JP7088795B2 - Information processing equipment, information processing methods, and programs - Google Patents

Description

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

In the field of natural language processing, distributed expressions of words and sentences obtained by using neural networks are used for machine learning and the like. On the other hand, a technique for deriving the degree of relevance of a plurality of queries to each other as a score is known (see, for example, Patent Document 1).

Japanese Unexamined Patent Publication No. 2017-228114

However, in the conventional technique, the distributed expression of a query input as a word or a sentence may not sufficiently express the characteristics of the query, and the accuracy of the distributed expression may not be sufficient. As a result, the accuracy of language processing using the distributed representation of queries may not be sufficient.

The present invention has been made in view of the above problems, and an object of the present invention is to provide an information processing apparatus, an information processing method, and a program capable of improving the accuracy of distributed representation of queries.

One aspect of the present invention includes an acquisition unit that acquires a plurality of queries input for searching information, a generation unit that generates a distributed representation of each of the plurality of queries acquired by the acquisition unit, and the plurality of units. It is an information processing apparatus including a selection unit for selecting a distributed expression related to each other from a plurality of distributed expressions generated by the generation unit based on the degree of duplication of a user who has input each of the queries.

According to one aspect of the present invention, the accuracy of the distributed representation of the query can be improved.

It is a figure which shows an example of the information processing system 1 including the information processing apparatus 100 of 1st Embodiment. It is a figure which shows an example of the structure of the information processing apparatus 100 in 1st Embodiment. It is a flowchart which shows the flow of a series of processing of the control unit 110 in 1st Embodiment. It is a figure which shows how to select the word vector distributed in the vicinity of the reference word vector schematically. It is a figure which shows how to derive the score of the degree of association with the query which is the generator of a reference word vector for each query which is the generator of each of N word vectors. It is a figure which shows typically the state when the process of selecting a word vector related to a reference word vector is repeated k a predetermined number of times. It is a figure which showed an example of the correct answer rate of this method. It is a figure which shows an example of the hardware composition of the information processing apparatus 100 of an embodiment.

Hereinafter, an information processing apparatus, an information processing method, and a program to which the present invention is applied will be described with reference to the drawings.

[Overview]
The information processing device is realized by one or more processors. The information processing device acquires a plurality of queries input for retrieving information, and generates a distributed representation of each of the acquired plurality of queries. The distributed representation of a query is information that represents the query by a plurality of features, and is represented by, for example, a multidimensional vector having the plurality of features as elements.

When the information processing device generates a distributed representation of a plurality of queries, the information processing device can generate a distributed representation from the plurality of distributed representations based on a predetermined index value for a user who has input each of the plurality of queries that are the generation sources of the plurality of distributed representations. Select distributed representations that are related to each other. The predetermined index value for the user may be, for example, the degree of duplication of the user who entered each of the plurality of queries, or the degree of similarity of the user who entered each of the plurality of queries.

Further, the information processing apparatus selects a distributed expression related to each other from a plurality of distributed expressions in place of the predetermined index value related to the user or in addition to the predetermined index value related to each of the plurality of queries. You may. The predetermined index value for the query may be, for example, the time each of the plurality of queries is entered by the user.

In this way, the information processing device selects queries that are related to each other from the distributed representation of a plurality of queries based on index values such as the degree of duplication of users, the degree of similarity of users, and the input time of queries. The accuracy of the distributed representation of the query can be improved.

<First Embodiment>
[overall structure]
FIG. 1 is a diagram showing an example of an information processing system 1 including the information processing apparatus 100 of the first embodiment. The information processing system 1 in the first embodiment includes, for example, one or more terminal devices 10, an information providing device 20, and an information processing device 100. These devices are connected to each other, for example, via a network NW.

Each device shown in FIG. 1 transmits and receives various information via the network NW. The network NW includes, for example, the Internet, a WAN (Wide Area Network), a LAN (Local Area Network), a provider terminal, a wireless communication network, a wireless base station, a dedicated line, and the like. It should be noted that not all combinations of the devices shown in FIG. 1 need not be able to communicate with each other, and the network NW may include a local network in part.

