JP7413423B2 - Information processing system and information processing method - Google Patents

Description

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

Conventionally, segment analysis devices generate segment information indicating the behavior of visitors to a website, perform analysis using segment information that matches analysis conditions, and display the analysis results on a display unit. It has been disclosed (for example, see Patent Document 1).

Japanese Patent Application Publication No. 2013-235572

However, with the above-mentioned conventional technology, there are cases where it is not possible to obtain information that meets the user's intention.

The present invention has been made in consideration of such circumstances, and one of the objects is to provide an information processing system, an information processing method, and a program that can acquire information in accordance with the user's intention. Let's do one.

In one aspect of the present invention, an information processing system includes an acquisition unit that acquires a query specified by a user, a query acquired by the acquisition unit, and a name that includes the query or a similar name. one or more features associated with the introduction target, and one or more indicators based on the features included in behavioral information indicating the user's behavior toward the content when content including the introduction target is provided to the user; , a determining unit that determines the ranking of introduction targets for the query by inputting to a trained model that has been trained in advance.

According to one aspect of the present invention, information that meets the user's intention can be provided.

1 is a diagram showing an example of a functional configuration of an information processing system 1. FIG. 6 is a diagram showing an example of the contents of product information 62. FIG. 6 is a conceptual diagram of a learned model 64. FIG. FIG. 3 is a diagram for explaining a validation function. FIG. 2 is a diagram for explaining processing executed by the information processing system 1. FIG. FIG. 7 is a diagram for explaining processing executed by the information processing system 1 of the second embodiment.

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

[overview]
Information processing system 1 includes a first device and a second device. The first device and the second device are each realized by one or more processors. The first device and the second device may be one device. The first device acquires behavioral information indicating the user's behavior with respect to content including the introduction target and includes the introduction target and a feature used for ranking in other devices, and transmits the acquired behavioral information to the other device. Provided to. The second device includes a storage unit that stores information in which the introduced subject and the background provided by the first device are associated, an acquisition unit that acquires a query specified by the user, and a query that is acquired by the acquisition unit. By inputting one or more indicators based on a query and one or more features associated with an introduction target having a name that includes the query or a name similar to the query into a trained model that has been trained in advance, and a determining unit that determines the ranking of introduction targets for the query.

The introduction target is, for example, a product or a service. The background includes an aggregated background in which the user's actions are aggregated. This aggregated feature is the number of times the user accessed content focusing on the introduction target (for example, content introducing a product or content showing details of a product).

The trained model identifies referral targets with a high conversion rate based on the features included in the behavioral information acquired by the first device (for example, aggregated features) and the conversion results for the introduced targets in the content included in the behavioral information. This is a model trained to rank high in the ranking. Conversion refers to the fact that the user has paid attention to the introduced item, and is, for example, information that indicates that the user has added the introduced item to the cart or purchased it.

<First embodiment>
[composition]
FIG. 1 is a diagram showing an example of the functional configuration of an information processing system 1. As shown in FIG. The information processing system 1 includes, for example, a terminal device 10, a shopping server 30, a search device 50, and a processing system 100.

The first storage device 120 included in the processing system 100 is an example of a "first device," and the search device 50 is an example of a "second device." The first information processing device included in the processing system 100 is an example of a “processing device”.

The terminal device 10 and the shopping server 30 communicate with each other via the network NW. The network NW includes, for example, the Internet, a wireless base station, a provider, and the like. Furthermore, the shopping server 30 and the search device 50 or the processing system 100 (first storage device 120) are connected to a predetermined communication network (or network NW) such as a WAN (Wide Area Network), a LAN (Local Area Network), or a dedicated line. ).

[Terminal device]
The terminal device 10 is a device used by a user, and is, for example, a mobile phone such as a smartphone, a tablet computer, a notebook computer, a desktop computer, or the like. In the terminal device 10, a UA (User Agent) such as a browser or an application program is activated and performs the following operations. When a user operates the terminal device 10 to access the shopping server 30 and input a query, the terminal device 10 requests the shopping server 30 to search for products related to the input query. The terminal device 10 acquires the content provided by the shopping server 30 and displays the content on its own display unit. By operating the content, the user can view the details of the desired product or perform the process of purchasing the product.

