JP6723182B2 - Information processing apparatus, information processing method, and program - Google Patents

Description

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

2. Description of the Related Art In the field of electronic transactions, there is known a technology for ranking products displayed based on a query searched by a user according to the number of clicks (see Patent Document 1).

JP, 2005-179461, A

In this type of product search, the tendency of clicks and purchases after the search differs depending on the nature of the request according to the user's attribute, the category of the product being searched, and the like. The technology described in Patent Document 1 cannot estimate the category of the product corresponding to the query.

The present invention has been made in view of such circumstances, and an object thereof is to provide an information processing device, an information processing method, and a program capable of estimating a product category corresponding to a query. ..

One aspect of the present invention is the information of a query input to a product search page of a business that conducts electronic transactions, and the product clicked on a product recommendation page provided in response to the input of the query. A category estimation unit that refers to the acquisition unit that acquires information and the information acquired by the acquisition unit, and that estimates the product category corresponding to the query based on the relevance between the query and the clicked product. An information processing apparatus including:

According to one aspect of the present invention, it is possible to estimate a product category corresponding to a query.

It is a block diagram of the shopping server 1 of embodiment. It is a figure which shows an example of the content of the store information 51 of embodiment. It is a figure which shows an example of the content of the log information 52 of embodiment. It is a figure which shows an example of the product category estimation result 54 of embodiment. It is a figure which shows the state in which a product category specialization model is constructed|assembled from the ranking learning data 55 of embodiment. It is a flow chart which shows a flow of processing in query learning of shopping server 1 of an embodiment. It is a flowchart which shows the flow of a process in the operation stage of the shopping server 1 of embodiment. It is a block diagram of the shopping server 2 which is a modification of embodiment. It is a block diagram of the shopping server 3 which is a modification of embodiment.

Hereinafter, embodiments of an information processing apparatus, an information processing method, and a program of the present invention will be described with reference to the drawings. The information processing device is realized by one or more processors. The information processing device is information on a query input to a product search page of a business operator that conducts electronic transactions via a network, and a product clicked on a product recommendation page provided in response to the input of the query. Is acquired, and the product category corresponding to the query is estimated based on the relationship between the query and the product obtained from the acquired information.

The business operator that conducts electronic transactions includes a business operator who provides a product sales page in response to a request from a business operator who operates an electronic store, a business operator who operates an electronic store, and the like. In the present specification, “product” includes “service”. The product sale page includes a product search page for the user to search for products and a product recommendation page for displaying recommended products displayed based on the query searched by the user.

The information processing device may be a web server that provides a web page in response to a request from a browser executed in the terminal device, or one of the application pages in response to a request from an application program executed in the terminal device. It may be an application server that provides all or part. In the following description, the information processing device is assumed to be a web server.

FIG. 1 is a configuration diagram of a shopping server 1 to which the information processing apparatus of the embodiment is applied. In the example of FIG. 1, the information processing device is included in the shopping server 1, but the information processing device and the shopping server 1 may be separate devices as described later.

The shopping server 1 includes, for example, a site providing unit 10, a log collecting unit 20, a category estimating unit 30, a ranking unit 40, an interface unit IF, and a storage unit 50.

Some or all of the site providing unit 10, the log collecting unit 20, the category estimating unit 30, and the ranking unit 40 are programs in which a shopping server processor such as a CPU (Central Processing Unit) is stored in the storage unit 50. It is realized by executing. Some or all of these may be realized by hardware such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), and FPGA (Field-Programmable Gate Array), or software and hardware. May be realized by the cooperation of

The interface unit IF includes, for example, a NIC (Network Interface Card). The shopping server 1 communicates with the user terminal 80 via the network NW using the interface unit IF. The network NW includes, for example, a wireless base station, Wi-Fi access point, communication line, provider, Internet and the like.

The user terminal 80 is, for example, a mobile phone such as a smartphone, a tablet terminal, or a computer device (communication device) such as a personal computer. The user terminal 80 transmits a request to the shopping server 1 via the network NW when a UA (User Agent) such as a web browser is activated and the user performs a predetermined operation. Then, the user terminal 80 generates a web screen based on the web page returned from the shopping server 1, and displays the web screen on the display unit. The request transmitted by the user terminal 80 includes information on the search query entered by the user in the search field on the product sales page.

The storage unit 50 is realized by, for example, a RAM (Random Access Memory), a ROM (Read Only Memory), an HDD (Hard Disk Drive), a flash memory, or a hybrid storage device in which a plurality of these are combined. Further, part or all of the storage unit 50 may be an external device accessible by the shopping server 1, such as a NAS (Network Attached Storage) or an external storage server.