The terminal device 10 is a terminal device including an input device, a display device, a communication device, a storage device, and an arithmetic unit, such as a mobile phone such as a smartphone, a tablet terminal, and various personal computers. The communication device includes a network card such as a NIC (Network Interface Card), a wireless communication module, and the like. In the terminal device 10, a UA (User Agent) such as a web browser or an application program is activated, and a request according to a user's input content is transmitted to the information providing device 20. Further, the terminal device 10 in which the UA is activated causes the display device to display various images based on the information acquired from the information providing device 20.

The information providing device 20 may be, for example, a web server that provides a web page to the terminal device 10 in response to a request from a web browser (for example, an HTTP (Hypertext Transfer Protocol) request or a query). Web pages contain content. The content may be, for example, document data posted on a blog, a website, or the like, or may be still image data, moving image data, audio data, or the like. Further, the information providing device 20 may be an application server that provides contents such as images and sounds to the terminal device 10 in response to a request from the application program. For example, the information providing device 20 is a user who uses a terminal device 10 via a website or an application to search for a desired content from a certain database (hereinafter referred to as a search service) such as a document search or an image search. To provide to. When a user inputs a query to the terminal device 10 when searching for desired content from a database using a search service, the information providing device 20 receives a query from the terminal device 10 and receives the query. A web page or the like including the content corresponding to the above is transmitted to the terminal device 10.

The information processing device 100 acquires a plurality of queries input when using the search service from the information providing device 20, and generates a distributed representation of each of the acquired plurality of queries. Then, the information processing apparatus 100 selects distributed representations related to each other from the generated plurality of distributed representations, and transmits information regarding the selection result to the information providing device 20. In response to this, the information providing device 20 improves the accuracy of the search service based on, for example, the information regarding the distributed representation received from the information processing device 100.

[Information processing device configuration]
FIG. 2 is a diagram showing an example of the configuration of the information processing apparatus 100 according to the first embodiment. As shown in the figure, the information processing apparatus 100 includes, for example, a communication unit 102, a control unit 110, and a storage unit 130.

The communication unit 102 includes, for example, a communication interface such as a NIC. The communication unit 102 communicates with the information providing device 20 and the like via the network NW.

The control unit 110 includes, for example, an acquisition unit 112, a generation unit 114, a selection unit 116, and an output control unit 118. These components are realized, for example, by a processor (or processor circuit) such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit) executing a program (software) stored in the storage unit 130. .. Further, some or all of the components of the control unit 110 are hardware (circuit unit: circuitry) such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), and the like. ), Or it may be realized by the collaboration of software and hardware. Further, the program referred to by the processor may be stored in the storage unit 130 in advance, or is stored in a removable storage medium such as a DVD or a CD-ROM, and the storage medium is the drive of the information processing apparatus 100. It may be installed in the storage unit 130 from the storage medium by being attached to the device.

The storage unit 130 is realized by, for example, a storage device such as an HDD (Hard Disc Drive), a flash memory, an EEPROM (Electrically Erasable Programmable Read Only Memory), a ROM (Read Only Memory), and a RAM (Random Access Memory). In addition to various programs such as firmware and application programs, the storage unit 130 stores search query information 132, action history information 134, generator information 136, and the like.

The search query information 132 is, for example, information in which an input date and time, identification information of the input user (hereinafter, user ID), and the like are associated with each query input when the search service is used.

The action history information 134 is, for example, information including the action history of each user who used the search service, and the date and time (search date and time) when the user used the search service or the user searches for the user ID of each user. This is information associated with a query entered at times, a URL (Uniform Resource Locator) of a web page viewed by a user, and the like.

Generator information 136 is information (program or data structure) that defines a generator for generating a distributed representation from a query. When a single word (word) representing a query or a phrase (phrase) consisting of a plurality of words is input, the generator is a multidimensional vector (hereinafter referred to as a word vector) having a predetermined number of elements (dimensions). ) Is realized by a neural network trained in advance to output as a distributed representation.

