JP2019053386A - Information processing apparatus, information processing method, and program - Google Patents

Abstract

To provide an information processing apparatus for improving search accuracy, an information processing method, and a program.SOLUTION: An information processing apparatus includes: an analysis unit which analyzes a query used for search, and divides the query into words; a conversion unit which converts the words obtained by the analysis unit into feature vectors; and a first estimation unit which learns relationship between feature vectors of the words included in the query in learning data and a category associated with the query in the learning data, to estimate a category associated with the query used for search.SELECTED DRAWING: Figure 1

Description

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

  In the field of web search, in accordance with a keyword (hereinafter referred to as “query”) input by a user, a technique for providing information on various products or services related to the query is performed. In addition, in order to provide highly accurate information on products or services intended by users, development of search methods using machine learning is also in progress.

JP-T-2002-530948

  However, in the related art, there is a case where information on a product or service unintended by the user is provided because the degree of relevance to the query is low. In addition, the conventional method using machine learning uses a simple linear model and cannot consider a combination of words included in a query. In addition, in the field of general machine learning, research on a method using an encoder-decoder model in addition to a linear model is underway, but application to the field of web search has not been assumed (for example, patents). Reference 1).

  The present invention has been made in consideration of such circumstances, and an object of the present invention is to provide an information processing apparatus, an information processing method, and a program capable of improving the search accuracy.

  In one aspect of the present invention, an analysis unit that analyzes a query used for search and divides the query into words, a conversion unit that converts the words divided by the analysis unit into feature vectors, and learning data A first estimation unit that estimates a category associated with the query used in the search by learning a relationship between a feature vector of a word included in the query and a category associated with the query in the learning data; An information processing apparatus is provided.

  According to one embodiment of the present invention, search accuracy can be improved.

It is a figure which shows the use environment of the search server 10 of 1st Embodiment. It is a figure which shows an example of the tree structure of the category of 1st Embodiment. It is a figure which shows an example of the flow of the learning process of the search server 10 of 1st Embodiment. It is a figure which shows an example of the flow of the category estimation process of the search server 10 of 1st Embodiment. It is a figure which shows typically the category estimation process in the category estimation part 36 of 1st Embodiment. It is a figure which shows an example of the search result page of 1st Embodiment. It is a figure explaining the model utilized in evaluation experiment. It is a figure which shows the result of evaluation experiment. It is a figure which shows the comparison result of the coincidence rate according to the number of clicks of evaluation experiment. It is a figure which shows an example of the category estimated using the model 1 of evaluation experiment. It is a functional block diagram of the search server 11 of 2nd Embodiment. It is a figure which shows an example of the search log data of 2nd Embodiment. It is a figure which shows an example of the flow of the category estimation process of the search server 11 of 2nd Embodiment. It is a functional block diagram of search server 12 of a 3rd embodiment.

  Hereinafter, embodiments of an information processing device, an information processing method, and a program will be described with reference to the drawings.

<1. First Embodiment>
The first embodiment of the present invention will be described below. The information processing apparatus according to the present embodiment divides a query used for search into words, converts the divided words into feature vectors, and estimates a category associated with the query. The “category” in the present embodiment is information indicating a field to which content contents (products or services) provided as a search result for a query used for a search belong. In the present embodiment, an example will be described in which the information processing apparatus is a search server that provides a search result of a product or service content.

<1-1. System configuration>
FIG. 1 is a diagram illustrating a use environment of a search server 10 (information processing apparatus) according to the present embodiment. The search server 10 is connected to the terminal device T and the like via the network NW. The network NW includes, for example, a WAN (Wide Area Network), a LAN (Local Area Network), the Internet, a dedicated line, a wireless base station, a provider, and the like.

  The terminal device T is operated by a user who uses a search service provided by the search server 10. The terminal device T is, for example, a computer device such as a personal computer, a mobile phone such as a smartphone, a tablet terminal, or a PDA (Personal Digital Assistant). In the terminal device T, a browser or an application program that operates based on a user operation transmits, for example, a query requesting information provision to the search server 10 and receives information associated with the query from the search server 10.

