JP2020129306A - Information processing device, information processing method, and information processing program - Google Patents

Abstract

To allow for efficiently interpreting a meaning of information.SOLUTION: An information processing device comprises a prediction unit and a storage unit. The prediction unit predicts first feature information indicating features of first input information from the first input information, assuming that a plurality of queries entered by a same user in a prescribed time should have similar features, using a learning model having learned the features that the plurality of queries has. The storage unit stores progress information, which is information on an interim progress of first feature information prediction processing by the prediction unit, as information used for prediction processing by the prediction unit to predict second feature information indicating second input information from the second input information.SELECTED DRAWING: Figure 1

Description

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

Conventionally, a technique of interpreting the meaning of information such as a search query is known. For example, the information regarding the search query used by the target user is classified into the first cluster based on the similarity between the vectors corresponding to the information regarding each search query used by the target user, and is classified into the information regarding each search query of other users. Information about search queries used by other users is classified into the second cluster based on the similarity between corresponding vectors. Then, based on the difference between the first cluster and the second cluster, a technique has been proposed for extracting a characteristic cluster, which is a cluster showing a characteristic behavior of the target user, from the first cluster.

Japanese Patent Laid-Open No. 2018-60469

However, the above-mentioned conventional techniques do not always make it possible to efficiently interpret the meaning of information such as a search query. For example, in the above-mentioned conventional technique, the information about the search query is only classified into clusters based on the similarity between the vectors corresponding to the information about each search query, and the meaning of the information such as the search query can be efficiently interpreted. You can't always do it. In addition, conventionally, it is a problem to efficiently interpret meanings of all kinds of information, not limited to search queries. Therefore, in the above-mentioned conventional technology, it is not always possible to efficiently interpret the meaning of information.

The present application has been made in view of the above, and an object thereof is to provide an information processing device, an information processing method, and an information processing program capable of efficiently interpreting the meaning of information.

The information processing apparatus according to the present application uses a learning model that learns the characteristics of the plurality of search queries, assuming that the plurality of search queries input by the same user within a predetermined time have similar characteristics, A predicting unit that predicts first characteristic information indicating a characteristic of the first input information from first input information, and a predicting unit predicts second characteristic information indicating a characteristic of the second input information from second input information. As information used for the prediction process, a storage unit that stores progress information that is information regarding the progress of the prediction process of the first feature information by the prediction unit is provided.

According to one aspect of the embodiment, there is an effect that the meaning of information can be efficiently interpreted.

FIG. 1 is a diagram illustrating an example of a storage process according to the embodiment. FIG. 2 is a diagram illustrating an example of information processing according to the embodiment. FIG. 3 is a diagram illustrating a configuration example of the information processing system according to the embodiment. FIG. 4 is a diagram illustrating a configuration example of the information processing apparatus according to the embodiment. FIG. 5 is a diagram illustrating an example of the query information storage unit according to the embodiment. FIG. 6 is a diagram illustrating an example of the category information storage unit according to the embodiment. FIG. 7 is a diagram illustrating an example of the classification definition storage unit according to the embodiment. FIG. 8 is a diagram illustrating an example of the model information storage unit according to the embodiment. FIG. 9 is a diagram illustrating an example of the progress information storage unit according to the embodiment. FIG. 10 is a diagram illustrating an example of the first learning model generation process according to the embodiment. FIG. 11 is a diagram illustrating an example of the first learning model generation process according to the embodiment. FIG. 12 is a diagram illustrating an example of a second learning model generation process according to the embodiment. FIG. 13 is a diagram illustrating an example of the first learning model according to the embodiment. FIG. 14 is a diagram illustrating an example of the second learning model according to the embodiment. FIG. 15 is a flowchart showing a procedure for generating the first learning model according to the embodiment. FIG. 16 is a flowchart showing a procedure for generating the second learning model according to the embodiment. FIG. 17 is a diagram illustrating an example of information processing according to the embodiment. FIG. 18 is a diagram illustrating an example of the prediction process according to the embodiment. FIG. 19 is a diagram illustrating an example of information processing according to the modification. FIG. 20 is a diagram illustrating an example of information processing according to the modification. FIG. 21 is a hardware configuration diagram illustrating an example of a computer that realizes the functions of the information processing device.

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

[1. Example of information processing]
First, an example of information processing according to the embodiment will be described with reference to FIGS. 1 and 2. 1 and 2 are diagrams illustrating an example of information processing according to the embodiment. The information processing shown in FIGS. 1 and 2 is performed by the user terminal 10 (see FIG. 3), the search server 50 (see FIG. 3), and the information processing device 100 (see FIG. 3).

The user terminal 10 is an information processing device used by a user. The user terminal 10 is realized by, for example, a smartphone, a tablet terminal, a notebook PC (Personal Computer), a mobile phone, a PDA (Personal Digital Assistant), or the like. In the following, the user terminal 10 may be identified with the user. That is, in the following, the user can be read as the user terminal 10.

In the following, the user identified by the user ID "U1" may be referred to as "user U1". As described above, in the following, when "user U* (* is an arbitrary numerical value)" is described, that user is a user specified by the user ID "U*". For example, when described as “user U2”, the user is the user specified by the user ID “U2”.

The search server 50 is a server device that provides a search service. For example, the search service provided by the search server 50 is a comprehensive search service that can search for any information. The search server 50 stores information about the search query input by the user. Specifically, the search server 50 stores information regarding the search history of the user.

The information processing device 100 is a server device that generates a first learning model (hereinafter, appropriately referred to as “first model”). Specifically, the information processing apparatus 100 acquires information regarding the search query input by the user from the search server 50. Subsequently, the information processing apparatus 100 extracts, from the search queries acquired from the search server 50, a plurality of search queries input by the same user within a predetermined time. Here, the information processing apparatus 100 treats the entire character group input in the search box for each search by the user as one search query input by the user. For example, when the search query including a plurality of character groups such as “Roppongi pasta” is input to the search box in one search by the user U1, the information processing apparatus 100 performs one search for the entire “Roppongi pasta”. Handle as a query. Further, the information processing apparatus 100 may execute a plurality of search queries for which the same user has input each search query within a predetermined time period (for example, within 2 minutes) for a predetermined time period by the same user. Extract as multiple search queries entered in. Subsequently, the information processing apparatus 100 learns that the extracted plurality of search queries have similar features to generate a first learning model that predicts the feature information of the predetermined search query from the predetermined search query. .. Specifically, the information processing apparatus 100 learns the first learning model so that the distributed expressions of the plurality of extracted search queries are similar to each other, and thus the distribution including the characteristic information of the predetermined search query from the predetermined search query. A first learning model that outputs an expression (vector) is generated. More specifically, the information processing apparatus 100 uses a DSSM (Deep Structured Semantic Model) that uses LSTM (Long Short-Term Memory), which is a type of RNN (Recurrent Neural Network) that is also called a recursive neural network, for distributed expression generation. The first learning model that outputs the distributed expression (vector) from the search query is generated by using the above technique. In the following, the LSTM as an extended version of the RNN will be referred to as “LSTM-RNN”. Further, the LSTM as a layer to be replaced by the hidden layer (intermediate layer) of the RNN is described as “LSTM layer”. In the example shown in FIG. 1, each block of the three vertically arranged blocks of each step is described as an LSTM layer. For example, the information processing apparatus 100 assumes that the correct data of the first learning model has a feature that a pair of search queries input by the same user within a predetermined time have similar characteristics, and a distributed expression of the predetermined search query ( A vector) and a distributed expression (vector) of another search query paired with a predetermined search query are learned so as to be similar.

The information processing apparatus 100 also generates a second learning model (hereinafter, appropriately referred to as “second model”). Specifically, the information processing apparatus 100 uses the first learning model to generate a second learning model that predicts a category to which a predetermined search query belongs from a predetermined search query. More specifically, when the information processing apparatus 100 generates the first learning model, the information processing apparatus 100 acquires the generated first learning model (model data MDT1 of the first learning model M1). When the information processing apparatus 100 acquires the first model M1, it uses the acquired first model M1 to generate a second learning model M2. The information processing apparatus 100 re-learns the first model M1 to generate a second learning model M2 having a connection coefficient that is a weight of the learning model different from that of the first model M1. For example, when the search query is input to the learning model, the information processing apparatus 100 learns so that the classification result of the distributed expression output by the learning model corresponds to the category to which the search query belongs, thereby performing a predetermined search. A second learning model M2 that predicts a category to which a predetermined search query belongs is generated from the query.

In addition, the information processing apparatus 100 uses a learning model that learns the characteristics of a plurality of search queries, assuming that the plurality of search queries input by the same user within a predetermined time have similar characteristics, and First feature information indicating the feature of the first input information is predicted from the first input information. The information processing apparatus 100 then uses the information processing apparatus 100 to predict the second characteristic information indicating the characteristic of the second input information from the second input information, as the information used in the prediction processing. The history information that is information about

From here, the flow of information processing is demonstrated using FIG. FIG. 1 is a diagram illustrating an example of information processing according to the embodiment. In FIG. 1, an example will be described in which the information processing apparatus 100 uses the second learning model M2 to predict the category to which the first search query belongs from the first search query. In addition, the information processing device 100 uses progress information that is information about the progress of the prediction process of the category to which the first search query belongs as the information used in the prediction process to predict the category to which the second search query belongs from the second search query. An example of storing will be described. The second search query refers to a search query that is input to the second learning model M2 after the first search query.

FIG. 1 shows the second learning model M2 generated by the information processing apparatus 100. The second learning model M2 is composed of three layers of LSTM-RNN. The layers to the right of the letters "LSTM" shown in FIG. 1 indicate the layers that are LSTM-RNN. That is, the set of three LSTM-RNNs shown in the three vertical layers on the right side of the character “LSTM” shown in FIG. 1 indicates the second learning model M2. Here, the LSTM-RNN is a type of RNN (Recurrent Neural Network). The RNN is a neural network including an input layer, a hidden layer, and an output layer, and is characterized in that the hidden layer has a return value. The LSTM-RNN is a neural network in which the hidden layer of the RNN is replaced with the LSTM layer. FIG. 1 is a diagram in which an operation in which the LSTM layer inputs the value of the hidden layer into the hidden layer again is developed in the progress direction of the process. For example, the processing step in which the character “6” is input to the second learning model M2 composed of a set of three vertical LSTM layers is referred to as “step 1”. Subsequently, the processing step in which the character "book" is input to the second learning model M2 is referred to as "step 2". Subsequently, the processing step in which the character "tree" is input to the second learning model M2 is referred to as "step 3". Subsequently, the processing step in which the character “□” is input to the second learning model M2 is referred to as “step 4”. Subsequently, the processing step in which the character “PA” is input to the second learning model M2 is referred to as “step 5”. Subsequently, the processing step in which the character "S" is input to the second learning model M2 is referred to as "step 6". Subsequently, the processing step in which the character "ta" is input to the second learning model M2 is referred to as "step 7". In this way, the second learning model M2 sequentially processes the character group, which is time series data, one character at a time from the beginning.

In the example illustrated in FIG. 1, the information processing apparatus 100 includes a first search query “Roppongi pasta” that is a character group in which “Roppongi” indicating a place name and “pasta” indicating a type of dish are separated by a space that is a delimiter. Is input to the second learning model M2 one by one from the beginning (step S1). As illustrated in FIG. 1, the second learning model M2 sequentially processes the first search query “Roppongi pasta”, which is a character group, character by character from the beginning.

Subsequently, the information processing apparatus 100 predicts the category to which the first search query “Roppongi pasta” belongs, as information used by the information processing apparatus 100 to predict the category to which the second search query belongs from the second search query. The progress information, which is information about the progress of the process, is stored. Specifically, the information processing device 100 stores information regarding the internal state of the second learning model M2 used in the processing of the neural network. Generally, information about the internal state of a learning model used in the processing of a neural network is roughly divided into information about connection coefficients (also called weights, parameters or weights) and information about activation. There are two types. Here, the activation means an intermediate calculation result of each layer during the calculation by the neural network. The information processing apparatus 100 stores information about activation of the second learning model M2 as information about the internal state of the second learning model M2. That is, the information processing apparatus 100 stores, as the information about the internal state of the second learning model M2, the information about the intermediate calculation result of each layer (each LSTM layer) forming the second learning model M2. In the example illustrated in FIG. 1, the information processing apparatus 100 stores information regarding intermediate calculation results of each layer (each LSTM layer) of the three LSTM layers that form the second learning model M2. For example, the information processing apparatus 100 determines whether the first search query “Roppongi pasta” includes a space that is a predetermined delimiter (hereinafter, the space is appropriately described by the symbol “□”). Subsequently, when the first search query “Roppongi pasta” includes a space that is a predetermined delimiter, the information processing apparatus 100 creates a character group “Roppongi□” that includes a space name that is a place name and “Roppongi”. The corresponding progress information is stored in the storage unit 120 (see FIG. 3) (step S2). Specifically, the information processing apparatus 100 stores information regarding the internal state of the second learning model M2 in the processing step “step 4” as the progress information corresponding to the character group “Roppongi□”. More specifically, the information processing apparatus 100 stores information regarding the activation of the second learning model M2 in the processing step “step 4” as the progress information corresponding to the character group “Roppongi□”. That is, the information processing apparatus 100 stores information regarding the intermediate calculation result of each layer forming the second learning model M2 in the processing step “step 4”. In the example illustrated in FIG. 1, the information processing apparatus 100 includes a vector (which is an intermediate calculation result of each layer (each LSTM layer) of the three LSTM layers forming the second learning model M2 in the processing step “step 4”. For example, 2048-dimensional vector values are stored for three sets (three layers).

Then, the information processing apparatus 100 stores the progress information and outputs the distributed expression of the first search query “Roppongi pasta” (step S3). In addition, the information processing apparatus 100 stores the distributed expression corresponding to the first search query “Roppongi pasta” in the storage unit 120 (FIG. 3) in case the same search query as the first search query “Roppongi pasta” is input. ).

When the information processing apparatus 100 extracts and outputs the distributed expression of the first search query “Roppongi pasta”, the distributed expression of the first search query “Roppongi pasta” is classified into each category as the output data of the second learning model M2. The probability of being output is output for each category (step S4). For example, in the information processing apparatus 100, the probability that the distributed expression of the first search query “Roppongi pasta” belongs to CAT11 (“Find a restaurant”) is “90(%)” and CAT12 (“Find a product”). The probability of belonging to "0 (%)", the probability of belonging to CAT13 ("reserve a restaurant") is "10 (%)", and the probability of belonging to CAT14 ("purchase a product") is "0 (%)". Is output. It should be noted that the fact that the search query belongs to CAT11 (“search for restaurants”) indicates that the search query is a search query that was input with the intention of searching for restaurants. Belonging to CAT12 (“search for product”) indicates that the search query is a search query input with the intention of searching for a product. Further, the fact that the search query belongs to CAT13 (“reserve a restaurant”) indicates that the search query is a search query input with the intention of reserving a restaurant. Further, the fact that the search query belongs to CAT14 (“purchase a product”) indicates that the search query is a search query input with the intention of purchasing the product.

Next, the flow of information processing will be described with reference to FIG. FIG. 2 is a diagram illustrating an example of information processing according to the embodiment. FIG. 2 shows information processing performed after the information processing shown in FIG. In the example illustrated in FIG. 2, the information processing apparatus 100 uses the second search query “Roppongi Okonomiyaki”, which is a character group in which “Roppongi” indicating a place name and “Okonomiki” indicating the type of food are separated by spaces as delimiters. Is input to the second learning model M2 one by one from the beginning (step S5).