In the generator information 136, for example, how the neurons (units) contained in the input layer, one or more hidden layers (intermediate layers), and the output layers constituting each neural network are connected to each other. It includes connection information and weight information such as how many connection coefficients are given to data input / output between connected neurons. The connection information includes, for example, the number of neurons contained in each layer, information that specifies the type of neuron to which each neuron is connected, the activation function that realizes each neuron, and information such as a gate provided between neurons in the hidden layer. including. The activation function that realizes the neuron may be, for example, a function that switches the operation according to the input code (ReLU function or ELU function), or may be a sigmoid function, a step function, or a hyperbolic tangent function. , May be an equal function. The gate selectively passes or weights the data transmitted between neurons, for example, depending on the value returned by the activation function (eg 1 or 0). The coupling coefficient is a parameter of the activation function, and is a weight given to the output data when data is output from a neuron in one layer to a neuron in a deeper layer, for example, in a hidden layer of a neural network. including. Further, the coupling coefficient may include a bias component peculiar to each layer.

The generator defined by the generator information 136 selects, for example, a reference word or phrase from a plurality of words or phrases contained in the corpus, and from the selected word or phrase, before or after the word or phrase. Skip-Gram that predicts words or phrases that appear in, or Countious Bag that focuses on a certain context contained in the corpus and predicts a certain word or phrase in that focused context from the words or phrases that appear before or after. -It is realized by a neural network such as Words (CBOW). Such neural networks are used in models such as fastText and word2vec. Unlike word2vec, fastText considers the grouping of words or partial words of words (the words and the characters that make up the words) at the time of learning, and the same words or words whose conjugations are different due to different conjugations. Treated as a phrase. The corpus may be, for example, a document included in an encyclopedia that aggregates information transmitted on the Internet, or may be a set of a plurality of queries (so-called query log) included in the search query information 132.

The generator information 136 is, for example, a fastText (hereinafter referred to as a first generator) that generates a word vector from a query by using a document included in an encyclopedia as a corpus, or a corpus of a set of queries and a word from a query. Define multiple generators such as fastText (hereinafter referred to as the second generator) that generates a vector, and word2vec (hereinafter referred to as the third generator) that generates a word vector from a query using a set of queries as a corpus. You can do it.

[Processing flow]
Hereinafter, the flow of a series of processes of the control unit 110 in the first embodiment will be described according to a flowchart. FIG. 3 is a flowchart showing a flow of a series of processes of the control unit 110 in the first embodiment. The processing of this flowchart may be repeated, for example, at a predetermined cycle.

First, the acquisition unit 112 obtains a plurality of queries input from the information providing device 20 when using the search service, the input date and time of each query, the user ID of the user who input each query, and the like via the communication unit 102. Acquired (S100), and these acquired information are stored in the storage unit 130 as search query information 132.

Next, the generator 114 refers to the search query information 132 and the generator information 136 stored in the storage unit 130, and each of the plurality of queries included in the search query information 132 is defined by the generator information 136. By inputting to the generator, the generator is made to generate a word vector for each query (S102).

For example, if the generator information 136 defines three generators, a first generator, a second generator, and a third generator, the generator 114 inputs a query to each of the three generators. By doing so, three types of word vectors different from each other may be generated for each query.

Next, the selection unit 116 selects a reference word vector (hereinafter referred to as a reference word vector) from the plurality of word vectors generated by the generation unit 114 (S104). The reference word vector is an example of the "first distributed representation".

Next, the selection unit 116 selects a predetermined number N of word vectors distributed in the vicinity of the selected reference word vector (S106). The neighborhood means that when each element included in the word vector is used as a basis, the relative distance from the reference word vector is close in the multidimensional space stretched by the basis.

For example, the selection unit 116 derives the similarity between the reference word vector and another word vector (for example, cosine similarity), and converts the derived similarity into a distance to obtain a higher-order predetermined degree close to the reference word vector. Select a word vector of number N. The word vector of the upper predetermined number N close to the reference word vector is an example of the “second distributed representation”.

FIG. 4 is a diagram schematically showing how a word vector distributed in the vicinity of the reference word vector is selected. In the figure, _VR represents a reference word vector. For example, when N = 9, the selection unit 116 selects the upper nine word vectors V ₁ to V ₉ that are close to the reference word vector _VR in the multidimensional space as shown in the illustrated example.

Next, the selection unit 116 derives the degree of relevance between the query that is the generator of each of the selected N word vectors and the query that is the generator of the reference word vector (S108). For example, the selection unit 116 derives the degree of relevance between queries based on the mathematical formula (1).