  The search server 10 transmits the page information of the search result associated with the query input from the terminal device T to the terminal device T. The search server 10 generates search search page information using search data that associates a query with reference information for referring to a page of a site associated with the query. The reference information includes, for example, a URL (Uniform Resource Locator).

  The search server 10 includes, for example, a communication unit 30, a query analysis unit 32 (analysis unit), a query conversion unit 34 (conversion unit), a category estimation unit 36 (first estimation unit, learning unit), and search result generation. A unit 38 (generation unit) and a storage unit 40 are provided. The communication unit 30 communicates with the terminal device T and the like via the network NW. The communication unit 30 receives a query from the terminal device T via the network NW, and transmits page information of the search result to the terminal device T. The communication unit 30 includes a communication interface such as a NIC, for example.

  The query analysis unit 32 analyzes the query input from the terminal device T and divides it into word levels. The query analysis unit 32 analyzes the query by morphological analysis, for example, and divides it into at least one word.

  The query conversion unit 34 converts each of the words divided by the query analysis unit 32 into a feature vector. The query conversion unit 34 converts each of the divided words into a feature vector using, for example, a tool (program) called word2vec. word2vec is a tool that uses a neural network, and converts text information included in the input corpus into a feature vector indicating the characteristics of the text information and outputs the converted feature information. In the present embodiment, it is assumed that a distributed expression of a query is learned in advance in word2vec.

  The category estimation unit 36 estimates a category associated with at least one feature vector input from the query conversion unit 34. The category estimation unit 36 estimates a category using, for example, an encoder / decoder model based on a recurrent neural network (RNN). In the following, an example in which the category estimation unit 36 uses an encoder / decoder model will be described.

  The category estimation unit 36 includes, for example, an encoder 42 and a decoder 44. The encoder 42 and the decoder 44 use, as learning data, a set of “word feature vector” included in the query used for the search and “category” selected by the user in the search result page provided in response to the query. Perform learning process.

  The encoder 42 performs an encoding process on at least one feature vector input from the query conversion unit 34 using a model (hereinafter referred to as an “encoding model”) that has been learned using the learning data described above. Performs an encoding process on the first feature vector when two or more feature vectors are input from the query conversion unit 34, and then performs an encoding process result of the first feature vector and a second feature vector. The encoder 42 inputs the result of the encoding process obtained by repeating the encoding process to the decoder 44.

  The decoder 44 selects a category to be associated with the query word based on the result of the encoding process input from the encoder 42 using the model (hereinafter referred to as “decode model”) that has been learned using the learning data described above. The category has, for example, a tree structure, and the fields to which the goods or services belong are defined in a hierarchical manner, and Fig. 2 is a diagram showing an example of the category tree structure. In the example shown, “Men's fashion”, “Ladies' fashion”, etc. are defined as the first hierarchy H1, and “Men's shoes” are defined as the second hierarchy H2 subordinate to “Men's fashion (first hierarchy H1)”. ”,“ Men's bag ”, etc. In addition, as the third hierarchy H3, which is lower than“ Men's shoes (second hierarchy H2) ”, Over "," business shoes ", such as" Sandals "," boots "are defined.

  The decoder 44 learns the relationship between the encoding process result (multidimensional vector) input from the encoder 42 and the category tree structure as shown in FIG. Based on the result of the encoding process, the decoder 44 estimates a category to be associated with the query word, and sequentially outputs the category from the upper hierarchy in the category tree structure.

  The search result generation unit 38 generates a search result page that is a search result based on the category estimated by the category estimation unit 36. The search result generation unit 38 transmits the generated search result page to the terminal device T via the communication unit 30.

  Each functional unit of the search server 10 is realized, for example, when a hardware processor such as a CPU (Central Processing Unit) executes a program (software) in a computer. Some or all of these components are hardware (circuits) such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), and GPU (Graphics Processing Unit). Part (including circuit)), or may be realized by cooperation of software and hardware.

  The storage unit 40 stores search data D1 in which a query is associated with reference information for referring to a page of a site associated with the query, category data D2 indicating a tree structure of categories, and the like. The storage unit 40 is realized by, for example, a random access memory (RAM), a read only memory (ROM), a hard disk drive (HDD), a flash memory, or a hybrid storage device in which a plurality of these are combined. Further, a part or all of the storage unit 40 may be an external device accessible by the search server 10 such as a NAS or an external storage server.

