JP7212728B1 - Information processing device, information processing method and information processing program - Google Patents

Abstract

An information processing device, an information processing method, and an information processing program capable of improving the quality of search results are provided. An information processing apparatus according to the present application includes a classification section, an identification section, and a search section. The classification unit classifies content vectors corresponding to each content to be searched into clusters. The identifying unit identifies clusters similar to the query vector corresponding to the search query input by the user. The search unit searches for content corresponding to the content vector included in the specified cluster as a search target. [Selection drawing] Fig. 5

Description

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

2. Description of the Related Art Conventionally, there is an information processing apparatus that vectorizes a query input by a user and content to be searched, and performs search processing based on the degree of similarity between the two vectors. For this type of information processing apparatus, a technique has been proposed for narrowing down the search range of a search target by an approximate nearest neighbor search (ANN).

JP 2021-86573 A

However, the conventional technology has room for further improvement in terms of improving the quality of search results.

The present application has been made in view of the above, and an object thereof is to provide an information processing apparatus, an information processing method, and an information processing program capable of improving the quality of search results.

An information processing apparatus according to the present application includes a classification section, an identification section, and a search section. The classification unit classifies content vectors corresponding to each content to be searched into clusters. The identifying unit identifies the clusters similar to a query vector corresponding to a search query input by a user. The search unit searches for content corresponding to the content vector included in the specified cluster as a search target.

According to one aspect of the embodiment, it is possible to improve the quality of search results.

FIG. 1 is a diagram showing information processing 1 according to the embodiment. FIG. 2 is a diagram illustrating information processing 2 according to the embodiment. FIG. 3 is a diagram illustrating information processing No. 3 according to the embodiment. FIG. 4 is a diagram illustrating a configuration example of an information processing system according to the embodiment; FIG. 5 is a diagram illustrating a configuration example of an information processing apparatus according to the embodiment; FIG. 6 is a diagram showing an example of content information. FIG. 7 is a diagram showing an example of user information. FIG. 8 is a flowchart illustrating a processing procedure of information processing 1 executed by the information processing apparatus according to the embodiment. FIG. 9 is a flowchart illustrating a processing procedure of information processing No. 2 executed by the information processing apparatus according to the embodiment. FIG. 10 is a flowchart illustrating a processing procedure of information processing No. 3 executed by the information processing apparatus according to the embodiment. FIG. 11 is a diagram illustrating an example of a hardware configuration;

Hereinafter, modes for implementing an information processing apparatus, an information processing method, and an information processing program according to the present application (hereinafter referred to as "embodiments") will be described in detail with reference to the drawings. The information processing apparatus, information processing method, and information processing program according to the present application are not limited to this embodiment. Also, in each of the following embodiments, the same parts are denoted by the same reference numerals, and overlapping descriptions are omitted.

(embodiment)
First, information processing executed by the information processing apparatus according to the embodiment will be described with reference to FIGS. 1 to 3. FIG. 1 to 3 are diagrams showing information processing 1 to 3 according to the embodiment. 1 to 3 show an operation example of an information processing system S including the information processing device 1 according to the embodiment. As shown in FIGS. 1 to 3, an information processing system S according to the embodiment includes an information processing device 1 and a user terminal 50. FIG.

First, information processing 1 will be described with reference to FIG. In the information processing 1 shown in FIG. 1, the search query input by the user and the filter condition specifying the search range of the search target are each vectorized, and the vector of the search query is corrected based on the filter condition vector. After that, search processing is performed.

Specifically, first, the user terminal 50 according to the embodiment transmits a search query input by the user and a filter condition specifying a search range to be searched (step S1). A search query is a text, an image, or the like. The filter condition is a condition for specifying a search target category (product type, price range, etc.), and is specified by, for example, the user selecting from preset filter conditions. Search targets are various contents such as products, documents, images, etc. on the shopping site.

Subsequently, the information processing apparatus 1 according to the embodiment converts the acquired search query and filter condition into a vector in a predetermined distance space (step S2). In the following description, a vector corresponding to a search query may be referred to as a "query vector", and a vector corresponding to a filter condition may be referred to as a "filter vector". Further, the predetermined metric space is a cosine similarity space, an inner product space, a Euclidean metric space, or the like.