Subsequently, the information processing apparatus 100 does the progress information stored in the storage unit 120 include progress information corresponding to a character group that is partially or entirely common to the second search query “Roppongi Okonomiyaki”? Determine whether or not. For example, the information processing apparatus 100 determines whether or not the character group stored in the storage unit 120 includes a character group that matches the second search query “Roppongi Okonomiyaki”. It is assumed that the information processing apparatus 100 determines that there is no character group that matches the second search query “Roppongi Okonomiyaki”. Subsequently, in the information processing apparatus 100, the character group “Roppongi□” included in the second search query “Roppongi Okonomiyaki” is common among the character groups stored in the storage unit 120. Determine if it exists. It is assumed that the information processing apparatus 100 determines that the character group “Roppongi□” common to the character group “Roppongi□” included in the second search query “Roppongi Okonomiyaki” exists. Subsequently, when the information processing apparatus 100 determines that the common character group “Roppongi□” exists, the information processing apparatus 100 acquires the progress information corresponding to the common character group “Roppongi□” from the storage unit 120 (step S6). Specifically, the information processing apparatus 100 acquires information regarding the internal state of the second learning model M2 in the processing step “step 4” of FIG. 1 as the progress information corresponding to the character group “Roppongi□”. More specifically, the information processing apparatus 100 acquires information regarding the activation of the second learning model M2 in the processing step “step 4” of FIG. 1 as the progress information corresponding to the character group “Roppongi□”. To do. That is, the information processing apparatus 100 acquires information regarding the intermediate calculation result of each layer forming the second learning model M2 in the processing step “step 4” of FIG. In the example illustrated in FIG. 2, the information processing apparatus 100 uses the intermediate calculation result of each layer (each LSTM layer) of the three LSTM layers forming the second learning model M2 in the processing step “step 4” of FIG. The value of a certain vector (for example, a 2048-dimensional vector) is acquired for three sets (three layers).

Subsequently, when the information processing apparatus 100 acquires the progress information corresponding to the character group “Roppongi□”, the information processing apparatus 100 uses the progress information corresponding to the acquired character group “Roppongi□” to search for the second search query “Roppongi Okonomiyaki”. The distributed expression is output (step S7). Specifically, when the information processing apparatus 100 acquires the information regarding the internal state of the second learning model M2 in the processing step “step 4” of FIG. 1, the internal state of the second learning model M2 in the processing step “step 4”. To reproduce. Subsequently, the information processing apparatus 100 starts the prediction process after the character group "Okonomiyaki" of the second search query "Roppongi Okonomiyaki" based on the information about the internal state of the second learning model M2 in the processing step "Step 4". To do. That is, the information processing apparatus 100, based on the information about the intermediate calculation result of each layer forming the second learning model M2 in the processing step “step 4”, the character group “okonomiyaki” of the second search query “Roppongi Okonomiyaki”. The subsequent prediction process is started. Then, the information processing apparatus 100 outputs the distributed expression of the second search query “Roppongi Okonomiyaki”.

Subsequently, when the information processing apparatus 100 extracts and outputs the distributed expression of the second search query “Roppongi Okonomiyaki”, the distributed expressions of the second search query “Roppongi Okonomiyaki” are output data of the second learning model M2. The probability of being classified into categories is output for each category (step S8). For example, in the information processing apparatus 100, the probability that the distributed expression of the second search query “Roppongi Okonomiyaki” belongs to CAT11 (“Search for a restaurant”) is set to “90(%)” and CAT12 (“Search for a product”). The probability of belonging to "0 (%)", the probability of belonging to CAT13 ("reserve a restaurant") is "10 (%)", and the probability of belonging to CAT14 ("purchase a product") is "0 (%)". Is output.

As described above, the information processing apparatus 100 uses the learning model in which the features of the plurality of search queries are learned, assuming that the plurality of search queries input by the same user within a predetermined time have similar features. Then, the first feature information indicating the feature of the first input information is predicted from the first input information. Then, the information processing apparatus 100 is information regarding the progress of the prediction process of the first feature information, as the information used in the prediction process of predicting the second feature information indicating the feature of the second input information from the second input information. Store information. With this, the information processing apparatus 100 can remember the prediction result until the middle of the prediction process, acquire caching as needed, and start the process from the middle. That is, the information processing apparatus 100 can efficiently perform the feature information prediction process by using the previously calculated result. Therefore, the information processing apparatus 100 can efficiently interpret the meaning of information.

Note that in FIG. 1, the information processing apparatus 100 relates to the internal state of the second learning model M2 for each unit delimited by a delimiter such as a space when the search query is a character group including a delimiter such as a space. An example of storing information has been described. However, the information processing apparatus 100 is not limited to a delimiter such as a space, and may divide the search query anywhere and store information regarding the internal state of the second learning model M2 in any unit. Specifically, the information processing apparatus 100 determines whether the first search query can be classified according to a certain rule (or a certain procedure). For example, the information processing apparatus 100 determines whether or not the first search query can be classified using morphological analysis. In addition, for example, the information processing apparatus 100 determines whether or not the first search query can be classified using BPE (Byte pair encoding). Subsequently, when the information processing apparatus 100 determines that the first search query can be classified according to a certain rule (or a certain procedure), the first search query is divided into units according to a certain rule (or a certain procedure). The information about the internal state of the second learning model M2 is stored in. For example, when the information processing apparatus 100 determines that the first search query can be classified by using the morpheme analysis, the information regarding the internal state of the second learning model M2 for each unit in which the first search query is divided by using the morpheme analysis. Memorize Further, for example, when the information processing apparatus 100 determines that the first search query can be classified using the BPE, the information regarding the internal state of the second learning model M2 for each unit in which the first search query is classified using the BPE. Memorize

In addition, in FIG. 2, when the information processing apparatus 100 has only one character group (character group “Roppongi□”) in the storage unit 120 that is common to the second search query “Roppongi Okonomiyaki”, the second search query. The example of acquiring the progress information corresponding to the character group “Roppongi □” common to the above has been described. Here, a case where a plurality of character groups common to the second search query exist in the storage unit 120 will be described. Specifically, when the second search query “Roppongi □ Okonomiyaki □ Recommendation” is entered, the character groups “Roppongi □ Okonomiyaki □” and the character group “Roppongi □” are common to the second search query. Consider a case where the two exist in the storage unit 120. When the information processing apparatus 100 refers to the storage unit 120 and determines that there are a plurality of character groups common to the second search query, it determines whether or not there is an inclusive relationship between the plurality of character groups. .. For example, the information processing apparatus 100 determines whether or not there is an inclusive relationship between the character group “Roppongi □ Okonomiyaki □” and the character group “Roppongi □”. Subsequently, when the information processing apparatus 100 determines that there is an inclusive relation between the plurality of character groups, it selects a character group that includes all of the other character groups from the plurality of character groups. Then, the information processing apparatus 100 acquires the progress information corresponding to the selected character group from the storage unit 120. For example, in the information processing apparatus 100, the character group "Roppongi □ Okonomiyaki □" includes the character group "Roppongi □" between the character group "Roppongi □ Okonomiyaki □" and the character group "Roppongi □" (the character group " Roppongi □” is included in the character group “Roppongi □ Okonomiyaki □”). Subsequently, in the information processing apparatus 100, the character group "Roppongi □ Okonomiyaki □" includes the character group "Roppongi □" between the character group "Roppongi □ Okonomiyaki □" and the character group "Roppongi □". When it is determined that the character group “Roppongi □ Okonomiyaki □” including the character group “Roppongi □” is selected. Then, the information processing apparatus 100 acquires the progress information corresponding to the selected character group “Roppongi □ Okonomiyaki □” from the storage unit 120. Note that, in FIG. 2, since the second learning model M2 is the LSTM-RNN, when there are a plurality of character groups that are common to the second search query, an inclusive relationship always exists between the character groups. Therefore, when there are a plurality of character groups common to the second search query, the information processing apparatus 100 selects the longest character group (the character group having the largest number of characters). Then, the information processing apparatus 100 acquires the progress information corresponding to the selected longest character group from the storage unit 120.

[2. Information processing system configuration]
Next, the configuration of the information processing system according to the embodiment will be described with reference to FIG. FIG. 2 is a diagram illustrating a configuration example of the information processing system according to the embodiment. As shown in FIG. 2, the information processing system 1 includes a user terminal 10, a search server 50, and an information processing device 100. The user terminal 10, the search server 50, and the information processing apparatus 100 are connected via a predetermined network N so as to be communicable by wire or wirelessly. Note that the information processing system 1 illustrated in FIG. 2 may include any number of user terminals 10, any number of search servers 50, and any number of information processing devices 100.

The user terminal 10 transmits the search query input by the user to the search server 50. Specifically, the user terminal 10 acquires a search page including a search box for inputting a search query from the search server 50 according to the operation by the user. Subsequently, when the user terminal 10 performs an operation of transmitting a search query following the operation of the user inputting characters in the search box, the user terminal 10 uses the characters input in the search box via the search page as a search query. It transmits to the search server 50. For example, when the user presses the send button of the search query or the enter key is pressed after the user inputs characters in the search box, the user terminal 10 displays the search page through the search page. The characters entered in the search box are transmitted to the search server 50 as a search query.

When the search server 50 receives the search query from the user terminal 10, the search server 50 selects the content corresponding to the received search query and output as the search result. Then, the search server 50 delivers the search result page including the selected content to the user terminal 10. Here, the content distributed by the search server 50 is not limited to the web page displayed by the web browser. For example, the content distributed by the search server 50 may be content displayed by a dedicated application installed in the user terminal 10. The contents distributed by the search server 50 are music contents, images (including not only still images but also moving images) contents, and text contents (including news articles and articles posted on SNS (Social Networking Service)). , Any content such as a content combining images and text, a game content, and the like.

Further, when the search server 50 receives the search query from the user terminal 10, the search server 50 registers the received search query, the user ID for identifying the user who is the transmission source of the search query, and the transmission date and time of the search query in association with each other. .. The search server 50 transmits information regarding the search query input by the user to the information processing apparatus 100 in response to a request from the information processing apparatus 100.

In addition, the user terminal 10 transmits the search query input by the user to the information processing device 100. Specifically, the user terminal 10 acquires, from the information processing apparatus 100, the content including the search box for inputting the search query, according to the operation by the user. Subsequently, when the user terminal 10 performs an operation of transmitting a search query following an operation of inputting a character in the search box by the user, the character group input in the search box via the content is used as a search query. The information is transmitted to the information processing device 100. For example, the user terminal 10 searches through the content when an operation of pressing a send button of a search query or an operation of pressing an enter key is performed after an operation of inputting characters in the search box by the user. The character group input in the box is transmitted to the information processing apparatus 100 as a search query.

The information processing device 100 is a server device that performs the information processing described in FIG. The information processing device 100 generates a first learning model. In addition, the information processing apparatus 100 uses the first learning model to generate the second learning model. The information processing apparatus 100 uses the learning model in which the characteristics of the plurality of search queries are learned as the first input, assuming that the plurality of search queries input by the same user within a predetermined time have similar characteristics. The first feature information indicating the feature of the first input information is predicted from the information. Then, the information processing apparatus 100 is information regarding the progress of the prediction process of the first feature information, as the information used in the prediction process of predicting the second feature information indicating the feature of the second input information from the second input information. Store information.

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

(Communication unit 110)
The communication unit 110 is realized by, for example, a NIC (Network Interface Card) or the like. The communication unit 110 is connected to a network by wire or wirelessly, and transmits and receives information between the user terminal 10 and the search server 50, for example.

(Storage unit 120)
The storage unit 120 is realized by, for example, a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory (Flash Memory), or a storage device such as a hard disk or an optical disk. As illustrated in FIG. 3, the storage unit 120 includes a query information storage unit 121, a category information storage unit 122, a classification definition storage unit 123, a model information storage unit 124, and a progress information storage unit 125.

(Query information storage unit 121)
The query information storage unit 121 stores various information regarding the search query input by the user. FIG. 5 shows an example of the query information storage unit according to the embodiment. In the example illustrated in FIG. 5, the query information storage unit 121 has items such as “user ID”, “date and time”, “search query”, and “search query ID”.

The “user ID” indicates identification information for identifying the user who has input the search query. The “date and time” indicates the date and time when the search server accepted the search query from the user. "Search query" indicates a search query input by the user. The “search query ID” indicates identification information for identifying the search query input by the user.

In the example shown in the first record of FIG. 5, the search query (search query Q11) identified by the search query ID “Q11” corresponds to the search query “Roppongi pasta” shown in FIG. The user ID “U1” indicates that the user who inputs the search query Q11 is the user (user U1) identified by the user ID “U1”. Further, the date and time "2018/9/1 PM 17:00" indicates that the date and time when the search server receives the search query Q11 from the user U1 is 17:00 pm on September 1, 2018. The search query "Roppongi pasta" indicates the search query Q11 input by the user U1.

(Category information storage unit 122)
The category information storage unit 122 stores various kinds of information regarding the search query and the category to which the search query belongs. FIG. 6 shows an example of the category information storage unit according to the embodiment. In the example shown in FIG. 6, the category information storage unit 122 has items such as “search query ID”, “large classification ID”, and “small classification ID”.

The “search query ID” indicates identification information for identifying the search query input by the user. “Large category ID” indicates identification information for identifying the major category of the category into which the search query input by the user is classified. “Small category ID” Indicates identification information for identifying the small category of the category into which the search query input by the user is classified.

In the example shown in the first record of FIG. 6, the search query (search query Q11) identified by the search query ID “Q11” corresponds to the search query “Roppongi pasta” shown in FIG.

(Classification definition storage unit 123)
The classification definition storage unit 123 stores various kinds of information regarding the definition of the category into which the search query is classified. FIG. 7 shows an example of the classification definition storage unit according to the embodiment. In the example shown in FIG. 7, the classification definition storage unit 123 has items such as “major classification ID”, “major classification”, “small classification ID”, and “small classification”.

The “major classification” indicates the major classification of the categories into which the search query is classified. The “major classification ID” indicates identification information for identifying the major classification. In the example shown in FIG. 7, the large classification “purchasing behavior type” corresponds to the large classification described in the example shown in the lower part of FIG. The large classification "purchasing behavior system" indicates a large classification of categories into which the search query is classified based on the purchasing behavior of the user. In the example shown in FIG. 7, the large classification “purchasing behavior type” further has four small classifications. The large classification ID “CAT1” indicates identification information for identifying the large classification “purchasing behavior type”.

“Minor classification” indicates a minor classification of the category into which the search query is classified. The “small category ID” indicates identification information for identifying the small category. In the example shown in FIG. 7, the small category “search for restaurants” is a category belonging to the large category “purchasing behavior”, and the search query classified into the small categories is input by the user with the intention of searching for restaurants. Indicates that the search query was performed. The small classification ID “CAT11” indicates identification information for identifying the small classification “search for a restaurant”.

The small category “search for products” is a category belonging to the large category “purchasing behavior type”, and indicates that the search queries classified into the small categories are search queries input by the user with the intention of searching for products. .. The small classification ID “CAT12” indicates identification information for identifying the small classification “search for products”.

The small category “reserve a restaurant” is a category belonging to the large category “purchasing behavior type”, and the search query classified into the small category is a search query input by the user with the intention of reserving a restaurant. Indicates that. The small classification ID “CAT13” indicates identification information for identifying the small classification “reserve a restaurant”.

The small category “Purchase goods” is a category belonging to the large category “Purchasing behavior type”, and the search queries classified into the small categories are search queries input by the user with the intention of purchasing the product. Show. The small classification ID “CAT14” indicates identification information for identifying the small classification “purchase a product”.

(Model information storage unit 124)
The model information storage unit 124 stores various kinds of information regarding the learning model generated by the information processing device 100. FIG. 8 shows an example of the model information storage unit according to the embodiment. In the example shown in FIG. 8, the model information storage unit 124 has items such as “model ID” and “model data”.

The “model ID” indicates identification information for identifying the learning model generated by the information processing device 100. The “model data” indicates model data of the learning model generated by the information processing device 100. For example, “model data” stores data for converting a search query into a distributed expression.

In the example shown in the first record of FIG. 8, the learning model identified by the model ID “M1” corresponds to the first model M1. The model data “MDT1” indicates the model data (model data MDT1) of the first model M1 generated by the information processing device 100. The details of the generation process of the first model M1 will be described later.