<1-2. Learning process>
Hereinafter, the learning process of the search server 10 will be described. FIG. 3 is a diagram illustrating an example of a learning process flow of the search server 10. The category estimation unit 36 learns the relationship between the feature vector of words included in the query and the category associated with the query.

  First, learning data is input to the category estimation unit 36 of the search server 10 (S101). As the learning data, for example, log data when the search service is used is used. As the learning data, for example, a set of “word feature vector” included in the query used for the search and “category” selected by the user in the search result page provided in response to the query is used.

  Next, when the encoder 42 of the category estimation unit 36 encodes the “feature vector of words included in the query” (S103) and the decoder 44 decodes the encoding processing result by the encoder 42 (S105), The parameters in the encoder model and the decoder model are adjusted so that the “category” as a set is output (S107).

  For example, as a result of the decoding process, the decoder 44 calculates a score indicating the degree of association with the query word for each category defined as the first hierarchy. The decoder 44 adjusts the parameters in the decoder model so that the score of the category paired with the query in the learning data is high. The category estimation unit 36 may adjust the parameter by, for example, a back propagation method. Similar processing is performed for the second and lower layers. Thus, the process of this flowchart is completed.

  In log data, out of the category data that exist for a certain query, exclude the data of the category with a small number of clicks (data whose number of clicks (number of selected times) is below a predetermined threshold) and exclude it. Later log data may be used as learning data. Thereby, the accuracy of the encoder model and the decoder model can be improved.

<1-3. Category estimation processing>
Hereinafter, the category estimation process of the search server 10 will be described. FIG. 4 is a diagram illustrating an example of the flow of category estimation processing of the search server 10.

  First, the query analysis unit 32 acquires a query input from the terminal device T via the communication unit 30 (S201). Next, the query analysis unit 32 analyzes the query input from the terminal device T and divides it into word levels (S203). The query analysis unit 32 analyzes the query by morphological analysis, for example, and divides it into word levels.

  Next, the query conversion unit 34 converts each of the words divided by the query analysis unit 32 into a feature vector (S205). The query conversion unit 34 converts each divided word into a feature vector using, for example, word2vec.

  Next, the encoder 42 performs an encoding process on the feature vector of the first word among the feature vectors converted by the query conversion unit 34 (S207). Next, the encoder 42 determines whether or not the encoding process for the feature vectors of all words has been completed (S209).

  If the encoder 42 determines that the encoding process has not been completed for the feature vectors of all words, the feature vector of the second word and the first word of the feature vectors converted by the query converter 34 are displayed. The encoding process is performed again using the encoding process result for the feature vector (S207). Next, the encoder 42 determines whether or not the encoding process for the feature vectors of all words has been completed (S209). Thereafter, the same processing is repeated until the encoding processing for the feature vectors of all words is completed. When the encoder 42 determines that the encoding process has been completed for the feature vectors of all words, the encoder 42 inputs the encoding process result obtained as a result of repeating the above-described encoding process to the decoder 44.

  Next, the decoder 44 performs a decoding process on the encoding process result input from the encoder 42 (S211). As a result of the decoding process, the decoder 44 calculates a score indicating the degree of association with the query word for each category defined as the first hierarchy. For example, the category with the highest score is estimated as the category associated with the query.

  Next, the decoder 44 determines whether or not the estimation of the lowermost category has been completed (S213). When the decoder 44 determines that the estimation of the lowermost category has not been completed, the decoder 44 performs the decoding process again using the estimated category information and the decoding process result (S211). Next, the encoder 42 determines whether or not the estimation of the lowermost category has been completed (S213). Thereafter, the same processing is repeated until the estimation of the lowermost category is completed. When the decoder 44 determines that the estimation of the lowermost category has been completed, the decoder 44 inputs information on the estimated category to the search result generation unit 38.