The storage unit 50 stores, for example, store information 51, log information 52, query learning data 53, and ranking learning data 55. The store information 51 is information on stores that are opened by businesses that conduct electronic transactions. FIG. 2 is a diagram showing an example of the contents of the store information 51. The store information 51 is, for example, information in which product information, inventory information, a delivery period, and the like are associated with store identification information (store ID). The store information is provided by the store manager.

The log information 52 includes, for example, a history of the user accessing the shopping server 1 using the user terminal 80. FIG. 3 is a diagram showing an example of the contents of the log information 52. The log information 52 is, for example, information in which information such as a query searched by the user, a product clicked by the user, and a URL of a page browsed by the user is associated with the identification information (user ID) of the user.

The query learning data 53 is data of a combination of a query input by the user and a product clicked on the product recommendation page presented corresponding to the query.

The ranking learning data 55 is data for learning an effective model for each product category. Is data including various information such as the PV (page view) of the recommended product page, the number of images, and the like.

The site providing unit 10 builds a product sales page (shopping site) based on the store information 51 and the query learning data 53 stored in the storage unit 50, and provides the product sales page (shopping site) to the user terminal 80 via the network NW. The product sales page includes pages of various layers. For convenience, pages focused on products corresponding to the query, which are provided according to the search query, are referred to as product recommendation pages, and pages other than the product recommendation page (including, for example, the top page of the product sales page) are referred to as product search pages. It should be noted that the product recommendation page often has a search field, but in the present specification, the above definition will be used for explanation.

The log collection unit (acquisition unit) 20 collects a history (log) of the user accessing the shopping server 1 using the user terminal 80. The log collection unit 20 may also collect history of access to devices other than the shopping server 1. In this case, the log collection unit 20 collects a user's access history to devices other than the shopping server 1 by using a cookie. The log information 52 also includes information on the query input to the product search page.

The category estimation unit 30 refers to the log information 52, and based on the relevance of the query and the product clicked among the plurality of products displayed on the product recommendation page by inputting the query, Estimate the product category that corresponds to the query.

The details of the processing by the category estimation unit 30 will be described below. The category estimation unit 30 refers to the log information 52 at an arbitrary timing and learns the relationship between the query and the product category set in advance for the product clicked after the query is input. The category estimation unit 30 generates a query vector from a query, for example, and generates and uses the query vector as a sparse vector (a vector in which only a few of the many elements have 1).

For example, it is assumed that the product A (category: home appliance) is clicked 100 times, the product B (category: home electric appliance) is clicked 50 times, and the product C (category: book) is clicked 20 times in response to the query “TV”. In this case, the category estimation unit 30 may determine the product category corresponding to the query “TV” to be the “appliances” with the largest number of appearances, or may score the product category according to the number of appearances. In the case of scoring, the category estimation unit 30 gives a category score such as 80 for “home appliances” and 20 for “books”.

FIG. 4 is a diagram showing an example of the product category estimation result 54. In the following description, it is assumed that the category estimation unit 30 adds a category score to the query, instead of narrowing it down to one product category.

When an unknown query whose relationship with the product category is unlearned is input, the category estimation unit 30 vectorizes the unknown query and creates it from other known queries that are similar to the query vector. The searched query vector may be searched, and the already-learned product category may be used for the query corresponding to the found known query vector. Alternatively, a known query that is similar to the input query and whose product category is estimated may be searched for by partial matching of words or rearrangement of word order.

It should be noted that, when the log information 52 related to the user's operation is small (such as when a new store opens), the product category and the category score are set in advance for the query.

The details of the processing by the ranking unit 40 will be described below. The ranking unit 40 ranks the products to be displayed on the product recommendation page using a model (product category specialized model) according to the product category estimated by the product category estimation unit 30 for each query.

FIG. 5 is a diagram showing a state in which a product category specialized model is constructed from the ranking learning data 55. The ranking unit 40 divides the ranking learning data 55 for each product category estimated by the category estimation unit 30 at an arbitrary timing.

The ranking unit 40 constructs a product category specialized model using the divided data as teacher data. Here, the model includes, for example, a set of weighting parameters that defines which of the features used for ranking has a higher specific gravity. The feature includes various information such as the PV of the recommended product page, the number of images, the point reduction rate, the price, the sales record, and the manufacturer name. The category-specific model is a model in which different weighting parameters are set for each product category.

The ranking unit 40 learns the weighting parameter having the highest index value such as CTR by machine learning based on the divided ranking learning data 55.