Subsequently, the information processing device 1 according to the embodiment generates a correction vector by correcting the query vector based on the filter vector (step S3). For example, the information processing device 1 generates a correction vector by correcting the orientation and norm of the query vector based on the orientation and norm of the filter vector.

In the example shown in FIG. 1, the information processing device 1 generates a correction vector in which the orientation and norm of the query vector are brought closer to the orientation and norm of the filter vector. Various variations are conceivable for the method of generating the correction vector, and the details of this point will be described later.

Subsequently, the information processing device 1 searches for a search target based on the generated correction vector (step S4). For example, the information processing device 1 calculates the distance (cosine similarity, inner product distance, Euclidean distance) between the correction vector and the search target vector, and generates search results based on the calculation results. Then, the information processing device 1 transmits the search result to the user terminal 50 (step S5).

As described above, according to the information processing 1 according to the embodiment, since the query vector itself is corrected using the filter vector, search targets that are close to the query vector, which is a conventional method, are extracted and then applied to the filter condition. Compared to the case of selecting matching search targets, the search results contain more content, and the filter conditions are already taken into consideration for each search target extracted by the correction vector. That is, according to the first information processing according to the embodiment, it is possible to improve the quality of search results.

Next, the second information processing will be described with reference to FIG. In information processing No. 2 shown in FIG. 2, search processing is performed after content is set as content to be subjected to inner product calculation for which the norm difference between the query vector and the vector of the content to be searched (content vector) is equal to or greater than a predetermined value. In the second information processing shown in FIG. 2, the predetermined metric space described above is assumed to be an inner product space.

Specifically, first, the user terminal 50 according to the embodiment transmits a search query input by the user (step S11). Note that the user terminal 50 may transmit the filter conditions described above in addition to the search query.

Subsequently, the information processing apparatus 1 according to the embodiment converts the search query and each content to be searched into vectors in the inner product space (step S12). In addition, hereinafter, the vector corresponding to the search query may be referred to as the "query vector", and the vector corresponding to each content may be referred to as the "content vector". This vector conversion process is realized by converting the inner product of the search query and each content vector into a vector in the same space (same number of dimensions) that makes sense as a score, and the vector is the value of the inner product. It is converted by a conversion procedure acquired by machine learning based on the magnitude. A search procedure using such an inner product space includes a procedure in which content with a large inner product value is more likely to remain in search results.

FIG. 2 shows an example in which the content vector is converted at the same timing as the query vector. may

Subsequently, the information processing apparatus 1 according to the embodiment extracts content vectors whose norm is equal to or greater than a predetermined value from each content vector, and compares the content corresponding to the extracted content vector with the query vector corresponding to the search query. The content is set as the target content for the inner product calculation (step S13).

In the example shown in FIG. 2, contents belonging to areas R2 and R3 are set as target contents. Also, the region R1 is a group of contents similar to the query vector (an inner product satisfies a predetermined condition) in step S14 described later, in other words, a target content for the inner product calculation.

Subsequently, the information processing apparatus 1 according to the embodiment performs search processing with target contents included in the regions R1 to R3 as search targets (step S14). For example, the information processing device 1 calculates the inner product of the content vector of the target content and the query vector, and generates search results based on the calculation results.

Subsequently, the information processing device 1 according to the embodiment transmits the generated search result to the user terminal 50 (step S15).

That is, in the information processing 2 according to the embodiment, the contents of the regions R2 and R3 whose norms are equal to or greater than a predetermined value are always included in the search target regardless of the query vector. This is because, in the inner product space, the similarity between the content vector and the query vector whose norm is equal to or greater than a predetermined value may be high (the search result is close to what the user desires). That is, according to the information processing device 1 according to the embodiment, it is possible to improve the quality of search results.

Next, information processing 3 will be described with reference to FIG. In the information processing 3 shown in FIG. 3, content vectors corresponding to each content to be searched are clustered in advance, and a cluster similar to the query vector is specified to perform search processing.

Specifically, first, the information processing apparatus 1 according to the embodiment classifies each content vector in advance (before acquiring a search query) so that similar content vectors are in the same cluster (step S21 ). In the example shown in FIG. 3, each of regions R1 to R9 is classified as a cluster.

Subsequently, the user terminal 50 according to the embodiment transmits the search query input by the user (step S22). Note that the user terminal 50 may transmit the filter conditions described above in addition to the search query.