  FIG. 5 is a diagram schematically showing category estimation processing in the category estimation unit 36. In the example illustrated in FIG. 5, the query analysis unit 32 divides the query “men's sneaker A company” input from the terminal device T into three words “men's”, “sneakers”, and “company A”. An example in which the conversion unit 34 converts each of these three words into a feature vector and inputs the feature vector to the category estimation unit 36 is shown.

  First, the encoder 42 performs an encoding process on the feature vector of the first word “men”, and obtains a first encoding process result R1. Next, the encoder 42 performs an encoding process using the feature vector of the second word “sneakers” and the first encoding process result R1 to obtain a second encoding process result R2. Next, the encoder 42 performs an encoding process using the feature vector of the third word “Company A” and the second encoding process result R2 to obtain a third encoding process result R3. The encoder 42 inputs the third encoding processing result R3 to the decoder 44.

  Next, the decoder 44 performs a decoding process on the third encoding process result R3 input from the encoder 42. Here, the decoder 44 calculates a score indicating the degree of association with the three words “men”, “sneakers”, and “Company A” for each category of the highest layer (first layer H1). The decoder 44 sets the category indicating the highest score among the scores calculated for each category to the first category (in this example, “Men”, “Sneakers”, and “Company A”) associated with the three words. Men's fashion ").

  Next, the decoder 44 performs a decoding process using the decoding process result (first decoding process result R4) for the third encoding process result R3 and data indicating the estimated category “men's fashion”. Here, the decoder 44 has three words “men”, “sneakers”, and “Company A” for each category of the second hierarchy H2 that is positioned below the estimated category “men's fashion (first hierarchy)”. A score indicating the degree of association with is calculated. The decoder 44 selects the category indicating the highest score among the scores calculated for each category as a second category (in this example, “Men”, “Sneakers”, and “Company A”). Men's shoes ").

  Next, the decoder 44 performs a decoding process using the decoding process result (second decoding process result R5) for the first decoding process result R4 and data indicating the estimated category “men's shoes”. Here, the decoder 44 has three words “men”, “sneakers”, and “Company A” for each category of the third hierarchy H3 that is located below the estimated category “men's shoes (second hierarchy)”. A score indicating the degree of association with is calculated. Of the scores calculated for each category, the decoder 44 assigns a category indicating the highest score to a third category (in this example, “Men”, “Sneakers”, and “Company A”). Sneakers)).

  As a result of the above processing shown in FIG. 5, categories associated with the three words “men's”, “sneakers”, and “Company A” are classified as “men's fashion (first level)” and “men's shoes (second level)”. ”And“ sneakers (third layer) ”are obtained.

  Next, the search result generation unit 38 generates a search result page, which is a search result, based on the category estimation result input from the decoder 44, and transmits it to the terminal device T (S215). FIG. 6 is a diagram illustrating an example of the search result page P1. In the example illustrated in FIG. 6, the search result generation unit 38 generates a search result page P1 in which “Sneakers”, which is the estimated category of the third hierarchy, is arranged at the highest visibility position in the page.

  In the example shown in FIG. 6, in addition to the estimated category “sneakers”, “boots” that is another category below “Men's shoes (second layer)” that is the estimated category of the second hierarchy. ", Sandals", "business shoes", etc. are displayed in order of increasing score, for example. The search result generation unit 38 transmits the generated search result page P1 to the terminal device T via the communication unit 30. The user of the terminal device T can acquire a search result according to the user's intention by operating the terminal device T and selecting one category included in the search result page P1. Thus, the process of this flowchart is completed.

  In the example illustrated in FIG. 6, the example in which the search result page P <b> 1 that allows the user to select the estimated category is transmitted to the terminal device T has been described. However, the search result generation unit 38 may transmit a page of products or services related to the estimated category to the terminal device T instead of the search result page P1 that allows the user to select the estimated category. .

<1-3. Evaluation Experiment>
In order to evaluate the category estimation accuracy of the search server 10 in this embodiment, an evaluation experiment using three models as shown in FIG. 7 was performed. Model 1 is an encoder / decoder model used in the search server 10 of the present embodiment. Model 2 is a linear model using a prior art linear classifier. Model 3 is a model based on the logs of the prior art (result-based). In this model 3, a category having a large number of clicks is selected in the search log. In this evaluation experiment, the category with the largest number of clicks among the categories clicked by the user on the search result page provided according to the actual use in the search was defined as the “correct answer category”. In this evaluation experiment, a query actually used for a search is input to the search server 10 that performs a category estimation process using each model, and the matching rate between the category estimated by the search server 10 and the correct category Was calculated.