The model data MDT1 includes an input layer to which a search query is input, an output layer, a first element belonging to any layer from the input layer to the output layer and other than the output layer, the first element, and the first element. A second element whose value is calculated based on the weight of one element, and a distributed expression of the search query input to the input layer is output from the output layer according to the search query input to the input layer. Thus, the information processing device 100 may be made to function.

Here, it is assumed that the model data MDT1 is realized by the regression model represented by “y=a1*x1+a2*x2+...+ai*xi”. In this case, the first element included in the model data MDT1 corresponds to the input data (xi) such as x1 and x2. The weight of the first element corresponds to the coefficient ai corresponding to xi. Here, the regression model can be regarded as a simple perceptron having an input layer and an output layer. When each model is regarded as a simple perceptron, the first element can be regarded as a node included in the input layer, and the second element can be regarded as a node included in the output layer.

Further, it is assumed that the model data MDT1 is realized by a neural network having one or a plurality of intermediate layers such as DNN (Deep Neural Network). In this case, the first element included in the model data MDT1 corresponds to any node included in the input layer or the intermediate layer. The second element corresponds to the node at the next stage, which is the node to which the value is transmitted from the node corresponding to the first element. The weight of the first element corresponds to the connection coefficient, which is a weight considered for the value transmitted from the node corresponding to the first element to the node corresponding to the second element.

The information processing apparatus 100 calculates the distributed expression using a model having an arbitrary structure such as the regression model and the neural network described above. Specifically, the model data MDT1 is set with a coefficient so as to output a distributed expression when a search query is input. The information processing apparatus 100 uses such model data MDT1 to calculate the distributed representation.

In the above example, the model data MDT1 is the model (hereinafter referred to as model X1) that outputs the distributed expression of the search query when the search query is input. However, the model data MDT1 according to the embodiment may be a model generated based on a result obtained by repeating input/output of data to/from the model X1. For example, the model data MDT1 may be a model (hereinafter, referred to as model Y1) that has been learned so that the search query is input and the distributed expression output by the model X1 is output. Alternatively, the model data MDT1 may be a model learned so that the search query is input and the output value of the model Y1 is output.

Further, when the information processing apparatus 100 performs the estimation process using GAN (Generative Adversarial Networks), the model data MDT1 may be a model forming a part of GAN.

In the example shown in the second record of FIG. 8, the learning model identified by the model ID “M2” corresponds to the second model M2 shown in FIG. The model data “MDT2” indicates model data (model data MDT2) of the second model M2 generated by the information processing apparatus 100. The details of the generation process of the second model M2 will be described later.

The model data MDT2 includes an input layer to which a search query is input, an output layer, a first element belonging to any layer from the input layer to the output layer and other than the output layer, the first element, and the first element. A second element whose value is calculated based on the weight of one element; and a probability that the search query input to the input layer belongs to each category according to the search query input to the input layer The information processing apparatus 100 may be caused to function so as to output from.

Here, it is assumed that the model data MDT2 is realized by the regression model represented by “y=a1*x1+a2*x2+...+ai*xi”. In this case, the first element included in the model data MDT2 corresponds to the input data (xi) such as x1 and x2. The weight of the first element corresponds to the coefficient ai corresponding to xi. Here, the regression model can be regarded as a simple perceptron having an input layer and an output layer. When each model is regarded as a simple perceptron, the first element can be regarded as a node included in the input layer, and the second element can be regarded as a node included in the output layer.

It is also assumed that the model data MDT2 is realized by a neural network having one or a plurality of intermediate layers such as DNN (Deep Neural Network). In this case, the first element included in the model data MDT2 corresponds to any node included in the input layer or the intermediate layer. The second element corresponds to the node at the next stage, which is the node to which the value is transmitted from the node corresponding to the first element. The weight of the first element corresponds to the connection coefficient, which is a weight considered for the value transmitted from the node corresponding to the first element to the node corresponding to the second element.

The information processing apparatus 100 calculates the probability that the search query belongs to each category using a model having an arbitrary structure such as the regression model and the neural network described above. Specifically, the model data MDT2 is set with a coefficient so as to output the probability that the search query belongs to each category when the search query is input. The information processing device 100 uses such model data MDT2 to calculate the probability that the search query belongs to each category.

In the example described above, the model data MDT2 is a model (hereinafter, referred to as model X2) that outputs the probability that the search query belongs to each category when the search query is input. However, the model data MDT2 according to the embodiment may be a model generated based on a result obtained by repeating input/output of data to/from the model X2. For example, the model data MDT2 may be a model (hereinafter, referred to as model Y2) that has been learned so that the search query is input and the probability that the model X2 is output is output. Alternatively, the model data MDT2 may be a model learned so that the search query is input and the output value of the model Y2 is output.

Further, when the information processing apparatus 100 performs the estimation process using GAN (Generative Adversarial Networks), the model data MDT2 may be a model forming a part of GAN.

(Progress information storage unit 125)
FIG. 9 shows an example of the progress information storage unit according to the embodiment. In the example shown in FIG. 9, the progress information storage unit 125 has items such as “character group”, “progress information”, and “date and time”.

“Character group” indicates a character group separated by a predetermined delimiter included in the first input information. The “progress information” indicates progress information that is information about the progress of the prediction process corresponding to each character group delimited by the predetermined delimiter included in the first input information. Specifically, the “progress information” indicates information regarding the internal state of the second learning model M2. More specifically, the “progress information” indicates information regarding activation of the second learning model M2. For example, the “progress information” indicates information about an intermediate calculation result of each layer (each LSTM layer) forming the second learning model M2. In the example illustrated in FIG. 1, the “progress information” indicates information about an intermediate calculation result of each layer (each LSTM layer) of the three LSTM layers forming the second learning model M2. The “date and time” indicates the date and time when the progress information corresponding to the character group is stored.

In the example shown in the first record of FIG. 9, the character group “Roppongi□” corresponds to the character group “Roppongi□” shown in FIG. In addition, the progress information “CDT11” indicates progress information that is information about the progress of the prediction process corresponding to the character group “Roppongi□”. The date and time “date and time #11” indicates that the date and time when the progress information “CDT11” corresponding to the character group “Roppongi□” is stored is “date and time #11”.

(Control unit 130)
Returning to the description of FIG. 3, the control unit 130 is a controller, and is stored in a storage device inside the information processing device 100 by, for example, a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). It is realized by executing various programs (corresponding to an example of an information processing program) stored in the RAM as a work area. The control unit 130 is a controller, and is realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

As illustrated in FIG. 3, the control unit 130 includes an acquisition unit 131, an extraction unit 132, a generation unit 133, and a prediction unit 134, and realizes or executes the operation of information processing described below. Note that the internal configuration of the control unit 130 is not limited to the configuration shown in FIG.

(Acquisition unit 131)
The acquisition unit 131 acquires various information. Specifically, the acquisition unit 131 acquires information regarding the search query input by the user from the search server 50. When acquiring the information about the search query input by the user, the acquisition unit 131 stores the acquired information about the search query in the query information storage unit 121.

The acquisition unit 131 also acquires category information that is correct answer data regarding the search query and the category to which the search query belongs. For example, the acquisition unit 131 acquires category information prepared in advance by the administrator of the information processing device 100. When acquiring the category information, the acquisition unit 131 stores the acquired category information in the category information storage unit 122.

Alternatively, the acquisition unit 131 may generate correct answer data regarding the search query and the category to which the search query belongs. Specifically, the acquisition unit 131 determines the correct category to which the search query belongs, based on the post-search behavior of the user who searched for the search query. More specifically, the acquisition unit 131 acquires, from the search server 50, information regarding the search history after the search of the predetermined search query by the user who searched for the predetermined search query. Subsequently, the acquisition unit 131 calculates the ratio of users who have performed a predetermined action after the search based on the information regarding the search history after the search of the predetermined search query. Subsequently, the acquisition unit 131 determines, as an action corresponding to the correct answer category, a predetermined action in which the ratio of users who have performed a predetermined action after the search exceeds a predetermined threshold with respect to the user who searched for the predetermined search query. To do. For example, as a ratio of users who searched for the search query Q11 (“Roppongi pasta”) to take a predetermined action after the search, a ratio of users who took action to search for a restaurant was 90%, and to search for a product after the search. It is assumed that the percentage of users who have caused a search is 0%, the percentage of users who have made an action to reserve a restaurant after the search is 10%, and the percentage of users who have made an action to purchase a product after the search is 0%. In this case, since the ratio of the users who have taken the action to search for the restaurant exceeds the predetermined threshold value (for example, 90%), the acquisition unit 131 searches for the action to search for the restaurant in the correct answer of the search query Q11 (“Roppongi pasta”). Determined as the action corresponding to the category. Since the acquisition unit 131 has determined the action of searching for a restaurant as the action corresponding to the correct category, the acquisition unit 131 determines that the correct category to which the search query Q11 (“Roppongi pasta”) belongs is CAT11 (“Search for a restaurant”). .. In this way, the acquisition unit 131 generates correct answer data regarding the search query and the category to which the search query belongs. Then, the acquisition unit 131 acquires the generated correct answer data. When acquiring the category information, the acquisition unit 131 stores the acquired category information in the category information storage unit 122.

The acquisition unit 131 also acquires classification definition information that defines the classification of the search query and the category to which the search query belongs. For example, the acquisition unit 131 acquires the classification definition information prepared in advance by the administrator of the information processing device 100. When acquiring the classification definition information, the acquisition unit 131 stores the acquired classification definition information in the classification definition storage unit 123.

The acquisition unit 131 also acquires the first learning model (model data MDT1) generated by the generation unit 133. Specifically, the acquisition unit 131 acquires the first learning model generated by the generation unit 133 with reference to the model information storage unit 124.

The acquisition unit 131 also acquires the second learning model (model data MDT2) generated by the generation unit 133. Specifically, the acquisition unit 131 refers to the model information storage unit 124 and acquires the second learning model generated by the generation unit 133.

Further, the acquisition unit 131 uses, as the information used by the prediction unit 134 in the prediction process of predicting the second feature information indicating the feature of the second input information from the second input information, the prediction process of the first feature information by the prediction unit 134. Acquire progress information that is information about the progress. Specifically, the acquisition unit 131 determines that the plurality of search queries input by the same user within a predetermined time period have similar features, and the internal state of the learning model that has learned the features of the plurality of search queries. Acquires the progress information that is the information about. More specifically, the acquisition unit 131 determines that a plurality of search queries input by the same user within a predetermined time period have similar features, and that the learning model has learned the features of the plurality of search queries. Get information about activation. The acquisition unit 131 regards each layer (each LSTM layer) that configures a learning model in which the characteristics of a plurality of search queries are learned, as a plurality of search queries input by the same user within a predetermined time have similar characteristics. ) To obtain information about intermediate calculation results. In the example illustrated in FIG. 1, the acquisition unit 131 acquires information regarding an intermediate calculation result of each layer (each LSTM layer) of the three LSTM layers forming the second learning model M2. For example, the acquisition unit 131 determines whether or not the first search query “Roppongi pasta” includes a space that is a predetermined delimiter (hereinafter, the space is appropriately described by a symbol “□”). Subsequently, when the acquisition unit 131 determines that the first search query “Roppongi pasta” includes a space that is a predetermined delimiter, the character group “Roppongi□” that includes the place name “Roppongi” and the delimiter space. Get the progress information corresponding to. Specifically, the acquisition unit 131 acquires information regarding the internal state of the second learning model M2 in the processing step “step 4” as the progress information corresponding to the character group “Roppongi□”. More specifically, the acquisition unit 131 acquires information regarding activation of the second learning model M2 in the processing step “step 4” as the progress information corresponding to the character group “Roppongi □”. That is, the acquisition unit 131 acquires information regarding the intermediate calculation result of each layer forming the second learning model M2 in the processing step “step 4”. In the example illustrated in FIG. 1, the acquisition unit 131 uses a vector (for example, an intermediate calculation result of each layer (each LSTM layer) of the three LSTM layers forming the second learning model M2 in the processing step “step 4” (for example, , 2048-dimensional vector) values for three sets (three layers) are acquired. Subsequently, when acquiring the progress information, the acquisition unit 131 stores the acquired progress information in the progress information storage unit 125.

Further, the acquisition unit 131 is not limited to a predetermined delimiter such as a space, and may delimit a search query anywhere and acquire information regarding the internal state of the second learning model M2 in any unit. Specifically, the acquisition unit 131 determines whether the first search query can be classified according to a certain rule (or a certain procedure). For example, the acquisition unit 131 determines whether or not the first search query can be classified using morphological analysis. In addition, for example, the acquisition unit 131 determines whether the first search query can be classified using BPE (Byte pair encoding). Subsequently, when the acquisition unit 131 determines that the first search query can be classified by a certain rule (or a certain procedure), the acquisition unit 131 divides the first search query by a unit by a certain rule (or a certain procedure). Information about the internal state of the second learning model M2 is acquired. For example, when the acquisition unit 131 determines that the first search query can be classified using morphological analysis, the acquisition unit 131 obtains information about the internal state of the second learning model M2 for each unit in which the first search query is classified using morphological analysis. get. In addition, for example, when the acquisition unit 131 determines that the first search query can be classified using the BPE, the acquisition unit 131 obtains information regarding the internal state of the second learning model M2 for each unit in which the first search query is classified using the BPE. get. Subsequently, when acquiring the information regarding the internal state of the second learning model M2, the acquisition unit 131 stores the acquired information regarding the internal state of the second learning model M2 in the progress information storage unit 125 in association with the character group.

The acquisition unit 131 acquires the progress information of the prediction process performed by sequentially processing the unit information included in the first input information including a plurality of unit information that is information for each processing unit by the prediction unit 134. More specifically, the acquisition unit 131 acquires the progress information for each unit information group divided by the predetermined unit information included in the first input information including a plurality of unit information. For example, the acquisition unit 131 is information about the progress of the prediction process for predicting the category of the first search query for each character group separated by a predetermined delimiter included in the first search query including a plurality of characters. Get information. Subsequently, when acquiring the progress information, the acquisition unit 131 stores the acquired progress information in the progress information storage unit 125 in association with the character group.

In addition, the acquisition unit 131 acquires the progress information for each predetermined number of unit information included in the first input information. For example, the acquisition unit 131 acquires information about the progress of the prediction process of the first characteristic information for each predetermined number of characters included in the first input information including the character that is the unit information. For example, the acquisition unit 131 acquires progress information that is information about the progress of the prediction process of predicting the category of the first search query for each character included in the first search query that includes a plurality of characters. Subsequently, when acquiring the progress information, the acquisition unit 131 stores the acquired progress information in the progress information storage unit 125 in association with the character.

(Extractor 132)
The extraction unit 132 extracts various information. Specifically, the extraction unit 132 extracts, from the search queries acquired by the acquisition unit 131, a plurality of search queries input by the same user within a predetermined time. For example, the extraction unit 132 extracts a plurality of search queries in which the time intervals at which each search query is input by the same user are within a predetermined time. Subsequently, the extraction unit 132 extracts a pair of search queries continuously input by the same user within a predetermined time from among a plurality of search queries input by the same user within a predetermined time. For example, the extraction unit 132 extracts a plurality of search queries in which a time interval in which each search query pair is input by the same user is within a predetermined time. For example, the extraction unit 132, among the search queries acquired by the acquisition unit 131, a search query Q11 (“Roppongi pasta”) that is four search queries continuously input by the same user U1 within a predetermined time. ), search query Q12 (“Roppongi Italian”), search query Q13 (“Akasaka pasta”), and search query Q14 (“Azabu pasta”). After arranging the search queries in the input order, the extraction unit 132 extracts the four search queries input in the order of the search query Q11, the search query Q12, the search query Q13, and the search query Q14. Subsequently, when the four search queries are extracted, the extraction unit 132 regards two search queries that are adjacent in time series as a pair of search queries and is a pair of three search queries (search query Q11, search query). Q12), (search query Q12, search query Q13), and (search query Q13, search query Q14) are extracted. Note that the extraction unit 132 may extract a plurality of search queries in which all search queries have been input by the same user within a predetermined time. Then, the extraction unit 132 selects two search queries from the plurality of extracted search queries, regardless of whether they are adjacent in time series, and sets the selected two search queries as a pair of search queries. You may extract.