[Shopping server]
In response to a request from the terminal device 10, the shopping server 30 requests the search device 50 to transmit product ranking results. The shopping server 30 provides the terminal device 10 with content including the ranking results provided by the search device 50.

[Search device]
The search device 50 includes, for example, an acquisition section 52, a processing section (determination section) 54, and a storage section 60. The acquisition unit 52 and the processing unit 54 are realized by a hardware processor such as a CPU (Central Processing Unit) executing a program stored in a storage device. Further, these functional units may be realized by hardware such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), GPU (Graphics Processing Unit), etc. It may also be realized by cooperation of software and hardware. Further, the above program may be stored in advance in a storage device or in a removable storage medium such as a DVD or CD-ROM, and the storage medium is loaded into the drive device of the search device 50. It may also be installed on a storage device.

The storage unit 60 is realized by, for example, a ROM (Read Only Memory), a flash memory, an SD card, a RAM (Random Access Memory), an HDD (Hard Disc Drive), a register, or the like. Further, part or all of the storage unit 60 may be a NAS (Network Attached Storage), an external storage server device, or the like. The storage unit 60 stores, for example, product information 62 and a learned model 64. The model is a predetermined function or a predetermined model such as a neural network.

The acquisition unit 52 acquires information provided by the shopping server 30 or the processing system 100. For example, the acquisition unit 52 acquires product information 62 from the processing system 100. The product information 62 is information provided by the second information processing device 150 included in the processing system 100 , and is information acquired from the shopping server 30 by the first storage device 120 included in the processing system 100 .

FIG. 2 is a diagram showing an example of the contents of the product information 62. The product information 62 is, for example, information in which a product ID, a product name, a product category, and a background (background information) are associated with each other. Features include, for example, product attributes or characteristics such as the product price, the number of PVs (Page Views) in which content containing the product was viewed, degree of similarity between the query and the product name, and information about the user (for example, demographic data). etc. For example, the processing unit 54 may obtain the above similarity by calculating a cosine similarity or the like based on the target query and the product name, or the similarity between each query and the product name may be determined in advance. It may be associated with product information 62. For example, PV is an example of aggregation features. The aggregated features may be aggregated data or statistical data other than PV. Further, the product information 62 may include conversion results.

The acquisition unit 52 also acquires the learned model 64 from the learning device 200 of the processing system 100.

The processing unit 54 generates product ranking results in response to a request from the shopping server 30, and provides the generated ranking results to the shopping server 30. The processing unit 54 inputs an index (for example, information obtained by vectorizing or normalizing the query or the feature) based on the query provided to the shopping server 30 and the feature corresponding to the query into the trained model 64, and The score of the product output by the model 64 is obtained.

FIG. 3 is a conceptual diagram of the learned model 64. The trained model 64 is a model that, when the above-mentioned index is input, outputs a score so that a product with a high conversion result has a high score. The trained model 64 is a model generated by the learning device 200 described later.

[Processing system]
Returning to the explanation of FIG. The processing system 100 includes, for example, a first storage device 120, a first information processing device 130, a second storage device 140, a second information processing device 150, and a learning device 200.

The first storage device 120 acquires the user behavior information acquired by the shopping server 30. Behavioral information is, for example, the user's behavior with respect to content including ranking results, and includes, for example, information about products, information (PV) indicating whether or not the content of the product has been viewed, and conversion results. . The behavior information is, for example, information that includes conversion results in addition to the contents of the product information 62 shown in FIG. 3 described above.

Behavior information is a data set in which features are organized according to predetermined rules (expressed in a predetermined format). Behavior information is information that is generated so that data can be searched by specifying predetermined conditions and characteristics. The predetermined rules are rules that depend on a programming language such as SQL. For example, the processing unit of the first storage device 120 aggregates information obtained from the shopping server 30 or performs statistical processing to generate behavioral information.

The first information processing device 130 acquires information included in the behavior information stored in the first storage device 120 at a predetermined period, and provides the acquired information to the second storage device 140 in order to provide it to the search device 50. do. For example, the first information processing device 130 provides the second storage device 140 with information or behavior information corresponding to the product information 62.