  FIG. 8 is a diagram showing the results of the evaluation experiment described above. As shown in FIG. 8, in any of the first hierarchy category, the second hierarchy category, and the third hierarchy category, the matching rate of the estimation results using the model 3 (log model) is the highest. The matching rate of the estimation results using Model 1 (encoder decoder model) was high, and the matching rate of the estimation results using Model 2 (Linear model) was the lowest.

  Moreover, FIG. 9 is a figure which shows the comparison result of the coincidence rate according to the number of clicks of the evaluation experiment. As shown in FIG. 9, in any click count, the matching rate of the estimation result using the model 3 is the highest, and then the matching rate of the estimation result using the model 1 is high, and the estimation using the model 2 is performed. The matching rate of the results was the lowest. It was found that even when the number of clicks is small (also in the so-called tail query), the matching rate of the estimation results using the model 3 is the highest.

  In the model 1 used in this evaluation experiment, the click frequency is not taken into consideration, and therefore the category with a large number of clicks and the category with a small number of clicks are processed in the same way, and therefore, it is vulnerable to noise included in the learning data. is assumed.

  Moreover, although the category estimated using the model 1 was different from the correct category in the search log, the category that may be determined to have been correctly estimated is included. FIG. 10 is a diagram illustrating an example of categories estimated using the model 1. As shown in FIG. 10, for example, for the query “honeymoon gift”, the category estimated using model 1 was a category related to kitchenware, whereas the correct category was related to a game. It was a category. This is due to the ambiguity of the query, and it can be considered that the estimation result of the model 1 may be determined to be correct.

  In addition, for example, for the query “enter large wallet passbook”, the category estimated using Model 1 was a category related to ladies fashion, while the correct category was a category related to men's fashion. there were. This is due to the existence of overlapping categories related to the product, and it can be considered that the estimation result of such a model 1 may be determined to be correct.

  For example, for the query “mobile phone cute strap”, the category estimated using model 1 was a category related to smartphones, whereas the correct category was a category related to ladies fashion. It was. This is due to the difficulty in selecting a category for the query, and it can be considered that the estimation result of the model 1 may be determined to be correct.

  In addition, as an example in which the estimation process of the model 1 is not determined to be correct, when the morphological analysis for the query by the query analysis unit 32 is not appropriate, or when a specific word included in the query has a large influence on the category estimation. In some cases, the relationship with Shikai and other words was not properly evaluated.

  In consideration of the category estimated using the model 1 as described above, although it is different from the “correct category” in the search log, it may be determined that the correct estimation process is performed. The coincidence rate of the category of the first layer in the case was 0.9238 (92.38 [%]). Considering this point, it can be determined that the accuracy of category estimation using the encoder / decoder model in the search server 10 of the present embodiment is high.

  Further, the search log coverage for the query in the estimation process using the model 3 was 0.3671 as a whole. For this reason, when the model 3 is used, an appropriate category may not be estimated if the query does not exist in the search log. On the other hand, in the estimation using the model 1, even if the query does not exist in the search log, the category can be estimated appropriately. For example, even if the query “men, sneakers, company B” is not learned, the query “men, sneakers, company A” is learned, and “company B” and “company B” If learning that “Company A” is a close word on the vector expression is made, it is possible to appropriately estimate the category for “Men, sneakers, Company B”.

  According to the search server 10 of this embodiment demonstrated above, the category linked | related with a query can be estimated appropriately, and search accuracy can be improved further. Further, according to the search server 10 of the present embodiment, it is possible to appropriately estimate a category even when an unknown query that has not been used for a search in the past is input.

  In addition, based on the encoding process result of the encoder 42, you may further provide the determination part which determines the break of the session of a group of access from the user who operates the terminal device T with respect to the search server 10. FIG. For example, when the position of the vector indicated by the encoding process result changes greatly in the vector coordinate space, it may be determined that the access session from the user has been switched.