Subsequently, the information processing device 1 according to the embodiment converts the search query into a query vector in a predetermined distance space (step S23).

Subsequently, the information processing device 1 according to the embodiment identifies clusters similar to the query vector (step S24). For example, the information processing device 1 calculates a representative vector of the content vectors included in the cluster, and specifies based on the degree of similarity between the representative vector and the query vector.

Subsequently, the information processing apparatus 1 according to the embodiment performs search processing with content included in the specified cluster as search targets (step S25). For example, the information processing device 1 calculates the inner product of the content vector included in the paired cluster and the query vector, and generates the search result based on the calculation result.

Subsequently, the information processing device 1 according to the embodiment transmits the generated search result to the user terminal 50 (step S26).

In other words, in the third information processing according to the embodiment, content vectors are clustered in advance, and by specifying clusters similar to the query vector, all content is scored (for example, inner product calculation) while narrowing down the search results. ) to generate similar search results. That is, according to the information processing device 1 according to the embodiment, it is possible to improve the quality of search results.

Next, a configuration example of the information processing system S according to the embodiment will be described with reference to FIG. FIG. 4 is a block diagram showing a configuration example of the information processing system S according to the embodiment. As shown in FIG. 4, in an information processing system S according to the embodiment, an information processing device 1 and a plurality of user terminals 50 are connected to a network N by wire or wirelessly. The network N is, for example, the Internet, a WAN (Wide Area Network), a LAN (Local Area Network), or the like.

The information processing device 1 is a search server that provides search results for various contents, and executes an information processing method. The search targets searched by the information processing apparatus 1 include, for example, products on shopping sites, documents, images, and the like.

The user terminal 50 is a terminal device possessed by a user who wants to search for content, and is a terminal device that issues a search request to the information processing apparatus 1 . The user terminal 50 can be any type of terminal device such as a smart phone, desktop PC, notebook PC, tablet PC, or the like.

Note that FIG. 4 shows a configuration example of the information processing system S including the information processing device 1 and the user terminal 50, but in addition to these configurations, a terminal device such as a business operator who registers content may be included in the information processing system S. may be included.

Next, a configuration example of the information processing apparatus 1 will be described with reference to FIG.

FIG. 5 is a diagram showing a configuration example of the information processing device 1 according to the embodiment. As shown in FIG. 5 , the information processing device 1 has a communication section 2 , a control section 3 and a storage section 4 . The control unit 3 includes an acquisition unit 31 , a conversion unit 32 , a generation unit 33 , a classification unit 34 , an identification unit 35 , a setting unit 36 , a search unit 37 and an output unit 38 . The storage unit 4 stores content information 41 and user information 42 .

The communication unit 2 is realized by, for example, a NIC (Network Interface Card) or the like. The communication unit 2 is connected to a network by wire or wirelessly.

The control unit 3 is a controller, and for example, various programs (information processing (corresponding to an example of a program) is executed using a RAM or the like as a work area. In addition, the control unit 3 is a controller, and may be implemented by an integrated circuit such as an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a GPGPU (General Purpose Graphic Processing Unit). .

The storage unit 4 is realized by, for example, a semiconductor memory device such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disc.

The content information 41 is information about each content to be searched. FIG. 6 is a diagram showing an example of the content information 41. As shown in FIG. As shown in FIG. 6, the content information 41 includes items such as "content ID", "category", and "detailed information".

"Content ID" is identification information for identifying each content. "Category" is information indicating the category of content, for example, information on the type of content (product type or document type), price range, and information mode such as character string or image. The “detailed information” is detailed information about the content, such as product descriptions, product images, document texts, and the like.

The user information 42 is information about users. FIG. 7 is a diagram showing an example of the user information 42. As shown in FIG. As shown in FIG. 7, the user information 42 includes items such as "user ID", "attribute information", and "behavior information".

“User ID” is identification information that identifies a user. The “attribute information” is information about user attributes, including psychographic attributes, demographic attributes, and the like. “Behavior information” is information about user behavior, and includes, for example, search behavior and purchase behavior.

Next, each function of the control unit 3 of the information processing device 1 (acquisition unit 31, conversion unit 32, generation unit 33, classification unit 34, identification unit 35, setting unit 36, search unit 37 and output unit 38) will be described. .