In addition, the extraction unit 132 extracts a predetermined search query and other search queries unrelated to the predetermined search query from the search queries acquired by the acquisition unit 131. For example, the extraction unit 132 extracts a predetermined search query from the search queries acquired by the acquisition unit 131. Subsequently, the extraction unit 132 randomly extracts another search query from the search queries acquired by the acquisition unit 131, regardless of the predetermined search query.

(Generator 133)
The generation unit 133 generates various information. Specifically, the generation unit 133 learns that among the search queries acquired by the acquisition unit 131, a plurality of search queries input by the same user within a predetermined time have similar characteristics. , A learning model for predicting the characteristic information of the predetermined search query from the predetermined search query is generated. Specifically, the generation unit 133 learns the learning model such that the distributed expressions of the plurality of search queries input by the same user within a predetermined time are similar to each other, thereby performing a predetermined search from the predetermined search query. A learning model for predicting query feature information is generated. For example, the generation unit 133 generates a learning model by learning so that the distributed expressions of a pair of search queries that are continuously input within a predetermined time period are similar to each other. For example, the generation unit 133 calculates the value of the similarity of the distributed expression (vector) before learning of the pair of search queries. The generation unit 133 also calculates the value of the similarity of the distributed representation (vector) after learning of the pair of search queries. Subsequently, the generation unit 133 trains the learning model such that the similarity value of the distributed expression (vector) after learning is larger than the similarity value of the distributed expression (vector) before learning. In this way, the generation unit 133 trains the learning model such that the two vectors, which are a pair of distributed expressions corresponding to the pair of search queries, are similar in the distributed expression space, and thus the distributed expression (vector ) Is output to generate a learning model. More specifically, the generation unit 133 generates a learning model that outputs a distributed expression (vector) from a search query using the DSSM technique that uses LSTM, which is a type of RNN, for distributed expression generation. For example, the generation unit 133 determines that, as correct answer data of the learning model, a pair of search queries input by the same user within a predetermined time have similar characteristics and a distributed expression (vector) of the predetermined search query. , A predetermined search query and a distributed expression (vector) of another search query paired with each other are learned so as to exist close to each other in the distributed expression space. Further, when the generation unit 133 generates the first learning model, the generation unit 133 stores the generated first learning model (model data MDT1) in the model information storage unit 124 in association with the identification information for identifying the first learning model.

[Example of Generation Processing of First Learning Model]
Here, the flow of the generation process of the first learning model will be described with reference to FIG. 10. FIG. 10 is a diagram illustrating an example of the first learning model generation process according to the embodiment. In the example illustrated in FIG. 10, the extraction unit 132 includes a pair of search queries Q11 “Roppongi pasta” and a search query Q12 “Roppongi Italian” that are continuously input by the same user U1 within a predetermined time. A search query is extracted (step S11).

Subsequently, the generation unit 133 inputs the search query Q11 extracted by the extraction unit 132 into the first model M1, and outputs the vector BQV11, which is a distributed expression of the search query Q11, from the first model M1. Here, the vector BQV11 is a distributed expression of the search query Q11 just output from the output layer of the first model M1, and represents a distributed expression before feedback is given to the first model M1 (before learning). Further, the generation unit 133 inputs the search query Q12 extracted by the extraction unit 132 into the first model M1, and outputs the vector BQV12, which is a distributed expression of the search query Q12, from the first model M1. Here, the vector BQV12 is a distributed expression of the search query Q12 just output from the output layer of the first model M1, and represents a distributed expression before feedback is applied to the first model M1 (before learning). In this way, the generation unit 133 outputs the vector BQV11 that is the distributed expression of the search query Q11 and the vector BQV12 that is the distributed expression of the search query Q12 (step S12).

Subsequently, the generation unit 133 includes a pair of search queries including a search query Q11 (“Roppongi pasta”) and a search query Q12 (“Roppongi Italian”) that are continuously input by the same user U1 within a predetermined time. Is presumed to be a search query input with a predetermined search intent (for example, a search intent of “searching for a restaurant at a certain place”), so that it is assumed that the search queries Q11 have similar features to each other. The first model M1 is trained so that the distributed expression (vector QV11) and the distributed expression (vector QV12) of the search query Q12 paired with the search query Q11 are similar. For example, the magnitude of the angle between the vector BQV11, which is the distributed expression of the search query Q11 before feedback (before learning) to the first model M1, and the vector BQV12, which is the distributed expression of the search query Q12, is Θ. Further, the size of the angle formed by the vector QV11, which is the distributed expression of the search query Q11 after feedback is applied to the first model M1 (after learning), and the vector QV12, which is the distributed expression of the search query Q12, is Φ. At this time, the generation unit 133 trains the first model M1 so that Φ becomes smaller than Θ. For example, the generation unit 133 calculates the value of the cosine similarity between the vector BQV11 and the vector BQV12. Further, the generation unit 133 calculates the value of the cosine similarity between the vector QV11 and the vector QV12. Next, the generation unit 133 causes the first model so that the value of the cosine similarity between the vectors QV11 and QV12 is larger than the value of the cosine similarity between the vectors BQV11 and BQV12 (so that the value approaches 1). Train M1. In this way, the generation unit 133 outputs the distributed expression (vector) from the search query by learning the first model M1 so that the two vectors, which are the pair of distributed expressions corresponding to the pair of search queries, are similar. The first model M1 to be generated is generated (step S13). The generation unit 133 is not limited to the cosine similarity and may calculate the similarity between the distributed expressions (vectors) based on any index as long as it is an index applicable as a distance measure between vectors. Good. Further, the generation unit 133 may train the first model M1 based on any index as long as it is an index applicable as a distance measure between vectors. For example, the generation unit 133 calculates a value of a predetermined distance function such as a Euclidean distance between distributed expressions (vectors), a distance in a non-Euclidean space such as a hyperbolic space, a Manhattan distance, a Mahalanobis distance, or the like. Subsequently, the generation unit 133 may train the first model M1 so that the value of the predetermined distance function between the distributed expressions (vectors) (that is, the distance in the distributed expression space) becomes small.

Next, the flow of the generation process of the first learning model will be described in more detail with reference to FIG. Note that in the description of FIG. 11, parts that overlap with the description of FIG. 10 are omitted as appropriate. FIG. 11 is a diagram showing a generation process of the first learning model according to the embodiment. The example illustrated in FIG. 11 illustrates a state in which the distributed representation (vector) output by the first model M1 generated by the information processing apparatus 100 is mapped in the distributed representation space. The information processing apparatus 100 trains the first model M1 so that a distributed expression of a predetermined search query and a distributed expression of another search query paired with the predetermined search query are closely mapped in the distributed expression space. To do.

In the example illustrated in the upper part of FIG. 11, the extraction unit 132 uses the search query Q11 (“Roppongi pasta”), which is four search queries continuously input by the same user U1 within a predetermined time, and the search query Q12. (“Roppongi Italian”), search query Q13 (“Akasaka pasta”), and search query Q14 (“Azabu pasta”) are extracted. The extraction unit 132 extracts four search queries whose time intervals at which each search query is input by the same user U1 are within a predetermined time. The extraction unit 132 extracts a plurality of search queries in which the same user U1 inputs a pair of search queries to be described later and the time interval is within a predetermined time. After arranging the search queries in the input order, the extraction unit 132 extracts the four search queries input in the order of the search query Q11, the search query Q12, the search query Q13, and the search query Q14. When the four search queries are extracted, the extraction unit 132 defines two search queries that are adjacent in time series as a pair of search queries, and is a pair of three search queries (search query Q11, search query Q12). (Search query Q12, search query Q13) and (search query Q13, search query Q14) are extracted (step S21-1). The extraction unit 132 may extract a plurality of search queries in which all the search queries have been input by the same user U1 within a predetermined time. Then, the extraction unit 132 selects two search queries from the plurality of extracted search queries regardless of whether or not they are adjacent in time series, and sets the selected two search queries as a pair of search queries. You may extract.

Subsequently, the generation unit 133 inputs the search query Q1k (k=1, 2, 3, 4) extracted by the extraction unit 132 into the first model M1, and the search query Q1k (k=1, 2, 3). 4) which is a distributed representation of the vector BQV1k (k=1, 2, 3, 4) is output from the first model M1. Here, the vector BQV1k (k=1, 2, 3, 4) is a distributed representation of the search query Q1k (k=1, 2, 3, 4) just output from the output layer of the first model M1. , A distributed expression before feedback is given to the first model M1 (before learning) (step S22-1).

Subsequently, the generation unit 133 determines that a pair of search queries continuously input by the same user U1 within a predetermined time period has a predetermined search intention (for example, “a place (in the vicinity of Minato-ku, Tokyo) at a restaurant). It is presumed that the search queries have been input according to the search intention of "searching for". Therefore, the search query Q11 is paired with the distributed expression (vector QV11) of the search query Q11 as having similar features. The first model M1 is trained so that the distributed expression (vector QV12) of the search query Q12 is similar. Further, the generation unit 133 trains the first model M1 so that the distributed expression (vector QV12) of the search query Q12 and the distributed expression of the search query Q13 (vector QV13) paired with the search query Q12 are similar. .. Further, the generation unit 133 trains the first model M1 so that the distributed expression (vector QV13) of the search query Q13 and the distributed expression (vector QV14) of the search query Q14 paired with the search query Q13 are similar. .. In this way, the generation unit 133 outputs the distributed expression (vector) from the search query by learning the first model M1 so that the two vectors, which are the pair of distributed expressions corresponding to the pair of search queries, are similar. The 1st model M1 which does is generated (step S23-1).

As a result of the information processing shown in the upper part of FIG. 11, the vector QV1k (k=1, 2, 3, 4), which is the distributed expression of the search query Q1k (k=1, 2, 3, 4), is close to the distributed expression space. It is shown that the position is mapped as the cluster CL11. For example, the search query Q1k (k=1, 2, 3, 4) is a set of search queries searched by the user U1 under the search intention of “searching for a restaurant in a certain place (near Minato-ku, Tokyo)”. Is estimated to be That is, the search query Q1k (k=1, 2, 3, 4) is a search query that is searched under the search intent of “searching for a restaurant in a certain place (in the vicinity of Minato-ku, Tokyo)”. , Are presumed to be search queries having similar features to each other. Here, the information processing apparatus 100 receives the position of the cluster CL11 when a predetermined search query input with the search intention of “search for a restaurant in a certain location (Near Minato-ku, Tokyo)” is input to the first model. It is possible to output a distributed expression that is mapped to. Thereby, for example, the information processing apparatus 100 extracts a search query corresponding to the distributed expression mapped to the position of the cluster CL11 to search for “search for a restaurant in a certain place (near Minato-ku, Tokyo)”. It is possible to extract a search query according to the intention. Therefore, the information processing apparatus 100 can appropriately interpret the meaning of the search query.

In the example illustrated in the lower part of FIG. 11, the extraction unit 132 includes the search query Q21 (“refrigerator 400L”) and the search query Q22, which are three search queries continuously input within the predetermined time by the same user U2. (“Refrigerator Medium-sized”) and search query Q23 (“Refrigerator Medium-sized recommended”) are extracted. After arranging the search queries in the input order, the extraction unit 132 extracts the three search queries input in the order of the search query Q21, the search query Q22, and the search query Q23. When the three search queries are extracted, the extraction unit 132 regards two search queries that are adjacent in time series as a pair of search queries and is a pair of two search queries (search query Q21, search query Q22), (Search query Q22, search query Q23) is extracted (step S21-2).

Subsequently, the generation unit 133 inputs the search query Q2m (m=1, 2, 3) extracted by the extraction unit 132 into the first model M1 to generate the search query Q2m (m=1, 2, 3). The vector BQV2m (m=1, 2, 3), which is a distributed expression, is output from the first model M1. Here, the vector BQV2m (m=1, 2, 3) is a distributed representation of the search query Q2m (m=1, 2, 3) just output from the output layer of the first model M1, and is the first model. A distributed expression before feedback is given to M1 (before learning) is shown (step S22-2).

Subsequently, the generation unit 133 inputs a pair of search queries continuously input by the same user U2 within a predetermined time with a predetermined search intent (for example, a search intent to "search for a medium-sized refrigerator"). Since it is estimated that the search query is a search query that has been performed, it is assumed that the search query Q21 has a distributed expression (vector QV21) and a search query Q22 that forms a pair with the search query Q22 (vector QV22). ) Trains the first model M1 so that and are similar. Further, the generation unit 133 trains the first model M1 so that the distributed expression of the search query Q22 (vector QV22) and the distributed expression of the search query Q23 (vector QV23) paired with the search query Q22 are similar. .. In this way, the generation unit 133 outputs the distributed expression (vector) from the search query by learning the first model M1 so that the two vectors, which are the pair of distributed expressions corresponding to the pair of search queries, are similar. The first model M1 to be generated is generated (step S23-2).

As a result of the information processing shown in the lower part of FIG. 11, the vector QV2m (m=1, 2, 3), which is the distributed expression of the search query Q2m (m=1, 2, 3), is located in the cluster CL21 near the distributed expression space. Is mapped as. For example, the search query Q2m (m=1, 2, 3) is presumed to be a set of search queries searched by the user U2 under the search intention of “searching for a medium-sized refrigerator”. That is, Q2m (m=1, 2, 3) is a search query having similar characteristics in that it is a search query searched under the search intention of “searching for a medium-sized refrigerator”. Presumed. Here, when the predetermined search query input with the search intention of “search for a medium-sized refrigerator” is input to the first model, the information processing apparatus 100 creates a distributed expression that is mapped to the position of the cluster CL21. Can be output. Thereby, for example, the information processing apparatus 100 extracts the search query corresponding to the distributed expression mapped to the position of the cluster CL21, thereby extracting the search query according to the search intention of “inspecting the medium-sized refrigerator”. be able to. Therefore, the information processing apparatus 100 can appropriately interpret the meaning of the search query.

In addition, the generation unit 133 considers that the plurality of randomly extracted search queries are search queries that are searched under different search intentions, and considers that the search queries have mutually different characteristics. 1 Learn the model M1. Specifically, the generation unit 133 causes the distributed expression of the predetermined search query to be different from the distributed expression of the search query that is randomly extracted regardless of the predetermined search query (for example, on the distributed expression space). The first model M1 is trained (so that it is mapped far away in.). In the example illustrated in FIG. 11, it is assumed that the extraction unit 132 extracts the search query Q21 when the search query is randomly extracted regardless of the search query Q11. In this case, the generation unit 133 is configured such that the distributed expression (vector QV11) of the search query Q11 and the distributed expression (vector QV21) of the search query Q21 randomly extracted irrespective of the search query Q11 are different from each other. Train the model M1. As a result, the vector QV1k, which is a distributed expression of the search query Q1k (k=1, 2, 3, 4), which is searched under the search intention of “searching for a restaurant in a certain place (Near Minato-ku, Tokyo)”. It is a distributed representation of a cluster CL11 including (k=1, 2, 3, 4) and a search query Q2m (m=1, 2, 3) searched under the search intention of “searching for a medium-sized refrigerator”. The cluster CL21 including the vector QV2m (m=1, 2, 3) is mapped far in the distributed representation space. That is, the information processing apparatus 100 according to the present invention distributes the distributed expressions of the search queries with different search intentions by learning the first model M1 so that the distributed expressions of the plurality of randomly extracted search queries are different. It is possible to output to a distant position in the expression space.

As a result of the distributed representation (vector) output by the first model M1 generated by the information processing apparatus 100 being mapped in the distributed representation space, in addition to the cluster CL11 and the cluster CL21 described above, the same user can specify a predetermined value. A cluster CL12 or a cluster CL22, which is a set of distributed expressions (vectors) of a plurality of search queries input within the time period, is generated.