Score (A, B) in the formula (1) is a score representing the degree of relevance of certain queries A and B to each other. It is assumed that one of the query A and the query B is the query from which the reference word vector is generated. As shown in the formula (1), Score (A, B) is the number of users who input both query A and query B (Auser∩Buser), and the number of users who input query A (Auser). It is derived as a value obtained by dividing the divided value (molecule in the formula) and further dividing the number of users who input the query B (Buser) by the number of all users (ALLuser) (denominator in the formula). To.

Since the score Score (A, B) of the degree of relevance is based on User ∩ Buser, it is also an index value indicating the degree of overlap (multiplicity) between the user who input the query A and the user who input the query B. The greater the multiplicity of users who entered each of the different queries, the higher the score Score (A, B), and the smaller the multiplicity of the users who entered each of the different queries, the higher the score Score (A, B). B) becomes smaller. That is, the greater the degree of duplication of the users who input each of the different queries, the larger the score Score (A, B) (= the degree of relevance between the query A and the query B), and the greater the degree of duplication of the users who input each of the different queries. The smaller the degree of duplication, the smaller the score Score (A, B) (= the degree of relevance between query A and query B).

FIG. 5 is a diagram schematically showing how the score of the degree of association with the query that is the generator of the reference word vector is derived for each query that is the generator of each of the N word vectors. As in the example of FIG. 4 described above, when N = 9, the query that is the generator of each of the nine word vectors V ₁ to V ₉ is the query that is the generator of the reference word vector _VR . The relevance scores S1 to S9 are derived based on the degree of duplication of the user who entered the two queries.

When the selection unit 116 derives the score Score of the degree of association with the query that is the generation source of the reference word vector for each query that is the generation source of each of the N word vectors, the selection unit 116 derives the score Score from the N word vectors. Delete the word vector whose relevance score Score is less than the threshold of the reference word vector query (S110), and relate the word vector whose relevance score Score to the reference word vector query is equal to or higher than the threshold value to the reference word vector. Select as the word vector.

In this way, in addition to the condition that the degree of similarity between word vectors is large, the condition that the degree of duplication of the user who input the query from which the word vector is generated is large (that is, the degree of relevance between the queries is large) is satisfied. By selecting only the word vector, the accuracy of the distributed representation of the query can be improved.

Next, the output control unit 118 controls the communication unit 102, and among the N word vectors selected by the selection unit 116, the word vectors remaining without being deleted based on the relevance of the query, that is, A word vector having a score equal to or higher than a threshold value of the degree of association with the query of the reference word vector is transmitted to the information providing device 20 as a word vector related to the reference word vector (S112).

In addition, the output control unit 118 may transmit the query which is the generation source of the word vector to be transmitted to the information providing device 20 in place of or in addition to the word vector. When the information providing device 20 receives a word vector or a query from the information processing device 100, the information providing device 20 stores the received information in a storage device such as an HDD provided therein. Then, when the information providing device 20 receives a query for search from the terminal device 10 that provides the search service via a web page or the like, the terminal device 10 is among the word vectors or queries received from the information processing device 100. A word vector or query having a high degree of relevance to the query received from is extracted, and the extracted information is transmitted to the terminal device 10 as a query recommending a search or the like. This allows the user to use the search service to search for content of interest or interest.

Next, the control unit 110 determines whether or not the processing of S104 to S112 has been repeated k a predetermined number of times (S114), and if it is determined that the processing has not been repeated k a predetermined number of times, returns the processing to S104 and the previously selected reference word. Another word vector different from the vector is selected as a new reference word vector, and the score Score (A, B) of the degree of association between queries is a threshold value from among the top N word vectors similar to the new reference word vector. The above word vector is selected as the word vector related to the new reference word vector. On the other hand, when the control unit 110 determines that the processes of S104 to S112 have been repeated k for a predetermined number of times, the control unit 110 ends the process of this flowchart.

