 100 can enable efficient search processing.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 calculates the second distance for the nodes in the search range indicating the target range to be extracted as the neighboring nodes.

As described above, the information processing apparatus 100 according to the first embodiment has the accuracy of the search by calculating the distance not vector-quantized only for the nodes whose vector-quantized distance is within the search range. It is possible to suppress an increase in search processing time while maintaining the above. Therefore, the information processing apparatus 100 can enable efficient search processing.

Further, in the information processing apparatus 100 according to the first embodiment, the search processing unit 133 calculates the second distance for the nodes in the search range indicating the target range of the search processing.

As described above, the information processing apparatus 100 according to the first embodiment calculates the distance not vector-quantized only for the nodes whose vector-quantized distance is within the search range, thereby achieving the accuracy of the search. It is possible to suppress an increase in the search processing time while maintaining the above. Therefore, the information processing apparatus 100 can enable efficient search processing.

[4. Hardware configuration]
The information processing devices 100 and 100A according to each of the above-described embodiments are realized by, for example, a computer 1000 having a configuration as shown in FIG. FIG. 20 is a hardware configuration diagram showing an example of a computer that realizes 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.

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

The HDD 1400 stores a program executed by the CPU 1100, data used by such a program, and the like. The communication interface 1500 receives data from another device via the network N and sends it to the CPU 1100, and transmits the data generated by the CPU 1100 to the other device via the network N.

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

The media interface 1700 reads a program or data stored in the recording medium 1800 and provides the program or data to the CPU 1100 via the RAM 1200. The CPU 1100 loads the program from the recording medium 1800 onto the RAM 1200 via the 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 PD (Phase change rewritable Disk), a magneto-optical recording medium such as MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory. And so on.

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

Although some of the embodiments of the present application have been described in detail with reference to the drawings, these are examples, and various modifications are made based on the knowledge of those skilled in the art, including the embodiments described in the disclosure line of the invention. It is possible to carry out the present invention in other modified forms.

[5. others〕
Further, among the processes described in the above-described embodiment, all or a part of the processes described as being automatically performed can be manually performed, or the processes described as being manually performed can be performed. All or part of it can be done automatically by a known method. In addition, information including processing procedures, specific names, various data and parameters shown in the above documents and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each figure is not limited to the information shown in the figure.

Further, each component of each of the illustrated devices is a functional concept, and does not necessarily have to be physically configured as shown in the figure. That is, the specific form of distribution / integration of each device is not limited to the one shown in the figure, and all or part of them may be functionally or physically distributed / physically in any unit according to various loads and usage conditions. Can be integrated and configured.

In addition, the processes described in the above-described embodiments can be appropriately combined as long as the processing contents do not contradict each other.

Further, the above-mentioned "section, module, unit" can be read as "means" or "circuit". 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 Information processing device 120, 120A Storage unit 121 Object information storage unit 122, 122A Graph information storage unit 123, 123A Quantized information storage unit 124 Codebook information storage unit 125 Blob Information storage unit 130, 130A Control unit 131 Acquisition unit 132, 132A Generation unit 133, 133A Search processing unit 134 Providing unit

Claims (20)