As described above, the information processing device 100 acquires the search query input by the user. In addition, the information processing apparatus 100 learns that a plurality of search queries input by the same user within a predetermined time have similar features among the acquired search queries, and thus the predetermined search query is performed. Generate a first model that predicts the feature information of the search query. That is, the information processing apparatus 100 according to the present invention is similar to each other in that the plurality of search queries continuously input within a predetermined time are search queries searched under a predetermined search intention. The first model is trained by regarding it as a search query having a characteristic. Specifically, the information processing apparatus 100 learns the first model so that the distributed expressions of the plurality of search queries input by the same user within a predetermined time are similar to each other, so that the predetermined search query is performed. A first model that outputs a distributed expression including the characteristic information of the search query is generated. That is, the information processing apparatus 100 according to the present invention learns the first model M1 so that the distributed expressions of a plurality of search queries continuously input within a predetermined time are similar to each other so that a predetermined search intention is obtained. The distributed expression of the search query searched below can be output to a close position in the distributed expression space. This allows the information processing apparatus 100 to output (interpret) the meaning (search intent) of the search query in accordance with the context of the user who inputs the search query. Therefore, the information processing apparatus 100 can appropriately interpret the meaning of the search query. Further, the information processing apparatus 100 extracts a search query corresponding to a distributed expression that is mapped in the vicinity of the distributed expression that includes the characteristic information of the predetermined search query, and accordingly, the information processing apparatus 100 responds to the search intention by which the predetermined search query is searched. Search queries can be extracted. That is, the information processing apparatus 100 enables the analysis of the search trend of the user in consideration of the search intention and the context of the user who inputs the search query. Therefore, the information processing apparatus 100 can improve the accuracy of analysis of the search trend of the user.

Further, the first model M1 generated by the information processing device 100 can be made to function as a part of the search system. Alternatively, the information processing apparatus 100 distributes the search query output by the first model M1 as input information to another system (for example, a search engine) that uses the feature information of the search query predicted by the first model M1. You can also provide an expression. Accordingly, the search system can select the content output as the search result based on the characteristic information of the search query predicted by the first model M1. That is, the search system can select the content output as the search result in consideration of the search intention or context of the user who inputs the search query. Further, the search system calculates the similarity between the distributed expression of the character group included in the content output as the search result and the distributed expression of the search query, based on the characteristic information of the search query predicted by the first model M1. It will be possible. Then, the search system can determine the display order of the content output as the search result based on the calculated similarity. That is, the search system can determine the display order of the content output as the search result in consideration of the search intention or context of the user who inputs the search query. Therefore, the information processing apparatus 100 can improve usability in the search service.

[Example of Second Learning Model Generation Processing]
Next, the flow of generation processing of the second learning model will be described with reference to FIG. FIG. 12 is a diagram illustrating an example of a second learning model generation process according to the embodiment. In the following, the second learning model will be appropriately referred to as the second model (or the second model M2). In the example shown in the upper part of FIG. 12, the extraction unit 132 uses the search query Q11 (“Roppongi pasta”), which is four search queries continuously input by the same user U1 within a predetermined time, and the search query Q12. (“Roppongi Italian”), search query Q13 (“Akasaka pasta”), and search query Q14 (“Azabu pasta”) are extracted. The extraction unit 132 extracts a plurality of search queries in which the time intervals at which the search queries are input by the same user U1 are within a predetermined time. In addition, the extraction unit 132 extracts a plurality of search queries in which the time interval at which each search query pair is input by the same user U1 is within a predetermined time. Here, it is assumed that the four search queries are search queries in which the search query Q11, the search query Q12, the search query Q13, and the search query Q14 are input in this order by the user U1 within a predetermined time. When the four search queries are extracted, the extraction unit 132 defines two search queries that are adjacent in time series as a pair of search queries, and is a pair of three search queries (search query Q11, search query Q12). (Search query Q12, search query Q13), (search query Q13, search query Q14) are extracted. When the extraction unit 132 extracts three pairs of search queries, the generation unit 133 inputs the extracted search query Q1k (k=1, 2, 3, 4) to the first model M1 (step S31). .. The extraction unit 132 may extract a plurality of search queries in which all the search queries have been input by the same user U1 within a predetermined time. Then, the extraction unit 132 selects two search queries from the plurality of extracted search queries, regardless of whether they are adjacent in time series, and sets the selected two search queries as a pair of search queries. You may extract.

Subsequently, the generation unit 133 outputs the vector BQV1k (k=1, 2, 3, 4), which is the distributed expression of the search query Q1k (k=1, 2, 3, 4), as the output data of the first model M1. Yes (step S32). Here, the vector BQV1k (k=1, 2, 3, 4) is a distributed representation of the search query Q1k (k=1, 2, 3, 4) just output from the output layer of the first model M1. , A distributed expression before feedback is applied to the first model M1 (before learning).

Here, the search query Q1k (k=1, 2, 3, 4) continuously input by the same user U1 within a predetermined time is, for example, “some place (near Minato-ku, Tokyo)” by the user U1. It is presumed to be a set of search queries searched under the search intention of "searching for restaurants in." That is, the search query Q1k (k=1, 2, 3, 4) is a search query that is searched under the search intent of “searching for a restaurant in a certain place (in the vicinity of Minato-ku, Tokyo)”. , Are presumed to be search queries having similar features to each other. Therefore, the generation unit 133 generates the first model that predicts the characteristic information of the predetermined search query from the predetermined search query by learning that the continuously input search queries have similar characteristics ( Step S33). Specifically, the generation unit 133 learns that the distributed expressions of continuously input search queries are similar to each other, and thus the first model M1 that predicts the distributed expression of the predetermined search query from the predetermined search query. To generate. For example, the generation unit 133 may make the distributed representation of the search query Q11 (vector QV11) and the distributed representation of the search query Q12 (vector QV12) that is paired with the search query Q11 similar to each other in the distributed representation space. The model M1 is trained. In addition, the generation unit 133 makes the distributed representation of the search query Q12 (vector QV12) and the distributed representation of the search query Q13 (vector QV13) paired with the search query Q12 similar to each other in the distributed representation space. The model M1 is trained. In addition, the generation unit 133 makes the distributed expression of the search query Q13 (vector QV13) and the distributed expression of the search query Q14 (vector QV14) paired with the search query Q13 similar to each other in the distributed expression space. The model M1 is trained.

On the right side of the upper part of FIG. 12, the variance of the search query Q1k (k=1, 2, 3, 4) input by the same user U1 within a predetermined time as the output result of the learned first model M1. It is shown that the expression vector QV1k (k=1, 2, 3, 4) is mapped as the cluster CL11 in the distributed expression space. In this way, the generation unit 133 generates the first learning model M1 that has learned the characteristics of the plurality of search queries input by the same user within a predetermined time period.

After generating the first model M1, the generation unit 133 acquires the generated first model M1 (model data MDT1 of the first model M1). When the generating unit 133 acquires the first model M1, the generating unit 133 generates the second learning model M2 using the acquired first model M1. Specifically, the generation unit 133 re-learns the first model M1 to generate the second model M2 having a connection coefficient that is a weight of the learning model different from that of the first model M1. More specifically, the generation unit 133 uses the first model M1 to generate a second learning model M2 that predicts a category to which a predetermined search query belongs from a predetermined search query (step S34).

In the example illustrated in the lower part of FIG. 12, when the search query is input to the second model M2, the generation unit 133 includes CAT11 (“search for a restaurant”), CAT12 (“searches for a product”), and CAT13 (“ A second model M2 that predicts which of the four categories of "reserve a restaurant") and CAT14 ("purchase a product") is generated. Specifically, when the search query is input to the second model M2 as the input information, the generation unit 133 generates the second model M2 that outputs the probability that the search query belongs to the category as the output information for each category. To do. For example, the generation unit 133 learns a set of a search query and a category (one of CAT11 to CAT14) to which the search query belongs as the correct answer data of the second model M2.

It should be noted that the fact that the search query belongs to CAT11 (“search for restaurants”) indicates that the search query is a search query that was input with the intention of searching for restaurants. Belonging to CAT12 (“search for product”) indicates that the search query is a search query input with the intention of searching for a product. Further, the fact that the search query belongs to CAT13 (“reserve a restaurant”) indicates that the search query is a search query input with the intention of reserving a restaurant. Further, the fact that the search query belongs to CAT14 (“purchase a product”) indicates that the search query is a search query input with the intention of purchasing the product.

Specifically, when the search query is input to the learning model, the generation unit 133 learns such that the classification result of the distributed expression output by the learning model corresponds to the category to which the search query belongs, and thus the predetermined result is obtained. The second model M2 that predicts the category to which the predetermined search query belongs is generated from the search query of. Then, for example, when the search query is input as the input information to the second model M2, the generation unit 133 outputs the probability that the search query belongs to the category as the output information for each of the categories CAT11 to CAT14. To generate.

For example, when the search query Q11 (“Roppongi pasta”) is input to the second model M2 as input information (step S35), the generation unit 133 outputs the distributed expression of the search query Q11 (“Roppongi pasta”) as output information. Is output from the second model M2. Here, the vector BQV11 is a distributed expression of the search query Q11 just output from the output layer of the second model M2, and represents a distributed expression before feedback is applied to the second model M2 (before learning). Here, it is assumed that the correct category to which the search query Q11 (“Roppongi pasta”) belongs is CAT11 (“Find a restaurant”). In this case, the generation unit 133 causes the probability that the vector BQV11, which is the distributed expression of the output search query Q11 (“Roppongi pasta”), is classified into CAT11 (“Find a restaurant”) exceeds a predetermined threshold. The second model M2 is trained. The generation unit 133 trains the second model using the correct answer data prepared in advance. Alternatively, the generation unit 133 may train the second model M2 using the correct answer data of the second model M2 generated by the acquisition unit 131.

For example, when the search query Q11 (“Roppongi pasta”) is input to the second model M2 before learning, the generation unit 133 classifies the vector BQV11, which is a distributed expression, into CAT11 (“Find a restaurant”). 80% probability, 0% probability of being classified as CAT12 (“find merchandise”), 20% probability of being classified as CAT13 (“reserve food”), CAT14 (“buy merchandise”) It is assumed that the probability of being classified as is output as 0%. In this case, the generation unit 133 trains the second model M2 so that the probability that the vector BQV11, which is a distributed expression, is classified into CAT11 (“search for a restaurant”) exceeds a predetermined threshold value (for example, 90%). .. In addition, the generation unit 133 performs learning so that the probability that the vector BQV11, which is a distributed expression, is classified into CAT11 (“search for a restaurant”) exceeds a predetermined threshold value (for example, 90%), and the distribution is performed. The second model M2 is trained so that the probability that the expression vector BQV11 is classified into another category CAT13 (“reserve a restaurant”) is reduced to 10%.

As described above, when the predetermined search query is input as the input information, the generation unit 133 is configured so that the probability that the distributed expression of the predetermined search query is classified as the correct category as the output information exceeds the predetermined threshold. Train the model. Then, when a predetermined search query is input as input information, the generation unit 133 sets, as a predetermined search query category, a category whose probability that the distributed expression of the predetermined search query belongs to that category exceeds a predetermined threshold value. Output. For example, when the search query Q11 (“Roppongi pasta”) is input as input information to the learned second model M2, the generation unit 133 generates a vector BQV11 that is a distributed expression of the search query Q11 (“Roppongi pasta”). Since the probability of belonging to the category CAT11 (“search for restaurants”) exceeds 90%, the category to which the search query belongs is output as output information as CAT11 (“searches for restaurants”) (step S36). In this way, the generation unit 133 generates a second model that predicts a category of a predetermined search query from the predetermined search query by learning a set of the search query and the correct category of the search query (step S37). ..

Generally, it is considered that a user performs a search a plurality of times with a certain intention, and therefore it is assumed that search queries continuously input within a predetermined time have similar search intentions. Therefore, the information processing apparatus 100 according to the present invention is similar to each other in that a plurality of search queries continuously input within a predetermined time period are search queries searched under a predetermined search intention. The first model M1 is trained on the assumption that the search query has a characteristic. As a result, the information processing apparatus 100 can cause the first model M1 to learn the characteristics of the search query in consideration of the search intention. Then, the information processing apparatus 100 utilizes the first model M1 that has learned the characteristics of the search query in consideration of the search intention, and efficiently uses the second model that predicts the category of the predetermined search query from the predetermined search query. Can be generated. As a result, the information processing apparatus 100 enables the search query to be classified into categories in consideration of the search intention of the user who inputs the search query. Further, conventionally, it has been necessary to prepare a sufficient amount of correct answer data in order to classify search queries into categories and obtain high classification accuracy. However, since the search queries themselves are diverse and have long tail characteristics, it is very troublesome and difficult to label the correct answer categories corresponding to a large number of search queries. Here, instead of labeling the correct answer category, the information processing apparatus 100 learns the second model that predicts the category of the search query by using the user's search intention (context of the user who inputs the search query) as a kind of correct answer. Can be made Thereby, the information processing apparatus 100 can train the second model without manually labeling the correct category of the search query. That is, the information processing apparatus 100 can obtain sufficient classification accuracy even when the correct answer data is small. In addition, the information processing apparatus 100 can obtain higher classification accuracy when there are many correct data. Therefore, the information processing device 100 can improve the classification accuracy of the search query.

[Example of first learning model]
Next, an example of the first learning model generated by the information processing device 100 will be described with reference to FIG. FIG. 13 is a diagram illustrating an example of the first learning model according to the embodiment. In the example illustrated in FIG. 13, the first learning model M1 generated by the information processing device 100 is composed of three layers of LSTM-RNN. In the example illustrated in FIG. 13, the extraction unit 132 includes a pair of search queries Q11 “Roppongi pasta” and a search query Q12 “Roppongi Italian” that are continuously input by the same user U1 within a predetermined time. Extract the search query. The generation unit 133 inputs the search query Q11 extracted by the extraction unit 132 into the input layer of the first learning model M1 (step S41).

Then, the generation unit 133 outputs a 256-dimensional vector BQV11, which is a distributed expression of the search query Q11, from the output layer of the first learning model M1. Further, the generation unit 133 inputs the search query Q12 extracted by the extraction unit 132 into the input layer of the first learning model M1. Subsequently, the generating unit 133 outputs the 256-dimensional vector BQV12, which is the distributed expression of the search query Q12, from the output layer of the first learning model M1 (step S42).

Subsequently, the generating unit 133 learns so that the distributed expressions (vectors) of the two consecutively input search queries are similar to each other, and thus the first learning model M1 that outputs the distributed expressions (vectors) from the search query. Is generated (step S43). For example, the magnitude of the angle between the vector BQV11, which is the distributed expression of the search query Q11 before feedback (before learning) to the first learning model M1, and the vector BQV12, which is the distributed expression of the search query Q12, is Θ. In addition, the size of the angle formed by the vector QV11, which is the distributed expression of the search query Q11 after feedback is applied to the first learning model M1 (after learning), and the vector QV12, which is the distributed expression of the search query Q12, is Φ. .. At this time, the generation unit 133 trains the first learning model M1 so that Φ becomes smaller than Θ. For example, the generation unit 133 calculates the value of the cosine similarity between the vector BQV11 and the vector BQV12. Further, the generation unit 133 calculates the value of the cosine similarity between the vector QV11 and the vector QV12. Then, the generation unit 133 sets the learning model M1 such that the value of the cosine similarity between the vectors QV11 and QV12 is larger than the value of the cosine similarity between the vectors BQV11 and BQV12 (so that the value approaches 1). To learn. As described above, the generation unit 133 learns the first learning model M1 so that two vectors, which are a pair of distributed expressions corresponding to a pair of search queries, are similar in the distributed expression space, and thus the search query is distributed. A first learning model M1 that outputs an expression (vector) is generated. The generation unit 133 is not limited to the cosine similarity and may calculate the similarity between the distributed expressions (vectors) based on any index as long as it is an index applicable as a distance measure between vectors. Good. Further, the generation unit 133 may train the learning model M1 based on any index as long as it is an index applicable as a distance measure between vectors. For example, the generation unit 133 calculates a value of a predetermined distance function such as a Euclidean distance between distributed expressions (vectors), a distance in a non-Euclidean space such as a hyperbolic space, a Manhattan distance, a Mahalanobis distance, or the like. Subsequently, the generation unit 133 may train the learning model M1 so that the value of the predetermined distance function between the distributed expressions (vectors) (that is, the distance in the distributed expression space) becomes small.