FIG. 6 is a diagram schematically showing a state when the process of selecting a word vector related to a reference word vector is repeated k a predetermined number of times. As shown in the illustrated example, when the set of word vectors of each of the plurality of queries included in the search query information 132 is used as the population, the control unit 110 is the population in each of the first to kth processes. From among, _Ni word vectors similar to a certain reference word vector V _Ri are selected, and from the _Ni word vectors, the score Score of the degree of association with the query of the reference word vector is the threshold. The remaining ( _Ni − _Di ) word vectors obtained by subtracting the less than _Di word vectors are selected as the word vectors associated with the reference word vector V _Ri . i is a number of 1 or more and k or less. For example, in each process from the first time to the kth time, when the word vector related to the reference word vector V _Ri is selected from the population, the query of the selected word vector under a certain criterion is the reference word vector V. A human or the like may determine whether the meaning is the same as or similar to the _Ri query, and evaluate this method. In the illustrated example, in each process from the first time to the kth time, among the queries of the word vector related to the reference word vector V _Ri , the number of queries having the same meaning as or similar to the query of the reference word vector V _Ri is determined as the reference word. The percentage divided by the total number of word vector queries related to the vector V _Ri is derived as the correct answer rate.

FIG. 7 is a diagram showing an example of the correct answer rate of this method. As described above, this method in the figure means that when selecting a word vector related to a reference word vector from a population, first, the upper N word vectors similar to the reference word vector are selected, and the selection thereof is performed. This is a method of further narrowing down the word vectors based on the score Score (A, B) of the degree of relevance between the queries from the N word vectors. On the other hand, in the comparison method, when selecting a word vector related to a reference word vector from the population, first, the upper N word vectors similar to the reference word vector are selected, and then the selected N word vectors are selected. Is a method of making a word vector related to a reference word vector.

For example, when a plurality of word vectors as a population are generated by using the first generator, the correct answer rate of the word vector selected as related to the reference word vector is 84 [%] in this method. Yes, it was 76 [%] in the comparison method, and the difference was +8 [%]. Further, when a plurality of word vectors as a population are generated by using the second generator, the correct answer rate of the word vector selected as related to the reference word vector is 66 [%] in this method. Yes, it was 30 [%] in the comparison method, and the difference was +36 [%]. Further, when a plurality of word vectors as a population are generated by using the third generator, the correct answer rate of the word vector selected as related to the reference word vector is 81 [%] in this method. Yes, it was 67 [%] in the comparison method, and the difference was +14 [%]. In this way, this method can select distributed expressions whose meanings are closer to each other than the comparison method.

According to the first embodiment described above, a plurality of queries input for searching information are acquired, a word vector which is a distributed representation of each of the acquired plurality of queries is generated, and each of the plurality of queries is generated. By selecting word vectors that are related to each other from a plurality of generated word vectors based on the degree of duplication of the user who entered, the accuracy of the distributed representation of the query can be improved.

<Second Embodiment>
Hereinafter, the second embodiment will be described. In the first embodiment described above, after selecting the upper N word vectors similar to the reference word vector, the query of the generator of each of the N word vectors and the query of the generator of the reference word vector are obtained. It was described as narrowing down N word vectors based on the degree of duplication of the input user (score Score (A, B) of the degree of relevance between queries).

On the other hand, in the second embodiment, after selecting the upper N word vectors similar to the reference word vector, the user who inputs the query of the generator of each of the N word vectors and the reference word vector. It differs from the above-mentioned first embodiment in that N word vectors are narrowed down based on the similarity with the user who input the query of the generation source of. Hereinafter, the differences from the first embodiment will be mainly described, and the points common to the first embodiment will be omitted. In the description of the second embodiment, the same parts as those of the first embodiment will be described with the same reference numerals.

The generation unit 114 of the second embodiment generates the user vector as the distributed representation of the user who input the query, in the same manner as generating the word vector as the distributed representation of the query. Specifically, the generation unit 114 may refer to the action history information 134 and generate a multidimensional vector having the number of times the content is viewed as an element from the action history of each user as a user vector.

When the user vector of each user is generated by the generation unit 114, the selection unit 116 of the second embodiment selects the upper N word vectors similar to the reference word vector, and then each of the N word vectors. Based on the similarity between the user who entered the query that generated the reference word vector and the user who entered the query that generated the reference word vector, the word vector related to the reference word vector is selected from the N word vectors. select. For example, the selection unit 116 has a cosine similarity between the user vector of the user A who has input the query of the generation source of each of the N word vectors and the user vector of the user B who has input the query of the generation source of the reference word vector. Is derived, and the word vector corresponding to the combination of user vectors whose cosine similarity is equal to or higher than the threshold value is selected, and the word vector related to the reference word vector is selected.