For example, according to the product category specialized model corresponding to the product category “home appliances”, it is assumed that a high weighting parameter is given to PV, price, sales record, maker name, etc. among the features.

Further, according to the product category specialized model in which the product category corresponds to, for example, “book”, for example, it is assumed that a high weighting parameter is given to the point reduction rate among the features.

In addition to this, for example, according to the product category specialized model corresponding to “fashion”, it is assumed that a high weighting parameter is given to the number of images related to the product on the product recommendation page among the features.

The ranking unit 40 selects and ranks products by any of the following methods, for example.

(1) The product score is calculated using the product category specialized model corresponding to the product category to which the maximum category score is given in the product category estimation result 54.

(2) A product score is calculated for each product category using one or more product category specialized models corresponding to one or more product categories to which a non-zero category score is assigned in the product category estimation result 54, and the category is calculated. The final product score is calculated by calculating the weighted sum using the score as a weight.

(3) Using one or more product category specialized models corresponding to one or more product categories to which a non-zero category score is assigned in the product category estimation result 54, a product score is calculated for each product category, and the category is calculated. A product corresponding to each product category is selected at a ratio according to the score.

The ranking unit 40 may obtain and use the PV for each query when using the PV. For example, the PV of the product for the query “TV” may be a value excluding the PV due to being displayed according to another query. The ranking unit 40 may also generate different product category specialized models for each user.

Further, the ranking unit 40 may rank a plurality of products in the product category based on data preset in the ranking learning data 55 at an initial stage where learning has not progressed.

When the ranking for each product category is determined in this way, the site providing unit 10 selects a predetermined number of products according to the rank determined by the ranking unit 40 for the input query and recommends the product. Display on page.

Next, the operation of the shopping server 1 will be described.

FIG. 6 is a flowchart showing the flow of processing in the query learning of the shopping server 1. The log collection unit 20 acquires the log information 52 (S100).

The category estimation unit 30 refers to the log information 52 to learn the relationship between the input query and the product category set in advance for the product clicked after the query is input, and the product corresponding to the query is learned. The category is estimated (S110).

The processing in the learning stage of FIG. 6 does not necessarily have to be performed each time a query is input, and the category estimation unit 30 determines the product category according to the query at a desired timing, such as when certain data is accumulated. It may be estimated.

FIG. 7 is a flowchart showing the flow of processing in the operation stage of the shopping server 1. The user accesses the product sales page on the user terminal 80, inputs a query for searching for products on the product search page, and instructs the search (S120). The ranking unit 40 acquires the product category corresponding to the query from the product category estimation result 54 (S130).

The ranking unit 40 selects a product to be displayed on the product recommendation page from a plurality of products belonging to the estimated product category using the product category specialized model corresponding to the acquired product category, and ranks the products. (S140).

According to the ranking of the ranking unit 40, the site providing unit 10 displays a plurality of products within the product category in a ranked state on the product recommendation page for each product category (S150).

Further, among the processes in the operation stage of FIG. 7, the processes of S130 and S140 do not necessarily have to be performed each time a query is input, and once a product is selected and ranked for a certain query, the ranking is changed. The assigning unit 40 may select and rank the products after a certain amount of data is accumulated.

As described above, according to the shopping server 1, by learning the product category related to the query, it is possible to estimate the product category that is highly relevant to the input query. As a result, the product category is assigned based on the actual behavior of the user, as compared to the case where the product category is fixedly assigned to the query, so a query that applies to multiple product categories is entered. Even then, product categories can be assigned based on the click tendencies of the majority of users. In other words, it is possible to allocate the product category that reflects the actual situation. Further, according to the shopping server 1, since the product category can be learned based on the user's click action, the product can be appropriately presented to the user in accordance with the user's preference.

Further, according to the shopping server 1, by ranking products by using different category-specific models for each product category, the products are appropriately ranked according to the property of the product category related to the input query. Therefore, the user can be encouraged to purchase the product.

Although the category estimation unit 30 and the ranking unit 40 are described as internal functions of the shopping server 1 in the above embodiment, one or both of them may be a function of an external device (information processing device). Good.

[Modification 1]
In the first modification, for example, a case where the category estimation unit 30 and the ranking unit 40 are included in the information processing device 90 will be illustrated.

FIG. 8 is a configuration diagram of the shopping server 2 which is a modified example. The information processing device 90 includes a category estimation unit 30, a ranking unit 40, and a storage unit 60 that stores the product category estimation result 54, and is connected to the shopping server 2 via the network NW.