The generation unit 133 also learns that a plurality of search queries that are input by the same user within a predetermined time have a similar feature in a plurality of search queries that include a character group separated by a predetermined delimiter. By doing so, the first learning model is generated. For example, the generation unit 133 uses a search query “Roppongi pasta” in which “Roppongi” indicating a place name and “pasta” indicating the type of dish are separated by a space that is a delimiter, and “Roppongi” indicating a place name. The first learning model is generated by learning that the search query “Roppongi Italian”, which is delimited by a space that is a delimiter between the characters “Italian” indicating the type of food, has similar characteristics.

In addition, the generation unit 133 generates a first learning model by learning that among the search queries acquired by the acquisition unit 131, a plurality of randomly extracted search queries have different characteristics. Specifically, the generation unit 133 generates a first learning model by learning so that the distributed expressions of a pair of search queries that are randomly extracted from among the search queries acquired by the acquisition unit 131 are different. To do. For example, the generation unit 133 causes the distributed expression of the predetermined search query extracted by the extraction unit 132 and the distributed expression of the search query randomly extracted irrespective of the predetermined search query to be far apart in the distributed expression space. The first learning model M1 is trained so as to be mapped.

The generation unit 133 also generates a second learning model. Specifically, the generation unit 133 refers to the model information storage unit 124 and acquires the first learning model (model data MDT1 of the first learning model M1) generated by the generation unit 133. Subsequently, the generation unit 133 uses the acquired first learning model to generate a second learning model that predicts a category to which the predetermined search query belongs from the predetermined search query. When the generating unit 133 acquires the first model M1, the generating unit 133 generates the second learning model M2 using the acquired first model M1. The generation unit 133 re-learns the first model M1 to generate a second model M2 having a connection coefficient that is a weight of the learning model different from that of the first model M1. Specifically, when the search query is input to the learning model, the generation unit 133 learns such that the classification result of the distributed expression output by the learning model corresponds to the category to which the search query belongs, and thus the predetermined result is obtained. The second model M2 that predicts the category to which the predetermined search query belongs is generated from the search query of.

Specifically, when the search query is input to the learning model, the generation unit 133 learns such that the classification result of the distributed expression output by the learning model corresponds to the category to which the search query belongs, and thus the predetermined result is obtained. A second learning model that predicts a category to which a predetermined search query belongs is generated from the search query. The generation unit 133 generates a second learning model that outputs the probability of each category to which the search query belongs as output information when the search query is input as input information to the learning model. For example, the generation unit 133 uses the first model M1 to determine the probability that the distributed expression of the search query is classified as the output information into the category when the predetermined search query is input as the input information into the learning model. A second model M2 to be output for each is generated. When the predetermined search query is input as the input information, the generation unit 133 trains the second model so that the probability that the distributed expression of the predetermined search query is classified as the correct category as the output information exceeds the predetermined threshold. .. Then, when a predetermined search query is input as input information, the generation unit 133 sets, as a predetermined search query category, a category whose probability that the distributed expression of the predetermined search query belongs to that category exceeds a predetermined threshold value. The second model M2 to be output is generated. Further, when the generation unit 133 generates the second learning model, the generation unit 133 stores the generated second learning model (model data MDT2) in the model information storage unit 124 in association with the identification information for identifying the second learning model.

For example, the generation unit 133 refers to the model information storage unit 124 illustrated in FIG. 8 and acquires the first model M1 (model data MDT1 of the first model M1). Subsequently, the generation unit 133 refers to the classification definition storage unit 123 illustrated in FIG. 9 and selects a large classification of categories into which the search query is classified. Subsequently, when selecting the large classification, the generation unit 133 learns a set of the search query and the small classification to which the search query belongs as the learning data of the second model M2.

For example, it is assumed that the correct category to which the search query Q11 (“Roppongi pasta”) belongs is CAT11 (“Find a restaurant”). When the search query Q11 (“Roppongi pasta”) is input to the second model M2 as input information, the generation unit 133 uses the distributed representation of the search query Q11 (“Roppongi pasta”) from the output layer of the second model M2. It outputs a certain vector BQV11. Here, the vector BQV11 is a distributed expression of the search query Q11 just output from the output layer of the second model M2, and represents a distributed expression before feedback is applied to the second model M2 (before learning). In this case, the generation unit 133 causes the probability that the vector BQV11, which is the distributed expression of the output search query Q11 (“Roppongi pasta”), is classified into the correct category CAT11 (“Find a restaurant”) exceeds a predetermined threshold. Thus, the second model M2 is trained.

For example, when the search query Q11 (“Roppongi pasta”) is input to the second model M2 before learning, the generation unit 133 classifies the vector BQV11, which is a distributed expression, into CAT11 (“Find a restaurant”). 80% probability, 0% probability of being classified as CAT12 (“find merchandise”), 20% probability of being classified as CAT13 (“reserve food”), CAT14 (“buy merchandise”) It is assumed that the probability of being classified as is output as 0%. In this case, the generation unit 133 trains the second model M2 so that the probability that the vector BQV11, which is a distributed expression, is classified into CAT11 (“search for a restaurant”) exceeds a predetermined threshold value (for example, 90%). .. In addition, the generation unit 133 performs learning so that the probability that the vector BQV11, which is a distributed expression, is classified into CAT11 (“search for a restaurant”) exceeds a predetermined threshold value (for example, 90%), and the distribution is performed. The second model M2 is trained so that the probability that the expression vector BQV11 is classified into another category CAT13 (“reserve a restaurant”) is reduced to 10%. Subsequently, when the search query Q11 (“Roppongi pasta”) is input as input information to the learned second model M2, the generation unit 133 generates a vector BQV11 that is a distributed expression of the search query Q11 (“Roppongi pasta”). Belongs to the category CAT11 (“search for restaurants”) exceeds 90%, the category to which the search query belongs is output as output information as CAT11 (“searches for restaurants”).

Note that the generation unit 133 may select an arbitrary number of large categories as the large categories. Then, when the search query is input to the second model M2 as the input information, the generation unit 133 determines the probability of belonging to each of the small categories belonging to the arbitrary number of the large categories selected by the search query as the output information for each of the small categories. The second model M2 to be output to may be generated. Further, the generation unit 133 may select all the large classifications as the large classifications. Then, the generation unit 133 may generate the second model M2 that outputs the probability of belonging to each small classification for every small classification when the search query is input to the second model M2.

[Example of Second Learning Model]
Next, an example of the second learning model generated by the information processing device 100 will be described with reference to FIG. FIG. 14 is a diagram illustrating an example of the second learning model according to the embodiment. In the example illustrated in FIG. 14, the second learning model M2 generated by the information processing device 100 is generated using the first learning model M1. That is, the information processing apparatus 100 re-learns the first learning model M1 to generate the second learning model M2 having a connection coefficient which is a weight of the learning model different from the first learning model M1.

More specifically, the second learning model M2 generated by the information processing apparatus 100 is composed of three layers of LSTM-RNN, like the first learning model M1. In the example illustrated in FIG. 14, the extraction unit 132 inputs the search query Q11 “Roppongi pasta” input by the user U1 into the input layer of the second learning model M2 (step S51).

Subsequently, the generation unit 133 outputs the 256-dimensional vector BQV11, which is the distributed expression of the search query Q11, from the output layer of the second learning model M2 (step S52).

Then, the generation unit 133 outputs the probability that the vector BQV11, which is the distributed expression of the search query Q11, is classified into each category (step S53).

Subsequently, the generation unit 133 learns the second learning model M2 so as to increase the probability that the vector BQV11, which is the distributed expression of the search query Q11, is classified into the correct answer category, thereby changing the category of the search query from the search query. A second model to be predicted is generated (step S54).

(Predictor 134)
The prediction unit 134 determines that the plurality of search queries input by the same user within a predetermined time period have similar features, and uses the learning model that has learned the features of the plurality of search queries to use the first input information. From the first feature information indicating the feature of the first input information. Specifically, the acquisition unit 131 refers to the model information storage unit 124 and acquires the second learning model generated by the generation unit 133. Subsequently, the prediction unit 134 uses the second learning model acquired by the acquisition unit 131 to generate the first search information, which is the first input information, from the first search query, which is the first feature information indicating the feature of the first search query. Predict search query categories.

Further, the prediction unit 134 uses the progress information stored in the storage unit 120 to predict the second characteristic information indicating the characteristic of the second input information including the plurality of unit information. Specifically, the prediction unit 134 determines that the plurality of search queries input by the same user within a predetermined time period have similar features, and the prediction unit 134 has learned the features of the plurality of search queries and is in the internal state of the learning model. The second characteristic information indicating the characteristic of the second input information including the plurality of unit information is predicted using the progress information which is the information on the second input information. More specifically, the prediction unit 134 determines that a plurality of search queries input by the same user within a predetermined time period have similar features, and the inside of the learning model that has learned the features of the plurality of search queries. The second characteristic information indicating the characteristic of the second input information including the plurality of unit information is predicted using the progress information that is information about the state. The prediction unit 134 determines that a plurality of search queries input by the same user within a predetermined time have similar features, and information regarding activation of a learning model that has learned the features of the plurality of search queries. Is used to predict the second characteristic information indicating the characteristic of the second input information including a plurality of unit information. The prediction unit 134 determines that the plurality of search queries input by the same user within a predetermined time period have similar features, and that each layer (each LSTM layer) that configures a learning model in which the features of the plurality of search queries are learned. The second characteristic information indicating the characteristic of the second input information including the plurality of unit information is predicted by using the information on the intermediate calculation result of (1). The prediction unit 134 refers to the progress information storage unit 125, and is performed by sequentially processing the unit information included in the first input information including a plurality of unit information that is information for each processing unit by the prediction unit 134. Acquires the progress information of the prediction process. Here, the sequential processing means that the information to be processed is processed according to a certain rule. As an example of a certain rule, there is a method of sequentially processing information including a plurality of processing units from the beginning. More specifically, the prediction unit 134 acquires the progress information for each unit information group divided by the predetermined unit information included in the first input information including a plurality of unit information.

For example, the prediction unit 134 is information about the progress of the prediction process of predicting the category of the first search query for each character group separated by a predetermined delimiter included in the first search query including a plurality of characters. Get information. Then, the prediction unit 134 uses the acquired progress information to predict the category of the second search query including a plurality of characters. Specifically, the prediction unit 134 predicts the second characteristic information by using the progress information corresponding to the unit information common to the second input information among the progress information corresponding to the first input information. For example, the prediction unit 134 uses the progress information, which is information about the internal state of the learning model up to the unit information common to the second input information, among the progress information corresponding to the first input information, to determine the second feature information. Predict. More specifically, the prediction unit 134 uses the progress information, which is the information about the internal state of the learning model up to the unit information common to the second input information, from the progress information corresponding to the first input information, 2 Predict feature information. The prediction unit 134 predicts the second feature information by using information about activation of the learning model up to the unit information common to the second input information among the progress information corresponding to the first input information. The prediction unit 134 uses the information regarding the intermediate calculation result of each layer (each LSTM layer) that constitutes the learning model up to the unit information common to the second input information among the progress information corresponding to the first input information. , The second feature information is predicted. For example, the prediction unit 134 uses the progress information corresponding to the character common to the second search query including the plurality of characters among the progress information corresponding to the first search query including the plurality of characters, and then uses the progress information corresponding to the second search query. Predict the category of. For example, the prediction unit 134 uses the progress information corresponding to the character group common to the second search query including the plurality of character groups from the progress information corresponding to the first search query including the plurality of character groups, and 2 Predict category of search query. For example, the prediction unit 134 uses the progress information up to the character that matches the second search query including the plurality of characters among the progress information corresponding to the first search query including the plurality of characters, and uses the progress information of the second search query. Predict categories. For example, the prediction unit 134 is information about the internal state of the second learning model up to the character that matches the second search query that includes a plurality of characters in the progress information that corresponds to the first search query that includes a plurality of characters. The history information is used to predict the category of the second search query. More specifically, the prediction unit 134 causes the internal state of the second learning model up to the character matching the second search query including the plurality of characters in the progress information corresponding to the first search query including the plurality of characters. The category of the second search query is predicted using the history information that is information about the second search query.

In the example illustrated in FIG. 2, the prediction unit 134 includes, in the progress information stored in the storage unit 120, progress information corresponding to a character group that is partially or entirely common to the second search query “Roppongi Okonomiyaki”. Determine if it exists. For example, the prediction unit 134 determines whether or not the character group stored in the storage unit 120 includes a character group that matches the second search query “Roppongi Okonomiyaki”. It is assumed that the prediction unit 134 determines that there is no character group that matches the second search query “Roppongi Okonomiyaki”. Then, the prediction unit 134 has a character group “Roppongi□” that is common to the character group “Roppongi□” included in the second search query “Roppongi Okonomiyaki” among the character groups stored in the storage unit 120. It is determined whether to do. It is assumed that the prediction unit 134 determines that the character group “Roppongi□” common to the character group “Roppongi□” included in the second search query “Roppongi Okonomiyaki” exists. Subsequently, when the prediction unit 134 determines that the common character group “Roppongi□” exists, the prediction unit 134 acquires the progress information corresponding to the common character group “Roppongi□” from the storage unit 120. Specifically, the prediction unit 134 acquires information about the internal state of the second learning model M2 in the processing step “step 4” of FIG. 1 as the progress information corresponding to the character group “Roppongi□”. More specifically, the prediction unit 134 acquires information regarding the activation of the second learning model M2 in the processing step “step 4” of FIG. 1 as the progress information corresponding to the character group “Roppongi□”. .. That is, the prediction unit 134 acquires information about the intermediate calculation result of each layer forming the second learning model M2 in the processing step “step 4” of FIG. In the example illustrated in FIG. 2, the prediction unit 134 is an intermediate calculation result of each layer (each LSTM layer) of the three LSTM layers that configure the second learning model M2 in the processing step “step 4” of FIG. Three sets (three layers) of vector (for example, 2048-dimensional vector) values are acquired.

Subsequently, when the prediction unit 134 acquires the progress information corresponding to the character group “Roppongi□”, the prediction unit 134 uses the progress information corresponding to the acquired character group “Roppongi□” to distribute the second search query “Roppongi Okonomiyaki”. Output the expression. Specifically, when the prediction unit 134 acquires the information about the internal state of the second learning model M2 in the processing step “step 4” of FIG. 1, the prediction unit 134 determines the internal state of the second learning model M2 in the processing step “step 4”. Reproduce. Subsequently, the prediction unit 134 starts the prediction process after the character group "Okonomiyaki" of the second search query "Roppongi Okonomiyaki" based on the information about the internal state of the second learning model M2 in the processing step "Step 4". .. That is, the prediction unit 134, based on the information about the intermediate calculation result of each layer forming the second learning model M2 in the processing step “Step 4”, the second search query “Roppongi Okonomiyaki” character group “Okonomiki” and subsequent characters. The prediction process of is started. Then, the prediction unit 134 outputs the distributed expression of the second search query “Roppongi Okonomiyaki”.

Subsequently, when the prediction unit 134 extracts and outputs the distributed expression of the second search query “Roppongi Okonomiyaki”, the distributed expression of the second search query “Roppongi Okonomiyaki” is output as the output data of the second learning model M2 in each category. The probability of being classified into is output for each category. For example, the prediction unit 134 assigns the probability that the distributed expression of the second search query “Roppongi Okonomiyaki” belongs to CAT11 (“Find a restaurant”) to “90(%)” and CAT12 (“Find a product”). The probability is “0 (%)”, the probability of belonging to CAT13 (“reserve a restaurant”) is “10 (%)”, and the probability of belonging to CAT14 (“purchase a product”) is “0 (%)”. Output.