The first information processing device 130 has a filtering function. The filtering function is a function in which the first information processing device 130 excludes background information that satisfies a predetermined condition and provides the excluded background information to the second storage device 140. The predetermined conditions include background information of users who repeatedly access the shopping server 30 in large quantities, such as bots, and background information of users who access the shopping server 30 at a predetermined frequency (high frequency). That's true. The first information processing device 130 also has a validation function, which will be described later.

The second storage device 140 acquires the background information provided by the first information processing device 130 and stores the acquired information.

The second information processing device 150 acquires store information provided by the manager of the virtual store opened in the shopping server 30. The store information is, for example, information about products sold at the virtual store, and is, for example, accompanying information such as product descriptions that accompany the products. The second information processing device 150 acquires the information stored in the second storage device 140, adds the above-mentioned accompanying information to the acquired information, and generates distribution information. Then, the second information processing device 150 converts the above distribution information into a format suitable only for processing by the search device 50, and provides the distribution information to the search device 50.

Thereby, the search device 50 can obtain the above-mentioned product information 62 based on the distribution information.

[Validation function]
The first information processing device 130 acquires behavioral information including the products provided by the first storage device 120 and aggregated features used for ranking in the search device 50 at a predetermined period, and transfers the acquired behavioral information to the second device. Provided to. At this time, the first information processing device 130 determines whether the aggregated features included in the behavior information provided by the first storage device 120 in the first cycle are the aggregated features obtained in the second cycle, which is earlier than the first cycle. If the deviation is more than a predetermined degree from the reference value derived based on the reference value, the behavior information is not provided to the second device.

For example, when providing background information regarding a predetermined product group to the search device 50, the first information processing device 130 is configured to detect a difference between the reference background information of the predetermined product group and the background information of the object to be provided by a predetermined degree or more. If there is, it is determined that there is an abnormality in the target's background information. In this case, the first information processing device 130 does not provide the target background information to the second storage device 140.

FIG. 4 is a diagram for explaining the validation function. In the illustrated example, the description will be made assuming that the background information is the PV number of product group A, but the background information processed by the validation function may be information different from the PV. For example, it may be aggregated data or statistical data other than PV. The reference PV is, for example, the median, average, or mode of the PV numbers of the product group A during a predetermined period. The target PV is the number of PVs in a target period (for example, a period with the same length as a predetermined period).

As shown in FIG. 4, if the number of target PVs is less than or greater than the reference number of PVs by a threshold value X or more, the number of target PVs is determined to be abnormal. The threshold value X is, for example, a value obtained based on the degree of variation in the number of PVs of the product group A, and is, for example, a standard deviation σ or multiple times σ.

As described above, the target PV in which the deviation is greater than or equal to the threshold value X may be an inaccurate PV. If inaccurate PV information is provided to the search device 50, the search device 50 uses this PV to obtain ranking results, and therefore may provide inaccurate ranking results to the user.

In contrast, in the present embodiment, as described above, the first information processing device 130 determines a PV that is estimated to be inaccurate as abnormal and does not provide this PV to the search device 50, so the inaccurate ranking Results are suppressed from being provided to users.

[Learning device]
Returning to the explanation of FIG. Learning device 200 includes, for example, a learning section 210 and a storage section 220. The learning unit 210 is realized, for example, by a hardware processor such as a CPU executing a program stored in a storage device. The storage unit 220 stores, for example, behavior information 222 and a learned model 224. The trained model 224 is a model generated by the learning unit 210 using the behavior information 222.

The learning unit 210 generates a learned model 224. The trained model 224 ranks referral targets with a high conversion rate based on the background information included in the behavioral information acquired by the first storage device 120 and the conversion results for the referral targets in the content included in the behavioral information. This is a model that has been trained to rank at the top.

For example, the learning unit 210 performs learning using learning data including background information and queries included in the behavior information 222 and teacher data including conversion results. The learning unit 210 adjusts the parameters included in the model so that when the background information and query are input to the model, products with high conversion results have a high score, and generates a trained model 224 with adjusted parameters. do. The generated trained model 224 is provided to the search device 50 and stored in the storage unit 60 of the search device 50 as a learned model 64.