The acquisition unit 31 acquires various types of information. The acquisition unit 31 acquires search queries and filter conditions from the user terminal 50 . Note that the acquisition unit 31 may acquire past search queries and filter conditions from a server that collects search logs. The acquisition unit 31 may also generate search queries and filter conditions that are expected to be input by the user.

The conversion unit 32 converts the search query and filter condition acquired by the acquisition unit 31 into a query vector and a filter vector, respectively. The conversion unit 32 also refers to the content information 41 and converts the content to be searched into a content vector.

As a vector conversion method, for example, a conversion method using a correspondence table for converting a token into a vector, such as word2vec, or a known conversion method such as Fasttext, BERT (Bidirectional Encoder Representations from Transformers), etc. can be used. .

The predetermined metric space in which vectors are arranged is a metric space of cosine similarity, an inner product space, a Euclidean metric space, or the like. In the information processing 2 shown in FIG. 2, the vector conversion is performed using the inner product space among the predetermined metric spaces.

Also, the filter vector may be converted using table information that holds the vector for each filter condition. Alternatively, the filter vector may be transformed by the transformation technique described above for the string of filter conditions.

The generation unit 33 generates a correction vector by correcting the query vector based on the filter vector converted by the conversion unit 32 . For example, the generation unit 33 generates the output of a function including the filter vector and the query vector as variables as the correction vector.

Specifically, such a function is a function that multiplies each of the filter vector and the query vector by a predetermined weight and sums them up. This weight may be, for example, a fixed value or a variable value.

A weight value as a variable value is determined, for example, according to the number of search targets that match the filter conditions. That is, the greater the number of search targets that match the filter condition, the greater the weight value to be multiplied by the filter vector, and the less the number of search targets, the smaller the weight value to be multiplied by the filter vector. As a result, when the search process is performed using the correction vector, it is possible to avoid, with high accuracy, a decrease in the number of contents included in the search results.

Further, when there are a plurality of filter conditions, the generation unit 33 may, for example, independently correct the query vector using a plurality of filter vectors corresponding to each filter condition. Alternatively, the generation unit 33 may calculate a representative vector (for example, the center of gravity) of a plurality of filter vectors, and correct the query vector using the representative vector.

Further, the generation unit 33 may generate the correction vector by adding the user information 42 . For example, the generation unit 33 may convert user attribute information and behavior information into a user vector, and correct the query vector based on the user vector and the filter vector.

Moreover, the generation unit 33 may generate the correction vector using user information of a user other than the user who has input the search query. For other users, other users with similar attribute information and behavior information are extracted.

Further, the generation unit 33 may estimate the attribute of the search query, convert it into an attribute vector, and generate the correction vector based on the attribute vector.

The classification unit 34 classifies the content vectors into clusters. For example, the classification unit 34 classifies contents vectors whose orientation and norm are within a predetermined range into the same cluster. Alternatively, the classification unit 34 divides the direction and norm of the vector into regions at predetermined intervals, and classifies the content vectors included in each region into the same cluster.

The identifying unit 35 identifies clusters similar to the query vector from among the clusters classified by the classifying unit 34 . For example, the identifying unit 35 calculates a representative vector of the content vectors included in the cluster, and identifies the cluster based on the similarity based on the distance between the representative vector and the query vector.

Note that the center of gravity of the content vectors included in the cluster or the average value can be used as the representative vector.

Further, for example, when the obtaining unit 31 obtains a filter condition, the identifying unit 35 may extract a cluster including a content vector that matches the filter condition, and perform the above-described identifying process on the extracted cluster. .

Further, the identifying unit 35 may identify the cluster when the norm or direction of the representative vector of the cluster is similar to the query vector.

The setting unit 36 extracts a content vector whose norm is equal to or greater than a predetermined value from the content vectors corresponding to each content, and selects the content corresponding to the extracted content vector as the target of the inner product calculation with the query vector corresponding to the search query. Set as content.

For example, the setting unit 36 sets, as the target content, the content whose norm difference is equal to or greater than a predetermined value and whose content vector norm is equal to or greater than a predetermined value. In other words, the setting unit 36 sets the content of the content vector whose norm is extremely longer than the query vector as the target content. Note that the setting unit 36 may set, as the target content, the content of the content vector whose norm is extremely shorter than the query vector.

Moreover, the setting unit 36 may extract, as target content, content whose norm is in the top few percent of all search targets, without being limited to the case where the norm is equal to or greater than a predetermined value.