When the prediction unit 134 refers to the storage unit 120 and determines that there are a plurality of character groups that are common to the second search query, the prediction unit 134 determines whether or not there is an inclusive relationship between the plurality of character groups. For example, the prediction unit 134, when the second search query “Roppongi □ Okonomiyaki □ recommended” is input, as the character group common to the second search query, the character group “Roppongi □ Okonomiyaki □” and the character group “Roppongi □”. 2” in the storage unit 120, it is determined whether or not there is an inclusive relation between the character group “Roppongi □ Okonomiyaki □” and the character group “Roppongi □”. Subsequently, when the prediction unit 134 determines that there is an inclusive relationship between the plurality of character groups, the prediction unit 134 selects a character group that includes all of the other character groups from the plurality of character groups. Then, the prediction unit 134 acquires the progress information corresponding to the selected character group from the storage unit 120. For example, the prediction unit 134 includes the character group “Roppongi □ Okonomiyaki □” and the character group “Roppongi □” in the character group “Roppongi □ Okonomiyaki □” (the character group “Roppongi □”). □” is included in the character group “Roppongi □ Okonomiyaki □”). Subsequently, the prediction unit 134 determines that the character group “Roppongi □ Okonomiyaki □” includes the character group “Roppongi □” between the character group “Roppongi □ Okonomiyaki □” and the character group “Roppongi □”. When judged, the character group "Roppongi □ Okonomiyaki □" including the character group "Roppongi □" is selected. Subsequently, the prediction unit 134 acquires the progress information corresponding to the selected character group “Roppongi □ Okonomiyaki □” from the storage unit 120. Note that, in FIG. 2, since the second learning model M2 is the LSTM-RNN, when there are a plurality of character groups that are common to the second search query, an inclusive relationship always exists between the character groups. Therefore, when there are a plurality of character groups common to the second search query, the prediction unit 134 selects the longest character group (the character group having the largest number of characters). Then, the prediction unit 134 acquires the progress information corresponding to the selected longest character group from the storage unit 120.

In addition, the prediction unit 134 refers to the progress information storage unit 125 and acquires the progress information for each predetermined number of unit information items included in the first input information. For example, the prediction unit 134 acquires information about the progress of the prediction process of the first characteristic information for each predetermined number of characters included in the first input information including the character that is the unit information. For example, the prediction unit 134 acquires progress information that is information about the progress of the prediction process of predicting the category of the first search query for each character included in the first search query that includes a plurality of characters. Subsequently, the prediction unit 134 predicts the category of the second search query including a plurality of characters by using the acquired progress information. Specifically, the prediction unit 134 predicts the second characteristic information by using the progress information corresponding to the unit information common to the second input information, among the progress information corresponding to the first input information. For example, the prediction unit 134 uses the progress information, which is information about the internal state of the learning model corresponding to the unit information common to the second input information, among the progress information corresponding to the first input information, and uses the second feature information. Predict. More specifically, the prediction unit 134 uses the progress information, which is the information about the internal state of the learning model corresponding to the unit information common to the second input information, among the progress information corresponding to the first input information, Predict the second feature information. For example, the prediction unit 134 uses the progress information corresponding to the character common to the second search query including the plurality of characters among the progress information corresponding to the first search query including the plurality of characters, and then uses the progress information corresponding to the second search query. Predict the category of. For example, the prediction unit 134 uses the progress information up to the character that matches the second search query including the plurality of characters among the progress information corresponding to the first search query including the plurality of characters, and uses the progress information of the second search query. Predict categories. For example, the prediction unit 134 is information about the internal state of the second learning model up to the character that matches the second search query that includes a plurality of characters in the progress information that corresponds to the first search query that includes a plurality of characters. The history information is used to predict the category of the second search query. More specifically, the prediction unit 134 causes the internal state of the second learning model up to the character matching the second search query including the plurality of characters in the progress information corresponding to the first search query including the plurality of characters. The category of the second search query is predicted using the history information that is information about the second search query.

[4. Flow of processing for generating first learning model]
Next, the procedure of the first learning model generation process according to the embodiment will be described with reference to FIG. 15. FIG. 15 is a flowchart showing a procedure for generating the first learning model according to the embodiment. In the example illustrated in FIG. 15, the information processing device 100 acquires the search query input by the user (step S101).

Subsequently, the information processing apparatus 100 extracts a plurality of search queries input by the same user within a predetermined time (step S102).

Subsequently, the information processing apparatus 100 learns that the plurality of extracted search queries have similar features, thereby generating a first learning model that predicts feature information of the predetermined search query from the predetermined search query. (Step S103).

[5. Flow of second learning model generation processing]
Next, the procedure of the second learning model generation process according to the embodiment will be described with reference to FIG. 16. FIG. 16 is a flowchart showing the procedure of the second learning model generation processing according to the embodiment. In the example illustrated in FIG. 16, the information processing device 100 acquires a first learning model (model data MDT1 of the first learning model M1) (step S201).

Then, the information processing apparatus 100 uses the first learning model to generate a second learning model that predicts a category of a predetermined search query from a predetermined search query (step S202).

[6. Information processing flow]
Next, a procedure of information processing according to the embodiment will be described with reference to FIG. FIG. 17 is a flowchart showing a procedure of information processing according to the embodiment. In the example illustrated in FIG. 17, the information processing device 100 determines whether a search query has been accepted (step S301). If the search query has not been received (step S301; No), the information processing apparatus 100 waits until the search query is received.

On the other hand, when the information processing apparatus 100 receives the search query (step S301; Yes), the information processing apparatus 100 determines whether the character group corresponding to the search query includes a predetermined delimiter (step S302). When the character group corresponding to the search query does not include a predetermined delimiter (step S302; No), the information processing apparatus 100 determines whether the progress information corresponding to the character group exists in the progress information storage unit 125. Yes (step S305). When the progress information corresponding to the character group is present in the progress information storage unit 125 (step S305; Yes), the information processing apparatus 100 ends the process. On the other hand, when the progress information corresponding to the character group does not exist in the progress information storage unit 125 (step S305; No), the information processing apparatus 100 stores the progress information corresponding to the character group in the progress information storage unit 125 (step S305). S306).

On the other hand, when the character group corresponding to the search query includes a predetermined delimiter (Yes in step S302), the information processing apparatus 100 stores the progress information corresponding to the character group delimited by the predetermined delimiter as the progress information storage unit. It is determined whether the file exists in 125 (step S303). When the progress information corresponding to the character group delimited by the predetermined delimiter exists in the progress information storage unit 125 (step S303; Yes), the information processing apparatus 100 ends the process. On the other hand, when the progress information corresponding to the character group delimited by the predetermined delimiter does not exist in the progress information storage unit 125 (step S303; No), the information processing apparatus 100 delimits the character group delimited by the predetermined delimiter. The progress information corresponding to is stored in the progress information storage unit 125 (step S304).

[7. Prediction processing flow]
Next, the procedure of the prediction process according to the embodiment will be described with reference to FIG. FIG. 18 is a flowchart showing the procedure of the prediction process according to the embodiment. In the example illustrated in FIG. 18, the information processing device 100 determines whether a search query has been accepted (step S401). If the search query has not been received (step S401; No), the information processing apparatus 100 waits until the search query is received.

On the other hand, when the information processing apparatus 100 receives the search query (step S401; Yes), the information processing apparatus 100 determines whether the character group corresponding to the search query includes a predetermined delimiter (step S402). When the character group corresponding to the search query does not include the predetermined delimiter (step S402; No), the information processing apparatus 100 determines whether the progress information corresponding to the character group exists in the progress information storage unit 125. Yes (step S406). If the historical information corresponding to the character group does not exist in the historical information storage unit 125 (step S406; No), the information processing apparatus 100 predicts the category of the search query (step S405). On the other hand, when the progress information corresponding to the character group exists in the progress information storage unit 125 (step S406; Yes), the information processing apparatus 100 acquires the progress information corresponding to the character group from the progress information storage unit 125 (step S406). S407). When the information processing apparatus 100 acquires the history information corresponding to the character group, the information processing apparatus 100 predicts the category of the search query using the acquired history information (step S405).

On the other hand, in the information processing apparatus 100, when the character group corresponding to the search query includes a predetermined delimiter (step S402; Yes), the progress information corresponding to the character group delimited by the predetermined delimiter is the progress information storage unit. It is determined whether the file exists in 125 (step S403). When the progress information corresponding to the character group delimited by the predetermined delimiter exists in the progress information storage unit 125 (step S403; Yes), the information processing apparatus 100 corresponds to the character group delimited by the predetermined delimiter. The progress information to be acquired is acquired from the progress information storage unit 125 (step S404). When the information processing apparatus 100 acquires the progress information corresponding to the character group separated by the predetermined delimiter, the information processing apparatus 100 predicts the category of the search query using the acquired progress information (step S405). On the other hand, when the historical information corresponding to the character group delimited by the predetermined delimiter does not exist in the historical information storage unit 125 (step S403; No), the information processing apparatus 100 predicts the category of the search query (step S405). ).

[8. Modified example)
The information processing system 1 according to the above-described embodiment may be implemented in various different forms other than the above-described embodiment. Therefore, other embodiments of the information processing system 1 will be described below. The same parts as those in the embodiment are designated by the same reference numerals and the description thereof will be omitted.

In the above-described embodiment, an example in which the second learning model M2 is configured by LSTM (Long Short-Term Memory) which is a type of recursive neural network (Recurrent Neural Network) has been described, but the second learning model M2 is LSTM. -Not limited to RNN. Here, a case where the second learning model M2 has another structure will be described.

[8-1. Recursive Neural Network)
Next, information processing according to the modification will be described with reference to FIG. FIG. 19 is a diagram illustrating an example of information processing according to the modification. The acquisition unit 131 acquires progress information, which is information about the progress of the prediction process of the first feature information predicted by the prediction unit 134, using a learning model that is a tree-structured recursive neural network. .. When acquiring the progress information, the acquisition unit 131 stores the acquired progress information in the progress information storage unit 125 in association with the character group.

In the example illustrated in FIG. 19, the prediction unit 134 describes a space in which “Roppongi” indicating a place name and “ramen” indicating a type of dish are predetermined delimiters (hereinafter, the space is appropriately represented by a symbol “□”). The first search query “Roppongi □ Ramen”, which is a character group delimited by ), is input to the second learning model M2A that is a tree-structured recursive neural network from the beginning (step S1A). When the first search query “Roppongi□Ramen” is input to the second learning model M2A, the prediction unit 134 determines whether or not the first search query “Roppongi□Ramen” includes a space that is a predetermined delimiter. To do. Subsequently, when the prediction unit 134 determines that the first search query “Roppongi □ ramen” includes a space that is a predetermined delimiter, the character group “Roppongi □” included in the first search query “Roppongi □ ramen”. And the character group "ramen" are sequentially processed separately. As illustrated in FIG. 19, the prediction unit 134 sequentially processes the character group “Roppongi□” and the character group “Ramen” included in the first search query “Roppongi□Ramen” separately. Specifically, the prediction unit 134 sequentially processes one character from the beginning of each character group.

For example, the prediction unit 134, in the processing step “step LA1-1”, the prediction result of the first two characters “Roppon” included in the character group “Roppongi□” that is a part of the input information of the second learning model M2A. Is output. Subsequently, the prediction unit 134, in the processing step “step LA2-1”, the third character “tree” included in the character group “Roppongi□” that is a part of the input information and the processing step “step LA1-1”. The prediction result of the first three characters "Roppongi" is output based on the prediction result of the above. Subsequently, the prediction unit 134, in the processing step “step LA3-1”, the fourth character “□” included in the character group “Roppongi□” that is a part of the input information and the processing step “step LA2-1”. The first progress information, which is information about the intermediate calculation result of the character group “Roppongi□” that is a part of the input information, is output based on the prediction result of 1.

In addition, for example, the prediction unit 134, in the processing step “step LA1-2”, the prediction result of the first two characters “Ra” included in the character group “Ramen” that is a part of the input information of the second learning model M2A. Is output. Subsequently, in the processing step “step LA2-2”, the prediction unit 134 executes the third character “me” included in the character group “ramen” that is a part of the input information and the processing step “step LA1-2”. Based on the prediction result, the prediction result of the first three characters “lame” is output. Subsequently, in the processing step “step LA3-2”, the prediction unit 134 executes the fourth character “n” included in the character group “ramen” that is a part of the input information and the processing step “step LA2-1”. Based on the prediction result, the second progress information, which is information about the intermediate calculation result of the character group “ramen” that is a part of the input information, is output.

Then, the acquisition unit 131 uses the prediction unit 134 in the process of predicting the category to which the first search query “Roppongi Ramen” belongs, as information used in the prediction process to predict the category to which the second search query belongs from the second search query. Acquires the progress information that is information about the progress. For example, the acquisition unit 131 uses the first progress information corresponding to the character group “Roppongi□” that includes a space name that is a delimiter and “Roppongi” that represents a place name (the first progress information in the step of the processing step “LA3-1” illustrated in FIG. The information about the intermediate calculation result of each layer forming the 2 learning model M2A) is acquired (step S2A-1). When the acquisition unit 131 acquires the first progress information, the acquisition unit 131 stores the acquired first progress information in the progress information storage unit 125 in association with the character group “Roppongi□”. In addition, the acquisition unit 131 configures the second progress information corresponding to the character group “ramen” that is the “ramen” that indicates the dish name (the second learning model M2A in the step of the processing step “LA3-2” illustrated in FIG. 19 is configured. Information regarding the intermediate calculation result of each layer) is acquired (step S2A-2). When acquiring the second progress information, the acquisition unit 131 stores the acquired second progress information in the progress information storage unit 125 in association with the character group “ramen.

Subsequently, when the storage unit 120 stores the historical information, the prediction unit 134 outputs the distributed expression of the first search query “Roppongi Ramen” (step S3A). In addition, the information processing apparatus 100 stores the distributed expression corresponding to the first search query “Roppongi Ramen” in the storage unit 120 (FIG. 3) in case the same search query as the first search query “Roppongi Ramen” is input. ).

Subsequently, when the information processing apparatus 100 extracts and outputs the distributed expression of the first search query “Roppongi Ramen”, the distributed expression of the first search query “Roppongi Ramen” is output as the output data of the second learning model M2A. The probability of being classified into categories is output for each category (step S4A). For example, in the information processing apparatus 100, the probability that the distributed expression of the first search query “Roppongi Ramen” belongs to CAT11 (“Find a restaurant”) is “90(%)” and CAT12 (“Find a product”). The probability of belonging is “0 (%)”, the probability of belonging to CAT13 (“reserves a restaurant”) is “10 (%)”, and the probability of belonging to CAT14 (“purchases a product”) is “0 (%)”. Is output.

Note that the acquisition unit 131 is not limited to a delimiter such as a space, and may delimit the search query anywhere and store information regarding the internal state of the second learning model M2A in any unit. Specifically, the prediction unit 134 determines whether the first search query can be classified according to a certain rule (or a certain procedure). For example, the prediction unit 134 uses morphological analysis to determine whether the first search query can be classified. Further, for example, the prediction unit 134 determines whether or not the first search query can be classified using BPE. Subsequently, when the prediction unit 134 determines that the first search query can be classified according to a certain rule (or a certain procedure), the prediction unit 134 is divided for each unit of the first search query according to a certain rule (or a certain procedure). , Each of which is processed separately. For example, when the prediction unit 134 determines that the first search query can be classified by using the morphological analysis, the prediction unit 134 sequentially and sequentially processes each of the units by which the first search query is classified by using the morphological analysis. In addition, for example, when the prediction unit 134 determines that the first search query can be classified using the BPE, the prediction unit 134 sequentially processes the first search query using the BPE separately. Subsequently, the acquisition unit 131 acquires corresponding progress information for each unit for which the prediction processing is performed by the prediction unit 134.

[8-2. Dilated Convolutional Neural Network)
Next, information processing according to the modification will be described with reference to FIG. FIG. 20 is a diagram illustrating an example of information processing according to the modification. The acquisition unit 131 acquires progress information, which is information about the progress of the prediction process of the first feature information predicted by the prediction unit 134, using a learning model that is an extended convolutional neural network (Dilated Convolutional Neural Network). When acquiring the progress information, the acquisition unit 131 stores the acquired progress information in the progress information storage unit 125 in association with the character group.