[Summary of processing executed by the information processing system]
FIG. 5 is a diagram for explaining processing executed by the information processing system 1. The first storage device 120 acquires behavior information from the shopping server 30 (S1). Furthermore, the first storage device 120 provides the acquired behavior information to the learning device 200 (S2), and the learning device 200 generates the learned model 224 based on the behavior information. This trained model 224 is provided to the search device 50 (S3).

Next, the first storage device 120 provides the search device 50 with at least the background information associated with the product among the behavior information (S4). In response to a request from the shopping server 30, the search device 50 inputs the background information and the query specified by the user into the trained model 224, and ranks the products based on the score output by the trained model 224. The determined ranking result is provided to the client (for example, the shopping server 30) (S5).

As described above, the first storage device 120 provides similar background information to the learning device 200 and the search device 50. The learning device 200 generates a learned model 224 using this feature information, and the search device 50 inputs the feature information used in learning to the generated learned model 224 to obtain a score.

As described above, since the types of features used in the learning device 200 and the search device 50 are the same, the search device 50 can obtain a score with higher accuracy. Furthermore, when adding a feature to be included in the feature information, the first storage device 120 can easily add the feature to the feature information provided to the learning device 200 and the search device 50 to obtain a highly accurate score. A system that outputs is constructed.

According to the first embodiment described above, the information processing system 1 includes the first storage device 120 that provides behavior information to other devices, and the search device 50, and the search device 50 is configured to send a query and a query. By inputting one or more indicators based on one or more features associated with a referral target with a name containing or a similar name into a pre-trained trained model, the referral target in response to a query can be By determining the ranking, it is possible to obtain information that meets the user's intentions, and this information can be provided to the user. As a result, the user can obtain information consistent with his or her intention.

<Second embodiment>
The second embodiment will be described below. In the second embodiment, a date and time are associated with a feature, and the information processing system 1 performs various processes (for example, learning) with reference to the date and time. Hereinafter, differences from the first embodiment will be mainly explained.

In the second embodiment, the latest dates of some of the aggregated features acquired by the search device 50 may not be available. The search device 50 inputs only the part in which all dates are the same among the aggregation features for each product to the learned model 64 that has been trained with the aggregation features whose final date is at least the matching date. In the second embodiment, the date associated with the aggregated features used to generate the trained model 224 is the same as the date associated with the aggregated features input into the trained model 224 in the search device 50.

FIG. 6 is a diagram for explaining processing executed by the information processing system 1 of the second embodiment. The shopping server 30 and search device 50 handle hundreds of millions of products. Information (background information, etc.) regarding hundreds of millions of products stored in the first storage device 120 is provided to the search device 50 as appropriate, but as mentioned above, there are hundreds of millions of products and the amount of data is enormous. Therefore, information regarding the product is not provided to the search device 50 in real time, and a time lag may occur.

The first storage device 120 sequentially provides behavior information for each product group to the search device 50, and the search device 50 provides behavior information for the first product group on the first specific day and the behavior information for the second product group on the first specific day. Obtain group behavior information on different days. As shown in FIG. 6, for example, assuming that the current date is January 5th, the first storage device 120 has obtained behavior information from January 1st to January 4th. Specifically, the PV of product A from January 1st to January 4th and the PV of product B from January 1st to January 4th are obtained. The search device 50 obtains the PV of product A from January 1st to January 4th and the PV of product B from January 1st to January 2nd, but the time tag shows that the PV of product B is from January 3rd. We have not received the PV for product B from January 4th.

Assuming such a situation, the learning unit 210 generates a trained model 224 (trained model 64) using the PVs of products A and B from January 1st to January 2nd as learning data. When obtaining a score for product A or product B, the search device 50 inputs the PV of product A or product B from January 1st to January 2nd into the learned model 64, and uses the learned model 64 to Generate product rankings based on the output scores.