<2. Second Embodiment>
Hereinafter, a second embodiment of the present invention will be described. The information processing apparatus (search server) according to the present embodiment differs from the first embodiment in that both the category estimation process based on the encoder / decoder model and the category estimation process based on log data are performed. For this reason, about the structure etc., the figure and related description which were demonstrated in 1st Embodiment are used, and detailed description is abbreviate | omitted.

<2-1. System configuration>
FIG. 11 is a functional block diagram of the search server 11 of this embodiment. The search server 11 further includes, for example, a query determination unit 46 and a second category estimation unit 48 (second estimation unit) in addition to the components of the search server 10 of the first embodiment.

  The query determination unit 46 refers to the search log stored in the storage unit 40 and determines whether or not the query input from the terminal device T exists in the past search log. In the search server 11, when the query determination unit 46 determines that the query input from the terminal device T does not exist in the past search log, the category estimation process based on the encoder / decoder model similar to the first embodiment is performed. . On the other hand, in the search server 11, when the query determination part 46 determines with the query input from the terminal device T existing in the past search log, the category estimation process based on log data is performed.

  For example, the second category estimation unit 48 refers to the search log data D3 stored in the storage unit 40 and estimates the category. For example, the second category estimation unit 48 selects the category having the largest number of clicks in the search log data D3 stored in the storage unit 40. FIG. 12 is a diagram illustrating an example of the search log data D3. In FIG. 12, when the query is “Ladies B company”, category 1 in the first hierarchy is “Ladies Fashion”, category 2 in the second hierarchy is “Wallet / Accessories”, and category 3 in the third hierarchy is The number of clicks (256999) in the category “wallet” is the largest. For this reason, when the query is “Ladies B company”, the second category estimation unit 48 sets the category 1 of the first hierarchy as “Ladies fashion” as the category associated with the query, and the category 2 of the second hierarchy. Is “wallet / accessory”, and category 3 in the third hierarchy estimates “wallet”.

<2-2. Category estimation processing>
Hereinafter, the category estimation process of the search server 11 will be described. FIG. 13 is a diagram illustrating an example of the flow of category estimation processing of the search server 11.

  First, the query analysis unit 32 acquires a query input from the terminal device T via the communication unit 30 (S301). Next, the query determination unit 46 refers to the search log stored in the storage unit 40, and determines whether or not the query input from the terminal device T exists in the past search log (S303).

  When the query determination unit 46 determines that the query input from the terminal device T exists in the past search log, the second category estimation unit 48 refers to the search log data D3 stored in the storage unit 40. The category is estimated (S305). The second category estimation unit 48 inputs the estimated category information to the search result generation unit 38.

  On the other hand, when the query determination unit 46 determines that the query input from the terminal device T does not exist in the past search log, the query analysis unit 32 analyzes the query input from the terminal device T to obtain the word level. Divide (S307). Next, the query conversion unit 34 converts each of the words divided by the query analysis unit 32 into a feature vector (S309).

  Next, the encoder 42 performs an encoding process on the feature vector of the first word among the feature vectors converted by the query conversion unit 34 (S311). Next, the encoder 42 determines whether or not the encoding process for the feature vectors of all words has been completed (S313).

  If the encoder 42 determines that the encoding process has not been completed for the feature vectors of all words, the feature vector of the second word and the first word of the feature vectors converted by the query converter 34 are displayed. The encoding process is performed again using the encoding process result for the feature vector (S311). Next, the encoder 42 determines whether or not the encoding process for the feature vectors of all words has been completed (S313). Thereafter, the same processing is repeated until the encoding processing for the feature vectors of all words is completed. When the encoder 42 determines that the encoding process has been completed for the feature vectors of all words, the encoder 42 inputs the encoding process result obtained as a result of repeating the above-described encoding process to the decoder 44.