In addition, the setting unit 36 clusters the content vectors for each norm, and when the norm difference between the representative vector of the cluster and the query vector is equal to or greater than a predetermined value, sets the content included in the cluster as the target content. good too.

In addition to the norm difference, the setting unit 36 may cluster the content vector for each content category, and set the content of the cluster that matches the category of the query vector as the target content.

Also, the setting unit 36 sets the content included in the cluster similar to the query vector as the target content. Specifically, when the norm and orientation of the cluster representative vector are similar to the norm and orientation of the query vector, the setting unit 36 sets the content as target content.

The search unit 37 calculates the distance between the target content set by the setting unit 36 and the query vector in a predetermined distance space. For example, the search unit 37 calculates the inner product of the query vector and the content vector of the target content.

Then, the search unit 37 generates a search result based on the degree of similarity according to the inner product. For example, the search unit 37 generates search results in which the higher the degree of similarity, the higher the ranking.

The output unit 38 outputs search results generated by the search unit 37 to the user terminal 50 .

Next, processing procedures of information processing 1 to 3 executed by the information processing apparatus 1 according to the embodiment will be described with reference to FIGS. 8 to 10. FIG. 8 to 10 are flow charts showing processing procedures of information processing 1 to 3 executed by the information processing apparatus 1 according to the embodiment.

First, the processing procedure of information processing 1 will be described with reference to FIG.

As shown in FIG. 8, the control unit 3 first acquires a search query and filter conditions from the user terminal 50 (step S101). Subsequently, the control unit 3 converts the search query and filter conditions into query vectors and filter vectors, respectively (step S102). Subsequently, the control unit 3 generates a correction vector by correcting the query vector based on the filter vector (step S103). Subsequently, the control unit 3 searches for a search target based on the correction vector (step S104). Subsequently, the control unit 3 outputs the search result to the user terminal 50 (step S105), and terminates the process.

Next, the processing procedure of the second information processing will be described with reference to FIG.

As shown in FIG. 9, the control unit 3 first acquires a search query from the user terminal 50 (step S201). Subsequently, the control unit 3 converts them into query vectors and content vectors in the inner product space (step S202). Subsequently, the control unit 3 sets content whose norm difference from the query vector is equal to or greater than a predetermined value as target content (step S204). Subsequently, the control unit 3 performs inner product calculation of the content vector of the target content and the query vector (step S205). Subsequently, the control unit 3 generates search results based on the calculation results (step S206). Subsequently, the control unit 3 outputs the search result to the user terminal 50 (step S207), and terminates the process.

Next, the processing procedure of information processing 3 will be described with reference to FIG.

As shown in FIG. 10, the control unit 3 first classifies content vectors into clusters (step S301). Subsequently, the control unit 3 acquires a search query and converts it into a query vector (step S302). Subsequently, the control unit 3 identifies a cluster similar to the query vector (step S303). Subsequently, the control unit 3 sets the content included in the identified cluster as target content (step S304). Subsequently, the control unit 3 performs inner product calculation of the content vector of the target content and the query vector (step S305). Subsequently, the control unit 3 generates search results based on the calculation results (step S306). Subsequently, the control unit 3 outputs the search result to the user terminal 50 (step S307), and ends the process.

〔others〕
Also, among the processes described in the above embodiments, some of the processes described as being automatically performed can also be performed manually. Alternatively, all or part of the processes described as being performed manually can be performed automatically by known methods. 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 drawing is not limited to the illustrated information.

Also, each component of each device illustrated is functionally conceptual, and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution and integration of each device is not limited to the one shown in the figure, and all or part of them can be functionally or physically distributed and integrated in arbitrary units according to various loads and usage conditions. Can be integrated and configured.

For example, part or all of the storage unit 4 shown in FIG. 5 may be held in a storage server or the like instead of being held by each device. In this case, each device acquires various information by accessing the storage server.