In the example illustrated in FIG. 20, the prediction unit 134 writes a space in which “Roppongi” that represents a place name and “ramen” that represents the type of dish are predetermined delimiters (hereinafter, the space is appropriately represented by a symbol “□”). The first search query “Roppongi □ ramen”, which is a character group delimited by ), is input to the second learning model M2B, which is an extended convolutional neural network (Dilated Convolutional Neural Network), character by character from the beginning (step S1B). When the first search query “Roppongi□Ramen” is input to the second learning model M2B, the prediction unit 134 determines whether or not the first search query “Roppongi□Ramen” includes a space that is a predetermined delimiter. To do. Subsequently, when the prediction unit 134 determines that the first search query “Roppongi □ ramen” includes a space that is a predetermined delimiter, the character group “Roppongi □” included in the first search query “Roppongi □ ramen”. And the character group "ramen" are sequentially processed separately. As illustrated in FIG. 20, the prediction unit 134 sequentially processes the character group “Roppongi□” and the character group “Ramen” included in the first search query “Roppongi□Ramen” separately.

For example, the prediction unit 134, in the processing step “step LB1-1”, the prediction result of the first two characters “Roppongi” included in the character group “Roppongi□” that is a part of the input information of the second learning model M2B. Is output. In addition, for example, the prediction unit 134, in the processing step “step LB1-2”, includes the latter two characters “tree □” included in the character group “Roppongi □” that is a part of the input information of the second learning model M2B. Output the prediction result. Subsequently, in the processing step “step LB2-1”, the prediction unit 134 sets one of the input information based on the prediction result of the processing step “step LB1-1” and the prediction result of the processing step “step LB1-2”. The first progress information, which is information about an intermediate calculation result of the character group “Roppongi□” that is a part, is output.

In addition, for example, the prediction unit 134, in the processing step “step LB1-3”, a prediction result of the first two characters “Ra” included in the character group “Ramen” that is a part of the input information of the second learning model M2B. Is output. Further, for example, the prediction unit 134, in the processing step “step LB1-4”, the prediction result of the latter two characters “men” included in the character group “ramen” that is a part of the input information of the second learning model M2B. Is output. Subsequently, in the processing step “step LB2-2”, the prediction unit 134 sets one of the input information based on the prediction result of the processing step “step LB1-3” and the prediction result of the processing step “step LB1-4”. Second progress information, which is information about an intermediate calculation result of the character group “ramen” that is a part, is output.

Then, the acquisition unit 131 uses the prediction process of the category to which the first search query “Roppongi □ ramen” belongs, as the information used by the prediction unit 134 to predict the category to which the second search query belongs from the second search query. Acquire progress information that is information about the progress. For example, the acquisition unit 131 acquires the first progress information corresponding to the character group “Roppongi□” that includes a space name that is a delimiter and “Roppongi” that represents a place name (step S2B-1). When the acquisition unit 131 acquires the first progress information, the acquisition unit 131 stores the acquired first progress information in the progress information storage unit 125 in association with the character group “Roppongi□”. Moreover, the acquisition part 131 acquires the 2nd progress information corresponding to the character group "ramen" which is "ramen" which shows a dish name (step S2B-2). When acquiring the second progress information, the acquisition unit 131 stores the acquired second progress information in the progress information storage unit 125 in association with the character group “ramen.

Subsequently, when the storage unit 120 stores the historical information, the prediction unit 134 outputs the distributed expression of the first search query “Roppongi□Ramen” (step S3B). Note that the information processing apparatus 100 stores the distributed expression corresponding to the first search query “Roppongi□Ramen” in the storage unit 120 (in preparation for the case where the same search query as the first search query “Roppongi□Ramen” is input. (See FIG. 3).

Subsequently, when the information processing apparatus 100 extracts and outputs the distributed expression of the first search query “Roppongi □ ramen”, the distributed expression of the first search query “Roppongi □ ramen” is output data of the second learning model M2B. The probability of being classified into each category is output for each category (step S4B). For example, in the information processing apparatus 100, the probability that the distributed expression of the first search query “Roppongi □ ramen” belongs to CAT11 (“Find a restaurant”) is “90 (%)”, CAT12 (“Find a product”). The probability of belonging to "0 (%)", the probability of belonging to CAT13 ("reserve a restaurant") is "10 (%)", and the probability of belonging to CAT14 ("purchase a product") is "0 (%)" Is output.

Note that the acquisition unit 131 may store the information about the internal state of the second learning model M2B in any unit, not limited to a delimiter such as a space, where the search query is delimited. Specifically, the prediction unit 134 determines whether the first search query can be classified according to a certain rule (or a certain procedure). For example, the prediction unit 134 uses morphological analysis to determine whether the first search query can be classified. Further, for example, the prediction unit 134 determines whether or not the first search query can be classified using BPE. Subsequently, when the prediction unit 134 determines that the first search query can be classified according to a certain rule (or a certain procedure), the prediction unit 134 is divided for each unit of the first search query according to a certain rule (or a certain procedure). , Each of which is processed separately. For example, when the prediction unit 134 determines that the first search query can be classified by using the morphological analysis, the prediction unit 134 sequentially and sequentially processes each of the units by which the first search query is classified by using the morphological analysis. In addition, for example, when the prediction unit 134 determines that the first search query can be classified using the BPE, the prediction unit 134 sequentially processes the first search query using the BPE separately. Subsequently, the acquisition unit 131 acquires corresponding progress information for each unit for which the prediction processing is performed by the prediction unit 134.

In addition, in the second learning model M2B, the prediction process is performed uniformly (for each unit such as a unit of 2 characters, a unit of 4 characters that collects it, or a unit of 8 characters that further collects it) regardless of the content of the first search query. Therefore, the prediction result in the lower part of the tree structure (in the example shown in FIG. 20, the prediction result of the processing step “step LB1-1” to the processing step “step LB1-4”) is the reuse of the progress information. It is expected that the merit obtained will be small. Therefore, the information processing apparatus 100 uses the prediction result of the lower character group (for example, in units of two characters) of the tree structure as the information used in the prediction process of predicting the category to which the second search query belongs from the second search query. Do not store the corresponding history information. In the example illustrated in FIG. 20, the information processing apparatus 100 has processing steps “step LB1-1” to processing step “step LB1-” as information used in prediction processing for predicting a category to which the second search query belongs from the second search query. The progress information corresponding to the prediction result of "4" is not stored. On the other hand, for the progress information corresponding to the prediction result of the character group (for example, in units of 4 characters or in units of 8 characters) that is somewhat higher in the tree structure, the information processing apparatus 100 only sets the second It is stored as information used in the prediction process of predicting the category to which the second search query belongs from the search query. In the example illustrated in FIG. 20, the information processing apparatus 100 determines whether or not the progress information corresponding to the prediction result of the processing step “step LB2-1” to the processing step “step LB2-2” satisfies the predetermined condition. It is stored as information used in the prediction process of predicting the category to which the second search query belongs from the two search queries. Specifically, the information processing apparatus 100 determines whether or not the character group that is somewhat higher in the tree structure matches a known word that is posted in a dictionary or the like. Subsequently, when the information processing apparatus 100 determines that the character group that is somewhat higher in the tree structure matches a known word published in a dictionary or the like, the second search query to the category to which the second search query belongs. As the information used in the prediction process for predicting, the progress information corresponding to the prediction result of the character group that is somewhat higher in the tree structure is stored. That is, the information processing apparatus 100 has a high possibility of reusing a character group that matches a known word published in a dictionary or the like in the prediction process of predicting the category to which the second search query belongs from the second search query. Store progress information. On the other hand, the information processing apparatus 100 predicts the category to which the second search query belongs from the second search query when the character group at the upper part of the tree structure does not match a known word published in a dictionary or the like. As information used in the prediction process, the progress information corresponding to the prediction result of the character group that is somewhat higher in the tree structure is not stored.

[9. effect〕
As described above, the information processing device 100 according to the embodiment includes the prediction unit 134 and the storage unit 120. The prediction unit 134 determines that the plurality of search queries input by the same user within a predetermined time period have similar features, and uses the learning model that has learned the features of the plurality of search queries to use the first input information. From the first feature information indicating the feature of the first input information. The storage unit 120 uses the prediction unit 134 as the information used in the prediction process of predicting the second feature information indicating the feature of the second input information from the second input information, and the progress of the prediction process of the first feature information by the prediction unit 134. The history information that is information about

With this, the information processing apparatus 100 can remember the prediction result until the middle of the prediction process, acquire caching as needed, and start the process from the middle. That is, the information processing apparatus 100 can efficiently perform the feature information prediction process by using the previously calculated result. Therefore, the information processing apparatus 100 can efficiently interpret the meaning of information.

In addition, the storage unit 120 stores the progress information of the prediction process performed by the prediction unit 134 sequentially processing the unit information included in the first input information including a plurality of unit information that is information for each processing unit. .. The prediction unit 134 predicts the second characteristic information indicating the characteristic of the second input information including the plurality of unit information, using the progress information stored in the storage unit 120.

Accordingly, the information processing apparatus 100 can remember the prediction result for each processing unit until the middle of the prediction process, acquire the caching for each processing unit as needed, and start the process from the middle.

The storage unit 120 also stores the progress information for each unit information group divided by the predetermined unit information included in the first input information.

Thus, the information processing apparatus 100 can remember the prediction result for each unit information group until the middle of the prediction process, acquire the caching for each unit information group as necessary, and start the process from the middle. ..

The storage unit 120 also stores the progress information for each predetermined number of unit information included in the first input information.

Thereby, the information processing apparatus 100 can remember the prediction result for each unit information until the middle of the prediction process, acquire the caching for each unit information as needed, and start the process from the middle.

In addition, the prediction unit 134 predicts the second characteristic information by using the progress information corresponding to the unit information common to the second input information among the progress information corresponding to the first input information.

As a result, the information processing apparatus 100 can efficiently perform the feature information prediction process by using the previously calculated result.

The storage unit 120 also stores information about the progress of the prediction process of the first characteristic information for each predetermined number of characters included in the first input information including the character that is the unit information.

With this, the information processing apparatus 100 can remember the prediction result for each character until the middle of the prediction process, acquire the caching for each character as needed, and start the process from the middle.

The storage unit 120 also stores information about the progress of the prediction process of the first feature information for each character group separated by a predetermined delimiter included in the first input information.

With this, the information processing apparatus 100 can remember the prediction result for each character group until the middle of the prediction process, acquire the caching for each character group as needed, and start the process from the middle.

In addition, the storage unit 120 uses a learning model that is a recursive neural network (Recurrent Neural Network), a tree-structured recursive neural network (Recursive Neural Network), or an extended convolutional neural network (Dilated Convolutional Neural Network). The progress information, which is information about the progress of the prediction process of the first feature information predicted by 134, is stored.

As a result, the information processing apparatus 100 can store the progress information according to the structure of the learning model, acquire the caching as needed, and start the processing from the middle.

[10. Hardware configuration]
Further, the information processing apparatus 100 according to the above-described embodiment is realized by, for example, a computer 1000 configured as shown in FIG. FIG. 21 is a hardware configuration diagram illustrating an example of a computer that realizes the information processing device 100. The computer 1000 includes a CPU 1100, RAM 1200, ROM 1300, HDD 1400, communication interface (I/F) 1500, input/output interface (I/F) 1600, and media interface (I/F) 1700.

The CPU 1100 operates based on a program stored in the ROM 1300 or the HDD 1400, and controls each unit. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 starts up, a program dependent on the hardware of the computer 1000, and the like.

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

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

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

For example, when the computer 1000 functions as the information processing device 100, the CPU 1100 of the computer 1000 realizes the function of the control unit 130 by executing the program loaded on the RAM 1200. The CPU 1100 of the computer 1000 reads these programs from the recording medium 1800 and executes them, but as another example, these programs may be obtained from other devices via a predetermined communication network.

As described above, some of the embodiments of the present application have been described in detail based on the drawings, but these are examples, and various modifications based on the knowledge of those skilled in the art, including the modes described in the section of the disclosure of the invention, It is possible to implement the present invention in other forms with improvements.

[11. Other]
Further, of the processes described in the above-described embodiment and modification, all or part of the processes described as being automatically performed may be manually performed, or described as manually performed. It is also possible to automatically carry out all or part of the above-mentioned processing by a known method. In addition, the processing procedures, specific names, information including various data and parameters shown in the above-mentioned 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.

Further, each component of each device shown in the drawings is functionally conceptual and does not necessarily have to be physically configured as shown. That is, the specific form of distribution/integration of each device is not limited to that shown in the figure, and all or part of the device may be functionally or physically distributed/arranged in arbitrary units according to various loads and usage conditions. It can be integrated and configured.

Further, the above-described embodiments and modified examples can be appropriately combined within a range in which the processing content is not inconsistent.

Also, the above-mentioned "section (module, unit)" can be read as "means" or "circuit". For example, the prediction unit can be read as a prediction unit or a prediction circuit.

1 information processing system 10 user terminal 50 search server 100 information processing device 110 communication unit 120 storage unit 121 query information storage unit 122 category information storage unit 123 category definition storage unit 124 model information storage unit 125 progress information storage unit 130 control unit 131 acquisition Part 132 Extraction part 133 Generation part 134 Prediction part

Claims (10)

It is assumed that the plurality of search queries input by the same user within a predetermined time have similar features, and a learning model that learns the features of the plurality of search queries is used to extract the first input information from the first input information. A prediction unit that predicts first characteristic information indicating characteristics of input information;
As the information used by the prediction unit to predict the second feature information indicating the feature of the second input information from the second input information, it is information about the progress of the prediction process of the first feature information by the prediction unit. A storage unit for storing progress information,
An information processing apparatus comprising: The storage unit is
The unit information included in the first input information including a plurality of unit information that is information for each processing unit stores the progress information of the prediction process performed by being sequentially processed by the prediction unit,
The prediction unit is
The information processing apparatus according to claim 1, wherein second characteristic information indicating a characteristic of the second input information including a plurality of unit information is predicted using the progress information stored in the storage unit. The storage unit is
The information processing apparatus according to claim 2, wherein the progress information is stored for each unit information group divided by predetermined unit information included in the first input information. The storage unit is
The information processing apparatus according to claim 2, wherein the progress information is stored for each of a predetermined number of unit information included in the first input information. The prediction unit is
Among the progress information corresponding to the first input information, the second feature information is predicted using the progress information corresponding to the unit information common to the second input information. The information processing apparatus described in any one of 1. The storage unit is
Information regarding the progress of the prediction process of the first feature information is stored for each of a predetermined number of characters included in the first input information that includes the character that is the unit information. The information processing device according to one. The storage unit is
The information regarding the progress of the prediction process of the first feature information is stored for each character group delimited by a predetermined delimiter included in the first input information. Information processing device described in. The storage unit is
The first predicted by the prediction unit using a learning model that is a recurrent neural network, a tree-structured recursive neural network, or a dilated convolutional neural network. The information processing apparatus according to any one of claims 1 to 7, characterized in that progress information, which is information relating to the progress of the feature information prediction process, is stored. An information processing method executed by a computer,
It is assumed that the plurality of search queries input by the same user within a predetermined time have similar features, and a learning model that learns the features of the plurality of search queries is used to extract the first input information from the first input information. A prediction step of predicting first characteristic information indicating characteristics of input information,
The information used in the prediction process of predicting the second feature information indicating the feature of the second input information from the second input information by the prediction process is information about the progress of the prediction process of the first feature information by the prediction process. A storage step of storing certain progress information,
An information processing method comprising: It is assumed that the plurality of search queries input by the same user within a predetermined time have similar features, and a learning model that learns the features of the plurality of search queries is used to extract the first input information from the first input information. Prediction means for predicting first characteristic information indicating characteristics of input information;
The information used by the prediction unit to predict the second characteristic information indicating the characteristic of the second input information from the second input information is information about the progress of the prediction process of the first characteristic information by the prediction unit. Storage means for storing certain progress information,
An information processing program that causes a computer to execute.

Images