According to the second embodiment described above, a plurality of queries input for searching information are acquired, a word vector which is a distributed representation of each of the acquired plurality of queries is generated, and each of the plurality of queries is generated. By selecting word vectors related to each other from a plurality of generated word vectors based on the similarity of the users who input the above, the accuracy of the distributed representation of the query is improved as in the above-described embodiment. Can be done.

<Third Embodiment>
Hereinafter, the third embodiment will be described. In the first embodiment described above, after selecting the upper N word vectors similar to the reference word vector, the query of the generator of each of the N word vectors and the query of the generator of the reference word vector are obtained. It was described as narrowing down N word vectors based on the degree of duplication of the input user (score Score (A, B) of the degree of relevance between queries). Further, in the second embodiment described above, the user who inputs the query of the generator of each of the N word vectors after selecting the upper N word vectors similar to the reference word vector, and the reference word vector. It was described as narrowing down N word vectors based on the similarity with the user who entered the query of the generation source.

On the other hand, in the third embodiment, after selecting the upper N word vectors similar to the reference word vector, the query of the generator of each of the N word vectors is based on the time input by the user. Therefore, it differs from the first embodiment and the second embodiment described above in that N word vectors are narrowed down. Hereinafter, the differences between the first embodiment and the second embodiment will be mainly described, and the points common to the first embodiment and the second embodiment will be omitted. In the description of the third embodiment, the same parts as those of the first embodiment and the second embodiment will be described with the same reference numerals.

After selecting the upper N word vectors similar to the reference word vector, the selection unit 116 of the third embodiment is based on the time when the query of the generator of each of the N word vectors is input by the user. , Select the word vector related to the reference word vector from the N word vectors. For example, when a user inputs a certain query A and a query B, the selection unit 116 determines that the times when the queries A and B are input are considered to be the same timing, the score Score of the degree of association between the queries Score ( If A and B) are increased and the times when the queries A and B are input can be regarded as different timings, the score Score (A, B) of the degree of association between the queries may be decreased.

Specifically, when the times when the queries A and B are input are included in the same session, the selection unit 116 considers that the queries A and B are input at the same timing, and interlies between the queries. If the relevance score Score (A, B) is increased and the times when queries A and B are entered are included in different sessions, it is considered that the queries A and B are entered at different timings. Therefore, the score Score (A, B) of the degree of association between queries may be reduced.

The session may be, for example, the period from the user viewing a certain content to the conversion. Conversion is, for example, purchasing content such as a product, viewing content such as a video, or browsing content such as an advertisement. In addition, the session may be a period from browsing a certain content to browsing another content, or from browsing a certain content to terminating an application such as a web browser. It may be a period, or it may be a period from browsing a certain content to the elapse of a predetermined time (timeout).

Further, the selection unit 116 increases the score Score (A, B) of the degree of relevance between queries as the period from the input of the query A to the input of the query B becomes shorter, and the query B is input because the query A is input. The longer the period until is input, the smaller the score Score (A, B) of the degree of association between queries may be.

In this way, in order to increase the score Score (A, B) of the degree of relevance between queries as the query input times are the same or closer, the word vector related to the reference word vector is selected from the N word vectors. When selecting, it is possible to remove word vectors that are less relevant to the reference word vector.

According to the third embodiment described above, a plurality of queries input for searching information are acquired, a word vector which is a distributed representation of each of the acquired plurality of queries is generated, and each of the plurality of queries is generated. By selecting word vectors related to each other from the plurality of generated word vectors based on the input time of, the accuracy of the distributed representation of the query can be improved as in the above-described embodiment.

<Fourth Embodiment>
Hereinafter, the fourth embodiment will be described. In the fourth embodiment, the degree of duplication of the user who input the query of the generation source of each of the plurality of word vectors as the population and the query of the generation source of the reference word vector (score Score (A) of the degree of relevance between the queries). , B)), which is different from the above-described first embodiment to the third embodiment in that each of the word vectors is weighted. Hereinafter, the differences from the first embodiment to the third embodiment will be mainly described, and the points common to the first to third embodiments will be omitted. In the description of the fourth embodiment, the same parts as those of the first to third embodiments will be described with the same reference numerals.