Assume that the trained model 64 (224) performs learning using the PVs of products A and B from January 1st to January 4th. In this case, the PV of product B from January 1 to January 2 inputted by the search device 50 and the PV of the learning data used by the trained model 64 will be different, which will be taken into consideration when generating the trained model 64. The learned model 64 will output a score based on the tendency of the PV that has not been updated.

On the other hand, in the information processing system 1 of this embodiment, a date is associated with background information. Then, the information processing system 1 generates a learned model 64 that is trained using the background information that has the same date as the background information that the search device 50 has, and uses the generated learned model 64 to generate a ranking. use As a result, more accurate product ranking results are generated.

According to the second embodiment described above, the search device 50 has the behavior information of the third introduction target on the second specific day and the third specific date, and the behavior information of the fourth introduction target on the second specific day. However, if it does not have the behavioral information of the fourth referral target on the third specific day, the background included in the behavioral information of the third referral target on the second specific day and the background included in the behavioral information of the fourth referral target By determining the ranking of the third introduction target and the introduction target using , it is possible to more accurately obtain information that meets the user's intention.

Although the mode for implementing the present invention has been described above using embodiments, the present invention is not limited to these embodiments in any way, and various modifications and substitutions can be made without departing from the gist of the present invention. can be added.

1‥Information processing system, 10‥Terminal device, 30‥Shopping server, 50‥Search device, 52‥Acquisition unit, 54‥Processing unit, 60‥Storage unit, 62‥Product information, 64‥Learned model, 100‥Processing System, 120..First storage device, 130..First information processing device, 140..Second storage device, 150..Second information processing device, 200..Learning device, 210..Learning unit, 220..Storage unit, 222..Action Information, 224‥Learned model

Claims (14)