[Hardware configuration]
Further, the information processing apparatus 1 according to the embodiment described above is implemented by a computer 1000 configured as shown in FIG. 11, for example. FIG. 11 is a diagram illustrating an example of a hardware configuration; A computer 1000 is connected to an output device 1010 and an input device 1020, and an arithmetic device 1030, a primary storage device 1040, a secondary storage device 1050, an output IF (Interface) 1060, an input IF 1070, and a network IF 1080 are connected via a bus 1090. have

The arithmetic device 1030 operates based on programs stored in the primary storage device 1040 and the secondary storage device 1050, programs read from the input device 1020, and the like, and executes various processes. The primary storage device 1040 is a memory device such as a RAM that temporarily stores data used by the arithmetic device 1030 for various calculations. The secondary storage device 1050 is a storage device in which data used for various calculations by the arithmetic device 1030 and various databases are registered. It is realized by

The output IF 1060 is an interface for transmitting information to be output to the output device 1010 that outputs various types of information such as a monitor and a printer. It is realized by a connector conforming to a standard such as HDMI (registered trademark) (High Definition Multimedia Interface). Also, the input IF 1070 is an interface for receiving information from various input devices 1020 such as a mouse, keyboard, scanner, etc., and is realized by, for example, USB.

Note that the input device 1020 includes, for example, optical recording media such as CDs (Compact Discs), DVDs (Digital Versatile Discs), PDs (Phase change rewritable discs), magneto-optical recording media such as MOs (Magneto-Optical discs), and tapes. It may be a device that reads information from a medium, a magnetic recording medium, a semiconductor memory, or the like. Also, the input device 1020 may be an external storage medium such as a USB memory.

Network IF 1080 receives data from other devices via network N and sends the data to arithmetic device 1030, and also transmits data generated by arithmetic device 1030 via network N to other devices.

The arithmetic device 1030 controls the output device 1010 and the input device 1020 via the output IF 1060 and the input IF 1070 . For example, arithmetic device 1030 loads a program from input device 1020 or secondary storage device 1050 onto primary storage device 1040 and executes the loaded program.

For example, when the computer 1000 functions as the information processing device 1 , the arithmetic device 1030 of the computer 1000 implements the functions of the control unit 3 by executing programs loaded on the primary storage device 1040 .

〔effect〕
As described above, the information processing device 1 according to the embodiment includes the conversion unit 32, the generation unit 33, and the search unit 37. The conversion unit 32 converts a search query input by the user into a query vector, and converts a filter condition specifying a search range of a search target into a filter vector. The generation unit 33 generates a correction vector by correcting the query vector based on the filter vector. The search unit 37 searches for a search target based on the correction vector. The generation unit 33 generates, as a correction vector, the output of a function including the filter vector and the query vector as variables. The generation unit 33 generates, as a correction vector, the output of a function that multiplies and adds weights to the variables of the filter vector and the query vector. A weight is set according to the number of search targets that match the filter condition. The generation unit 33 generates the center of gravity of the filter vector and the query vector as a correction vector. The generation unit 33 generates a correction vector based on the user vector obtained by vectorizing the user information about the user and the filter vector. User information includes at least one of user attribute information and behavior information. The generation unit 33 generates a correction vector based on a vector obtained by vectorizing user information of other users similar to the user. The conversion unit 32 converts into a plurality of filter vectors respectively corresponding to a plurality of filter conditions. The generator 33 generates correction vectors based on the plurality of filter vectors. Such a configuration can improve the quality of search results.

Further, as described above, the information processing apparatus 1 according to the embodiment includes the conversion section 32 and the setting section 36 . The conversion unit 32 converts the search query input by the user and each content to be searched into vectors in the inner product space. The setting unit 36 extracts a content vector whose norm is equal to or greater than a predetermined value from the content vectors corresponding to each content, and selects the content corresponding to the extracted content vector as the target of the inner product calculation with the query vector corresponding to the search query. Set as content. The setting unit 36 clusters the content vectors according to the norm, and sets the target content based on the norm difference between the representative vector of the content vectors included in the cluster and the query vector. The representative vector is the centroid of the content vectors included in the cluster. The setting unit 36 sets, as the target content, the content whose norm difference is equal to or greater than a predetermined value and whose content vector norm is equal to or greater than a predetermined value. The setting unit 36 sets, as the target content, the content whose norm difference is equal to or greater than a predetermined value and whose content vector norm is up to a predetermined percentage from the top. The setting unit 36 classifies the content vectors into clusters for each content category, and sets target content from clusters that match the category of the search query. Such a configuration can improve the quality of search results.