  Next, the decoder 44 performs a decoding process on the encoding process result input from the encoder 42 (S315). Next, the decoder 44 determines whether or not the estimation of the lowermost category has been completed (S317). When the decoder 44 determines that the estimation of the lowermost category has not been completed, the decoder 44 performs the decoding process again using the estimated category information and the decoding process result (S315). Next, the encoder 42 determines whether or not the estimation of the lowermost category is completed (S317). Thereafter, the same processing is repeated until the estimation of the lowermost category is completed. When the decoder 44 determines that the estimation of the lowermost category has been completed, the decoder 44 inputs information on the estimated category to the search result generation unit 38.

  Next, the search result generation unit 38 generates a search result page as a search result based on the category estimation result input from the decoder 44 or the second category estimation unit 48, and transmits it to the terminal device T (S319). ). Thus, the process of this flowchart is completed.

  According to the search server 11 of this embodiment demonstrated above, the category linked | related with a query can be estimated appropriately, and search accuracy can be improved further. Moreover, according to the search server 11 of this embodiment, when the query input from the terminal device T does not exist in the past search log, the category estimation process based on the encoder / decoder model is performed, and the query is input from the terminal device T. If the query exists in the past search log, the category estimation accuracy based on the log data can be performed to further improve the category estimation accuracy.

<3. Third Embodiment>
Hereinafter, a third embodiment of the present invention will be described. The information processing apparatus (search server) according to this embodiment is different from the first embodiment in that it includes one encoder and a plurality of estimation units (decoders). For this reason, about the structure etc., the figure and related description which were demonstrated in 1st Embodiment are used, and detailed description is abbreviate | omitted.

<3-1. System configuration>
FIG. 14 is a functional block diagram of the search server 12 of the present embodiment. Instead of the category estimation unit 36 of the search server 10 of the first embodiment, the search server 12 includes, for example, one encoder 50 and three estimation units (category estimation unit 52, price range estimation unit 54, brand estimation unit 56). ).

  The encoder 50 performs an encoding process on at least one feature vector input from the query conversion unit 34 using the learning data and using an encoding model. When two or more feature vectors are input from the query converter 34, the encoder 50 performs an encoding process on the first feature vector, and then encodes the result of the first feature vector and 2 Encoding processing is performed using the first feature vector. The encoder 42 inputs the encoding process result obtained by repeating the encoding process to the category estimation unit 52, the price range estimation unit 54, and the brand estimation unit 56.

  The category estimation unit 52 performs a decoding process on the encoding process result input from the encoder 50 using a decoding model that has been learned using learning data, and estimates a category to be associated with the query word. To the search result generation unit 38.

  The price range estimation unit 54 performs a decoding process on the encoding process result input from the encoder 50 using a decoding model learned using the learning data, and a product or service to be associated with the query word The price range is estimated and input to the search result generation unit 38. The learning data of the price range estimation unit 54 includes, for example, a “word feature vector” included in the query used for the search and a “product or service price” selected by the user on the search result page provided in response to the query. ”Is used.

  The brand estimation unit 56 performs decoding processing on the encoding processing result input from the encoder 50 using the decoding model that has been learned using the learning data, and the product or service to be associated with the query word. The brand is estimated and input to the search result generation unit 38. The learning data of the brand estimation unit 56 includes, for example, a “word feature vector” included in the query used for the search and a “product or service brand” selected by the user on the search result page provided in response to the query. Use the pair.

  The search result generation unit 38 is based on the category estimation result input from the category estimation unit 52, the price range estimation result input from the price range estimation unit 54, and the brand estimation result input from the brand estimation unit 56. Then, a search result page as a search result is generated and transmitted to the terminal device T.

  According to the search server 12 of this embodiment demonstrated above, the category linked | related with a query can be estimated appropriately, and search accuracy can be improved further. Further, according to the search server 12 of the present embodiment, the category estimation result input from the category estimation unit 52, the price range estimation result input from the price range estimation unit 54, and the brand estimation unit 56 input Since a search result page that is a search result is generated based on the brand estimation result, a search result page according to the user's intention can be provided.

  In addition to or instead of the category estimation unit 52, the price range estimation unit 54, and the brand estimation unit 56, an estimation unit that estimates other features may be provided.