an acquisition unit that acquires a query specified by a user;
When providing the user with content that includes the query acquired by the acquisition unit and one or more features associated with the referral target that has a name that includes the query or a similar name; and that includes the referral target; determining the ranking of introduction targets for the query by inputting one or more indicators based on features included in behavioral information indicating the user's behavior toward the content into a trained model that has been trained in advance; and,
The background includes an aggregated background in which the user's actions are aggregated;
A second device comprising the acquisition unit and the determination unit acquires behavior information including the introduction target provided by the first device at a predetermined period and the aggregated features used for ranking in other devices, and transmits the acquired behavior information to the second device, which includes the acquisition unit and the determination unit. further comprising a processing device for providing the two devices;
The processing device determines that the aggregated features included in the behavior information provided by the first device in the first cycle are derived based on the aggregated features obtained in the second cycle earlier than the first cycle. If the behavior information deviates from the value by a predetermined degree or more, the behavior information is not provided to the second device.
Information processing system. an acquisition unit that acquires a query specified by a user;
When providing the user with content that includes the query acquired by the acquisition unit and one or more features associated with the referral target that has a name that includes the query or a similar name; and that includes the referral target; determining the ranking of introduction targets for the query by inputting one or more indicators based on features included in behavioral information indicating the user's behavior toward the content into a trained model that has been trained in advance; and,
The background includes an aggregated background in which the user's actions are aggregated;
The aggregated features acquired by the second device including the acquisition unit and the determination unit may not have the latest dates,
The second device inputs only the part where all the dates are the same among the aggregated features for each introduction target to the learned model trained with the aggregated features whose final date is at least the aligned date.
Information processing system. an acquisition unit that acquires a query specified by a user;
When providing the user with content that includes the query acquired by the acquisition unit and one or more features associated with the referral target that has a name that includes the query or a similar name; and that includes the referral target; determining the ranking of introduction targets for the query by inputting one or more indicators based on features included in behavioral information indicating the user's behavior toward the content into a trained model that has been trained in advance; Department and
From among the backgrounds of the behavioral information, the background included in the behavioral information associated with the device that was accessing the providing server device that provides content including the introduction target to the user's terminal device with a frequency of a predetermined degree or more is determined. a processing device that excludes and provides the excluded features to a second device including the acquisition unit and the determination unit;
An information processing system equipped with. an acquisition unit that acquires a query specified by a user;
When providing the user with content that includes the query acquired by the acquisition unit and one or more features associated with the referral target that has a name that includes the query or a similar name; and that includes the referral target; determining the ranking of introduction targets for the query by inputting one or more indicators based on features included in behavioral information indicating the user's behavior toward the content into a trained model that has been trained in advance; Department and
A second device having a data set in which features are organized according to predetermined rules, extracting set features from the data set, and providing the extracted features to a second device including the acquisition unit and the determination unit. 1 device and
An information processing system equipped with. The trained model is configured to rank referral targets with a high conversion rate at the top of the ranking based on the characteristics included in the behavioral information and the conversion results for the referral targets in the content included in the behavioral information. A model trained to locate
The information processing system according to any one of claims 1 to 4 . The introduction target is a product or service;
The information processing system according to any one of claims 1 to 5 . The feature is the number of times the user accessed content focused on the introduction target;
The information processing system according to any one of claims 1 to 6 . The background includes information indicating the characteristics of the product or service to be introduced.
The information processing system according to any one of claims 1 to 7 . The feature includes information indicating the degree of similarity between the name of the product or service to be introduced and the query;
The information processing system according to any one of claims 1 to 8 . The above characteristics are
The number of times the user accessed content focused on the introduction target;
information indicating the characteristics of the product or service to be introduced; and information indicating the degree of similarity between the name of the product or service to be introduced and the query;
The information processing system according to any one of claims 1 to 9 . The computer that is the determining device that determines the ranking is
Get the query specified by the user,
The retrieved query and one or more features associated with an introduction target that has a name that includes the query or a similar name, and that is related to the content when the content that includes the introduction target is provided to the user. Determine the ranking of introduction targets for the query by inputting one or more indicators based on the background included in behavioral information indicating the user's behavior into a trained model that has been trained in advance,
The background includes an aggregated background in which the user's actions are aggregated;
The computer of the processing unit is
acquiring behavior information including the introduction target provided by the first device at a predetermined period and the aggregated features used for ranking in other devices, and providing the acquired behavior information to the determining device;
The aggregated features included in the behavior information provided by the first device in the first cycle are higher than the reference value derived based on the aggregated features obtained in the second cycle, which is earlier than the first cycle. If the deviation is greater than or equal to the above, the behavior information is not provided to the determination device;
Information processing method. The computer that is the determining device that determines the ranking is
Get the query specified by the user,
The retrieved query and one or more features associated with an introduction target that has a name that includes the query or a similar name, and that is related to the content when the content that includes the introduction target is provided to the user. Determine the ranking of introduction targets for the query by inputting one or more indicators based on the background included in behavioral information indicating the user's behavior into a trained model that has been trained in advance,
The background includes an aggregated background in which the user's actions are aggregated;
The aggregated features acquired by the determination device may not have the latest dates,
Inputting only the part where all dates are aligned among the aggregated features for each introduction target into the trained model trained with aggregated features whose final date is at least the aligned date;
Information processing method. The computer that is the determining device that determines the ranking is
Get the query specified by the user,
The retrieved query and one or more features associated with an introduction target that has a name that includes the query or a similar name, and that is related to the content when the content that includes the introduction target is provided to the user. Determine the ranking of introduction targets for the query by inputting one or more indicators based on the background included in behavioral information indicating the user's behavior into a trained model that has been trained in advance,
The computer of the processing unit is
From among the backgrounds of the behavioral information, the background included in the behavioral information associated with the device that was accessing the providing server device that provides content including the introduction target to the user's terminal device with a frequency of a predetermined degree or more is determined. excluding and providing the excluded features to the determining device;
Information processing method. The computer that is the determining device that determines the ranking is
Get the query specified by the user,
The retrieved query and one or more features associated with an introduction target that has a name that includes the query or a similar name, and that is related to the content when the content that includes the introduction target is provided to the user. Determine the ranking of introduction targets for the query by inputting one or more indicators based on the background included in behavioral information indicating the user's behavior into a trained model that has been trained in advance,
The computer of the first device is
having a data set in which features are organized according to predetermined rules, extracting set features from the data set, and providing the extracted features to the determining device;
Information processing method.

Images