An acquisition unit that acquires search queries for multiple objects to be searched for data,
In a search process for searching for a node in the vicinity of the search query using a graph in which a group of nodes corresponding to each of the plurality of objects is connected by an edge, a plurality of nodes selected by a predetermined criterion and the search query. A search processing unit that calculates the distance to and from using the vector information of the plurality of nodes that have been vector-quantified.
An information processing device characterized by being equipped with. The search processing unit
The information processing apparatus according to claim 1, wherein the calculation of the distance between the plurality of nodes and the search query is performed in parallel. The search processing unit
The information processing apparatus according to claim 2, wherein the calculation of the distance between the plurality of nodes and the search query, which is the number of batch processes determined based on the specifications of the information processing apparatus, is performed in parallel. The search processing unit
One of claims 1 to 3, wherein the distance between the plurality of nodes and the search query is calculated by using a codebook associated with a representative vector in which each of the plurality of nodes corresponds. The information processing device described in the section. The search processing unit
The information processing apparatus according to claim 4, wherein the distance between the plurality of nodes to which edges from one node are connected and the search query is calculated. The search processing unit
The reference information indicating the representative vector to which each of the plurality of nodes and the plurality of nodes is associated is associated with the one node, and the node information stored is used for the plurality of nodes and the search query. The information processing apparatus according to claim 5, wherein the distance between the node and the node is calculated. The search processing unit
5. Item 6. The information processing apparatus according to Item 6. The search processing unit
Using a plurality of codebooks associated with the representative vector corresponding to the vector for each division position of the plurality of subvectors in which each of the plurality of nodes is divided, the plurality of nodes and the search query are used. The information processing apparatus according to claim 7, wherein the distance is calculated. The search processing unit
Using the node information stored in association with the one node, reference information indicating the representative vector to which each of the plurality of nodes and the plurality of partial vectors of each of the plurality of nodes is associated is used. The information processing apparatus according to claim 8, wherein the distance between the plurality of nodes and the search query is calculated. The search processing unit
The plurality of nodes and the search query are subjected to the reference information associated with the list of the representative vectors to which each of the plurality of partial vectors of each node is associated with each of the plurality of nodes. The information processing apparatus according to claim 9, wherein the distance is calculated. The search processing unit
The transposition information in which the representative vector corresponding to each of the plurality of partial vectors of each of the plurality of nodes is arranged in a list for each division position and the correspondence indicating the positions in which each of the plurality of nodes corresponds in the list. The information processing apparatus according to claim 9, wherein the distance between the plurality of nodes and the search query is calculated by using reference information including additional information. The search processing unit
It is characterized in that the distance between the plurality of nodes and the search query is calculated by using one codebook associated with the representative vector to which each of the plurality of partial vectors of each of the plurality of nodes corresponds. The information processing apparatus according to claim 7. The search processing unit
Among the nodes processed in the search process, the second distance that is not vector-quantized is different from the first distance that is the distance that is vector-quantized for a predetermined node. The information processing apparatus according to any one of claims 1 to 12, wherein the information processing apparatus is calculated. The search processing unit
In the search process, a second number of nodes, which is a larger number than the first number of nodes to be extracted as a node in the vicinity of the search query, is extracted as a neighborhood candidate node, and the second distance is calculated for the neighborhood candidate node. The information processing apparatus according to claim 13, wherein the information processing apparatus is calculated. The search processing unit
The information processing apparatus according to claim 14, wherein the first number of nodes are extracted as nodes in the vicinity of the search query from the one having the shorter second distance among the neighborhood candidate nodes. The search processing unit
The information processing apparatus according to any one of claims 13 to 15, wherein the second distance is calculated for a node whose first distance is within a predetermined threshold value. The search processing unit
The information processing apparatus according to any one of claims 13 to 16, wherein the second distance is calculated for a node in a search range indicating a target range to be extracted as a node in the vicinity. The search processing unit
The information processing apparatus according to any one of claims 13 to 17, wherein the second distance is calculated for a node in a search range indicating a target range of the search process. It is an information processing method executed by a computer.
The acquisition process to acquire search queries for multiple objects to be searched for data, and
In a search process for searching for a node in the vicinity of the search query using a graph in which a group of nodes corresponding to each of the plurality of objects is connected by an edge, a plurality of nodes selected by a predetermined criterion and the search query. A search processing step of calculating the distance to and from using the vector information of the plurality of nodes subjected to vector quantization, and
An information processing method characterized by including. The acquisition procedure to acquire the search query for multiple objects to be searched for data, and the acquisition procedure.
In a search process for searching for a node in the vicinity of the search query using a graph in which a group of nodes corresponding to each of the plurality of objects is connected by an edge, a plurality of nodes selected by a predetermined criterion and the search query. A search processing procedure for calculating the distance to and from using the vector information of the plurality of nodes that have been vector-quantified.
An information processing program characterized by having a computer execute.

Images