  According to the embodiment described above, the query used for the search is analyzed, the analysis unit that divides the query into words, the conversion unit that converts the words divided by the analysis unit into feature vectors, A first estimation for estimating a category associated with the query used in the search by learning a relationship between a feature vector of a word included in the query in the learning data and a category associated with the query in the learning data. The search accuracy can be improved.

  As mentioned above, although the form for implementing this invention was demonstrated using embodiment, this invention is not limited to such embodiment at all, In the range which does not deviate from the summary of this invention, various deformation | transformation and substitution Can be added.

10, 11, 12 ... search server (information processing apparatus)
30 ... Communication unit 32 ... Query analysis unit (analysis unit)
34 ... Query conversion unit (conversion unit)
36 ... category estimation unit (first estimation unit, learning unit)
38 ... Search result generation unit (generation unit)
40: Storage unit 42 ... Encoder 44 ... Decoder 46 ... Query determination unit (determination unit)
48 ... Second category estimation unit (second estimation unit)
DESCRIPTION OF SYMBOLS 50 ... Encoder 52 ... Category estimation part 54 ... Price range estimation part 56 ... Brand estimation part

Claims (14)

An analysis unit that analyzes the query used for the search and divides the query into words;
A conversion unit that converts the words divided by the analysis unit into feature vectors;
A first estimation for estimating a category associated with the query used in the search by learning a relationship between a feature vector of a word included in the query in the learning data and a category associated with the query in the learning data. An information processing apparatus comprising: The first estimation unit includes
An encoder that performs an encoding process on the feature vector of the word;
A decoder that performs a decoding process on the processing result of the encoder and outputs a category associated with the query used for the search;
Comprising
The information processing apparatus according to claim 1. The encoder performs a first encoding process on a first feature vector among a plurality of feature vectors of the word, and then uses a result of the first encoding process and a second feature vector to perform a second Encoding process,
The information processing apparatus according to claim 2. The decoder performs a first decoding process on the processing result of the encoder to output a first category, performs a second decoding process on the result of the first decoding process, and performs a second decoding process. Output categories,
The information processing apparatus according to claim 2 or 3. The category has a tree structure;
The first category is a category defined at the highest level in the tree structure;
The second category is a category defined under the first category.
The information processing apparatus according to claim 4.   The information processing according to claim 1, further comprising: a generation unit that generates a search result page for the query used for the search based on the category estimated by the first estimation unit. apparatus.   The information processing apparatus according to claim 1, wherein the analysis unit analyzes the query used for the search by morphological analysis and divides the query into words. A learning unit that learns the relationship between the feature vector of the word of the query used for the search and the category selected by the user in the search result page for the query used for the search;
The learning unit excludes, from the learning data, a category in which the number of times selected by the user is equal to or less than a predetermined threshold value.
The information processing apparatus according to any one of claims 1 to 7. A second estimation unit configured to estimate a category associated with the query used for the search based on a log of the category selected by the user in the search result page for the query used for the search;
The information processing apparatus according to any one of claims 1 to 8. A determination unit that determines whether the query used for the search exists in the log;
When the determination unit determines that the query used for the search exists in the log, the second estimation unit estimates a category associated with the query used for the search,
When the determination unit determines that the query used for the search does not exist in the log, the first estimation unit estimates a category associated with the query used for the search.
The information processing apparatus according to claim 9. The first estimation unit includes
An encoder that performs an encoding process on the feature vector of the word;
A plurality of decoders for performing decoding processing on the processing result of the encoder;
Comprising
The information processing apparatus according to claim 1. An encoder that performs an encoding process on the feature vector of the word;
A determination unit that determines a break in a session of a group of accesses from a user to the information processing device based on a processing result of the encoder;
Further comprising
The information processing apparatus according to claim 1. Computer
Analyzing the query used for the search, breaking the query into words,
Converting the divided words into feature vectors;
Information processing for estimating a category associated with a query used in the search by learning a relationship between a feature vector of a word included in the query in the learning data and a category associated with the query in the learning data Method. On the computer,
Analyzing the query used for the search, splitting the query into words,
Converting the divided words into feature vectors;
A program for estimating a category associated with a query used in the search by learning a relationship between a feature vector of a word included in the query in learning data and a category associated with the query in the learning data.

Images