The selection unit 116 of the fourth embodiment is based on the degree of duplication of the user who input the query of the generation source of each of the plurality of word vectors as the population and the query of the generation source of the reference word vector, and between the queries. Derivation of the score Score (A, B) of the degree of relevance. Then, the selection unit 116 weights each of the plurality of word vectors according to the score Score (A, B) of the degree of association between the queries.

For example, the selection unit 116 multiplies the word vector by a weighting coefficient having a large value so that the greater the degree of overlap of the users, the greater the degree of similarity to the reference word vector (for example, the degree of cosine similarity). On the other hand, the selection unit 116 multiplies the word vector by a weighting coefficient having a small value so that the smaller the user's multiplicity is, the smaller the similarity with respect to the reference word vector (for example, the cosine similarity).

When each of the plurality of word vectors as the population is weighted, the selection unit 116 selects the upper N word vectors similar to the reference word vector from the weighted plurality of word vectors, and selects the upper N word vectors. N based on the degree of relevance between the query from which each of the N word vectors was generated and the query from which the reference word vector was generated, the similarity with the user who entered those queries, the input time of those queries, and so on. Select the word vector related to the reference word vector from the number of word vectors. By weighting a plurality of word vectors as the population in this way, the reference word vector has more meaning when selecting the upper N word vectors similar to the reference word vector from the population. You can select a word vector that is close to you. As a result, the accuracy of the distributed representation of the query can be improved as in the above-described embodiment.

<Hardware configuration>
The information processing apparatus 100 of the above-described embodiment is realized by, for example, a hardware configuration as shown in FIG. FIG. 8 is a diagram showing an example of the hardware configuration of the information processing apparatus 100 of the embodiment.

The information processing device 100 includes NIC100-1, CPU100-2, RAM100-3, ROM100-4, a secondary storage device 100-5 such as a flash memory or an HDD, and a drive device 100-6 on an internal bus or a dedicated communication line. It is configured to be interconnected by. A portable storage medium such as an optical disk is mounted on the drive device 100-6. A program stored in a portable storage medium mounted on the secondary storage device 100-5 or the drive device 100-6 is expanded in the RAM 100-3 by a DMA controller (not shown) or the like, and executed by the CPU 100-2. As a result, the control unit 110 is realized. The program referred to by the control unit 110 may be downloaded from another device via the network NW.

Although the embodiments for carrying out the present invention have been described above using the embodiments, the present invention is not limited to these embodiments, and various modifications and substitutions are made without departing from the gist of the present invention. Can be added.

1 ... Information processing system, 10 ... Terminal device, 20 ... Information providing device, 100 ... Information processing device, 102 ... Communication unit, 110 ... Control unit, 112 ... Acquisition unit, 114 ... Generation unit, 116 ... Selection unit, 118 ... Output control unit, 130 ... Storage unit

Claims (11)