Further, as described above, the information processing apparatus 1 according to the embodiment includes the classification section 34 , the identification section 35 and the search section 37 . The classification unit 34 classifies the content vectors corresponding to each content to be searched into clusters. The identifying unit 35 identifies clusters similar to the query vector corresponding to the search query input by the user. The search unit 37 searches for content corresponding to the content vector included in the specified cluster as a search target. The identifying unit 35 identifies a cluster in which the representative vector of the content vectors included in the cluster is similar to the query vector. The representative vector is the centroid of the content vectors included in the cluster. The identifying unit 35 extracts clusters that match a filter condition specifying a search range of a search target, and identifies clusters that are similar to the query vector from among the extracted clusters. The identifying unit 35 identifies clusters having similar query vector directions and norms. Such a configuration can improve the quality of search results.

As described above, some of the embodiments of the present application have been described in detail based on the drawings. It is possible to carry out the invention in other forms with modifications.

〔others〕
Further, among the processes described in the above embodiments, all or part of the processes described as being automatically performed can be manually performed, or the processes described as being performed manually can be performed manually. All or part of this can also be done automatically by known methods. 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 drawing is not limited to the illustrated information.

Also, each component of each device illustrated is functionally conceptual, and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution and integration of each device is not limited to the one shown in the figure, and all or part of them can be functionally or physically distributed and integrated in arbitrary units according to various loads and usage conditions. Can be integrated and configured.

Further, each processing described in the above-described embodiments can be appropriately combined within a range that does not contradict the content of the processing.

Also, the "section, module, unit" described above can be read as "means" or "circuit". For example, the control unit 3 can be read as control means or a control circuit.

1 information processing device 2 communication unit 3 control unit 4 storage unit 31 acquisition unit 32 conversion unit 33 generation unit 34 classification unit 35 identification unit 36 setting unit 37 search unit 38 output unit 41 content information 42 user information 50 user terminal N network S information processing system

Claims (8)

a conversion unit that converts a search query input by a user and each content to be searched into vectors in an inner product distance space;
a classification unit that classifies the content vectors corresponding to the respective contents into clusters;
an identifying unit that identifies the clusters similar to the query vector corresponding to the search query;
a search unit that searches for content corresponding to the content vector included in the identified cluster by calculating an inner product of the content vector and the query vector;
An information processing apparatus, comprising: a setting unit that sets, as content for inner product calculation in the search unit, content whose norm difference between the query vector and the content vector is equal to or greater than a predetermined value. The identification unit
The information processing apparatus according to claim 1, wherein the cluster in which a representative vector of the content vectors included in the cluster is similar to the query vector is specified. The representative vector is
3. The information processing apparatus according to claim 2, wherein the center of gravity of said content vector included in said cluster. The identification unit
4. The method according to any one of claims 1 to 3, wherein the clusters that match a filter condition specifying the search range of the search target are extracted, and the clusters that are similar to the query vector are specified from among the extracted clusters. The information processing apparatus according to any one of the above. The identification unit
The information processing apparatus according to any one of claims 1 to 4, wherein said clusters having similar orientations and norms of said query vectors are identified. a conversion unit that converts the search query into a query vector and a filter condition specifying the search range of the search target into a filter vector;
a generation unit that generates a correction vector that corrects the query vector based on the filter vector;
further comprising
The identification unit
The information processing apparatus according to any one of claims 1 to 5, wherein the cluster similar to the correction vector is specified. A computer-executed information processing method comprising:
a conversion step of converting a search query input by a user and each content to be searched into vectors in an inner product distance space;
a classification step of classifying content vectors corresponding to the respective contents into clusters;
an identifying step of identifying the clusters similar to a query vector corresponding to the search query;
a search step of searching content corresponding to the content vector included in the identified cluster as a search target by inner product calculation with the query vector;
and a setting step of setting a content whose norm difference between the query vector and the content vector is equal to or greater than a predetermined value as content to be subjected to inner product calculation in the search step. a conversion procedure for converting a search query input by a user and each content to be searched into vectors in an inner product distance space;
a classification procedure for classifying content vectors corresponding to the respective contents into clusters;
an identifying step of identifying the clusters similar to the query vector corresponding to the search query;
a search procedure for searching content corresponding to the content vector included in the identified cluster as a search target by inner product calculation with the query vector;
An information processing program for causing a computer to execute: a setting procedure for setting content having a norm difference between the query vector and the content vector equal to or greater than a predetermined value as target content for inner product calculation in the search procedure.

Images