In the information processing device 90, the category estimation unit 30 acquires the log information 52 via the network NW, learns the relationship between the query and the product category in the same manner as described above, and estimates the product category according to the query. The category estimation unit 30 updates the product category estimation result 54 in the information processing device 90.

The ranking unit 40 refers to the product category estimation result 54, performs selection/ranking for each query, and transmits the selection/ranking to the site providing unit 10.

Based on the result of the ranking unit 40 acquired from the information processing device 90, the site providing unit 10 displays a plurality of products in the product category in a ranked state on the product recommendation page for each product category.

[Modification 2]
The modification 2 is different from the modification 1 in that the category estimation unit 30 is in the information processing device 100, but the ranking unit 40 is in the shopping server 3.

FIG. 9 is a configuration diagram of a shopping server 3 which is a modified example. The information processing apparatus 100 includes, for example, a category estimation unit 30 and is connected to the shopping server 3 via the network NW.

In the information processing device 100, the category estimation unit 30 acquires the log information 52 via the network NW, learns the relationship between the query and the product category in the same manner as above, and estimates the product category according to the query. The category estimation unit 30 transmits the estimation result to 1. 1 stores the estimation result in the storage unit 50 as the product category estimation result 54.

The ranking unit 40 refers to the product category estimation result 54 and performs selection/ranking for each query.

Based on the result of the ranking unit 40 acquired from the information processing device 100, the site providing unit 10 displays a plurality of products in the product category in the product recommendation page for each product category in a ranked state.

According to each modification of the above-described information processing device, a shopping site can be provided with various device configurations via the network NW.

As described above, the embodiments for carrying out the present invention have been described using the embodiments, but the present invention is not limited to such embodiments, and various modifications and substitutions are made within the scope not departing from the gist of the present invention. Can be added. For example, the above product category estimation may be applied to business scenes other than shopping sites, such as auctions.

1, 2, 3... Shopping server 10... Site providing unit 20... Log collecting unit 30... Category estimating unit 40... Ranking unit 80... User terminals 90, 100... Information processing device IF... Interface unit NW... Network

Claims (5)

An acquisition unit that acquires information about a query input to a product search page of a business that conducts electronic transactions and information about a product clicked on a product recommendation page provided in response to the input of the query. When,
With reference to the information acquired by the acquisition unit, based on the relevance between the query and the clicked product, a category estimation unit that estimates a product category corresponding to the query,
Using a different method for each product category estimated by the category estimation unit, a ranking unit that ranks the products on the product recommendation page,
Equipped with
The ranking unit is
Using the product category specialized model constructed by machine learning with the data divided for each product category as teacher data, the products are ranked,
The product category specialization model includes a set of weight parameters that define which of the features to be used for ranking products are weighted more heavily,
The feature includes at least a page view of a product recommendation page, and the page view included in the feature is a value aggregated for each query,
Information processing device. The category estimation unit estimates a product category corresponding to the query, based on category information given in advance to the clicked product,
The information processing apparatus according to claim 1. The category estimation unit updates the product category based on the query and the number of clicks on products displayed on the product recommendation page corresponding to the query,
The information processing apparatus according to claim 1 . Computer
Acquiring the information of the query input to the product search page of the business operator conducting the electronic transaction, and the information of the product clicked on the product recommendation page provided in response to the input of the query,
Referring to the obtained information, based on the relevance of the query and the clicked item, estimates the product category that corresponds to said query,
Using a method that is different for each of the estimated product categories, rank the products on the product recommendation page,
When performing the ranking, the products are ranked using a product category specialized model constructed by machine learning with the data divided for each product category as teacher data,
The product category specialization model includes a set of weight parameters that define which of the features to be used for ranking products are weighted more heavily,
The feature includes at least a page view of a product recommendation page, and the page view included in the feature is a value aggregated for each query,
Information processing method. On the computer,
The information of the query input to the product search page of the business operator conducting the electronic transaction and the information of the product clicked on the product recommendation page provided in response to the input of the query are acquired,
With reference to the acquired information, the product category corresponding to the query is estimated based on the relevance between the query and the clicked product,
Using a method that is different for each of the estimated product categories, rank the products on the product recommendation page,
When the ranking is performed, the ranking of the products is performed using a product category specialized model constructed by machine learning with the data divided for each product category as teacher data,
The product category specialization model includes a set of weight parameters that define which of the features to be used for ranking products are weighted more heavily,
The feature includes at least a page view of a product recommendation page, and the page view included in the feature is a value aggregated for each query,
program.

Images