An acquirer that retrieves multiple queries entered to retrieve information,
A generation unit that generates a distributed representation of each of the plurality of queries acquired by the acquisition unit, and a generation unit.
A selection unit for selecting a distributed expression related to each other from a plurality of distributed expressions generated by the generation unit based on the degree of duplication of the user who entered each of the plurality of queries is provided.
The selection unit is
Weighting each of the plurality of distributed representations generated by the generator based on the user's multiplicity,
The greater the degree of duplication of the users, the larger the weight given to the distributed representation, and the smaller the degree of duplication of the users, the smaller the weight given to the distributed representation.
Select the distributed representations related to each other from the plurality of weighted distributed representations.
Information processing equipment. The selection unit is
From the plurality of distributed representations, a predetermined number of higher-order second distributed representations similar to the reference first distributed representation are selected.
From the selected predetermined number of second distributed representations, the second distributed representation related to the first distributed representation is selected based on the degree of duplication of the user who input the query that is the generator of the second distributed representation. select,
The information processing apparatus according to claim 1. The selection unit uses the second distributed representation generated from a query in which the degree of duplication of the user is equal to or greater than the threshold value from the predetermined number of second distributed representations, and the second distributed representation is related to the first distributed representation. Select as a distributed representation,
The information processing apparatus according to claim 2. The selection unit is
For each of the plurality of distributed representations generated by the generation unit, the greater the degree of duplication of the users, the more similar to the first distributed representation as a reference.
A second distributed representation related to the first distributed representation is selected from the plurality of weighted distributed representations.
The information processing apparatus according to claim 1 . The selection unit further selects distributed expressions related to each other from the plurality of distributed expressions generated by the generating unit based on the similarity of the user who entered each of the plurality of queries.
The information processing apparatus according to any one of claims 1 to 4 . The selection unit further selects distributed expressions related to each other from the plurality of distributed expressions generated by the generating unit based on the time when each of the plurality of queries is input by the user.
The information processing apparatus according to any one of claims 1 to 5 . An acquirer that retrieves multiple queries entered to retrieve information,
A generation unit that generates a distributed representation of each of the plurality of queries acquired by the acquisition unit, and a generation unit.
A selection unit for selecting a distributed expression related to each other from a plurality of distributed expressions generated by the generation unit based on the similarity of the user who entered each of the plurality of queries is provided.
The selection unit is
Based on the degree of duplication of the user who entered each of the plurality of queries, the distributed representations related to each other are further selected from the plurality of distributed representations generated by the generator.
Weighting each of the plurality of distributed representations generated by the generator based on the user's multiplicity,
The greater the degree of duplication of the users, the larger the weight given to the distributed representation, and the smaller the degree of duplication of the users, the smaller the weight given to the distributed representation.
Select the distributed representations related to each other from the plurality of weighted distributed representations.
Information processing equipment. An acquirer that retrieves multiple queries entered to retrieve information,
A generation unit that generates a distributed representation of each of the plurality of queries acquired by the acquisition unit, and a generation unit.
Each of the plurality of queries includes a selection unit that selects a distributed expression related to each other from the plurality of distributed expressions generated by the generation unit based on the time input by the user .
The selection unit is
Based on the degree of duplication of the user who entered each of the plurality of queries, the distributed representations related to each other are further selected from the plurality of distributed representations generated by the generator.
Weighting each of the plurality of distributed representations generated by the generator based on the user's multiplicity,
The greater the degree of duplication of the users, the larger the weight given to the distributed representation, and the smaller the degree of duplication of the users, the smaller the weight given to the distributed representation.
Select the distributed representations related to each other from the plurality of weighted distributed representations.
Information processing equipment. An acquirer that retrieves multiple queries entered to retrieve information,
A generation unit that generates a distributed representation of each of the plurality of queries acquired by the acquisition unit, and a generation unit.
From a plurality of distributed representations generated by the generator based on a predetermined index value for a user who has entered each of the plurality of queries, or a predetermined index value for each of the plurality of queries, they are related to each other. With a selection section to select a distributed representation ,
The selection unit is
Based on the degree of duplication of the user who entered each of the plurality of queries, the distributed representations related to each other are further selected from the plurality of distributed representations generated by the generator.
Weighting each of the plurality of distributed representations generated by the generator based on the user's multiplicity,
The greater the degree of duplication of the users, the larger the weight given to the distributed representation, and the smaller the degree of duplication of the users, the smaller the weight given to the distributed representation.
Select the distributed representations related to each other from the plurality of weighted distributed representations.
Information processing equipment. The computer
Get multiple queries entered to retrieve information,
Generate a distributed representation of each of the multiple queries obtained above
Based on the degree of duplication of the user who entered each of the plurality of queries, the distributed representations related to each other are selected from the plurality of generated distributed representations.
Weighting each of the generated distributed representations based on the user's multiplicity,
The greater the degree of duplication of the users, the larger the weight given to the distributed representation, and the smaller the degree of duplication of the users, the smaller the weight given to the distributed representation.
Select the distributed representations related to each other from the plurality of weighted distributed representations.
Information processing method. On the computer
The process of retrieving multiple queries entered to retrieve information,
The process of generating the distributed representation of each of the multiple queries obtained above,
A process of selecting a distributed representation related to each other from the generated distributed representations based on the degree of duplication of the user who entered each of the plurality of queries.
A process of weighting each of the generated plurality of distributed expressions based on the degree of duplication of the user.
The larger the degree of duplication of the users, the larger the weight given to the distributed expression, and the smaller the degree of duplication of the users, the smaller the weight given to the distributed expression.
The process of selecting the distributed representations related to each other from the plurality of weighted distributed representations,
A program to execute.

Images