JP6833540B2 - Extractor, extraction method and extraction program - Google Patents

Description

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

In recent years, with the rapid spread of the Internet, various contents have been actively provided via networks. For example, when the content is a product, the user searches by specifying the category of the product and accesses the introduction page or the sales page of the content displayed as the search result.

Here, as an example of the technology using the search result related to the content, the ranking of the target product is generated from the sales history of the target product group searched by the user, and the generated ranking is used as the text of the advertisement content. The technology is known.

JP-A-2015-79449

However, in the above-mentioned conventional technique, it is not always possible to more flexibly appeal the characteristics of the product. Specifically, in the above-mentioned conventional technology, the text of the advertisement content is automatically generated from the features other than the features that are the targets of the ranking, and the page (web page, etc.) that actually introduces the product is displayed. It does not present the characteristics of the product to the user. Therefore, it may not be possible to appeal the various features of the product to the user.

The present application has been made in view of the above, and an object of the present application is to provide an extraction device, an extraction method, and an extraction program capable of more flexibly appealing the characteristics of a product.

The extraction device according to the present application includes an acquisition unit that acquires product information associated with a product that is the target of commercial transactions on the network, and a predetermined set included in a set formed based on the contents specified by the user. The product is characterized by including an extraction unit that extracts feature information that satisfies a predetermined condition in the set from the product information of the predetermined product.

According to one aspect of the embodiment, there is an effect that the characteristics related to the product can be more flexibly appealed.

FIG. 1 is a diagram showing an example of an extraction process according to an embodiment. FIG. 2 is a diagram showing a configuration example of the extraction processing system according to the embodiment. FIG. 3 is a diagram showing a configuration example of the extraction device according to the embodiment. FIG. 4 is a diagram showing an example of a product information storage unit according to the embodiment. FIG. 5 is a diagram showing an example of the learning information storage unit according to the embodiment. FIG. 6 is a diagram showing an example of the provided information storage unit according to the embodiment. FIG. 7 is a diagram showing a configuration example of the web server according to the embodiment. FIG. 8 is a flowchart (1) showing a processing procedure according to the embodiment. FIG. 9 is a flowchart (2) showing a processing procedure according to the embodiment. FIG. 10 is a flowchart (3) showing a processing procedure according to the embodiment. FIG. 11 is a hardware configuration diagram showing an example of a computer that realizes the function of the extraction device.

Hereinafter, the extraction device, the extraction method, and the embodiment for implementing the extraction program according to the present application (hereinafter, referred to as “the embodiment”) will be described in detail with reference to the drawings. The extraction device, extraction method, and extraction program according to the present application are not limited by this embodiment. In addition, each embodiment can be appropriately combined as long as the processing contents do not contradict each other. Further, in each of the following embodiments, the same parts are designated by the same reference numerals, and duplicate description is omitted.

[1. Example of extraction process]
First, an example of the extraction process according to the embodiment will be described with reference to FIG. FIG. 1 is a diagram showing an example of an extraction process according to an embodiment. FIG. 1 shows an example in which a page (content) for introducing a product in an electronic commerce is extracted by the extraction device 100 according to the present application.

The extraction device 100 shown in FIG. 1 is a server device that generates a page (an example of content) that introduces a product in electronic commerce. Specifically, the extraction device 100 extracts information that characterizes the product based on a web page related to the product and information associated with the product. Then, the extraction device 100 provides a page (hereinafter, may be referred to as an "appeal page") capable of appealing the characteristics of the product to the user by adding the extracted features to the original web page or the like. Generate. In FIG. 1, a real estate rental property will be described as an example of a commercial material. Further, although a page (web page) is shown as an example of the content in FIG. 1, the content is not limited to the page, and includes various information contents such as an advertisement content and an information page provided via an application.

The user terminal 10 shown in FIG. 1 is an information processing terminal used by the user. For example, the user terminal 10 is a smartphone or tablet terminal. The user terminal 10 uses a service on the network according to the operation of the user. For example, the user terminal 10 can access an electronic commerce site (hereinafter referred to as "commercial transaction site") to make a purchase application or make an inquiry regarding various services. In the embodiment, the user terminal 10 may be read as a user. For example, the description "the user accesses the website" may actually indicate the situation that "the user terminal 10 operated by the user accesses the website".

Here, in a commercial transaction site such as a real estate information site illustrated in the embodiment, a plurality of businesses (for example, a real estate agent) upload various products (property in the example of FIG. 1) handled by each to the site. It is assumed that

In such a commercial transaction site, the title and description of the property are usually created by the business operator and associated with the property information (product information). In addition, the business operator also provides detailed information on the property, such as the rent of the property, the nearest station, the distance to the nearest station, the area (occupied area), the age of the property, and whether or not to allow pets. Correspond as product information of the property.

Then, the business operator uploads an image of the property and other product information along with the title and description of the property. The commercial transaction site side generates a property introduction page based on the information uploaded by the business operator. A user who uses the commercial transaction site accesses the commercial transaction site, refers to the title and description posted on the introduction page, and makes an application or inquiry to the property.

Here, in such a commercial transaction site, the creation of information to be given to the property (for example, the description of the property) may be entrusted to the business operator. For this reason, there is a problem that appropriate property information is not always provided to the user on the commercial transaction site. That is, when the business operator arbitrarily assigns an explanatory text, for example, by taking SEO measures (Search Engine Optimization) or the like, an excessively long explanatory text is given or an irrelevant phrase is inserted into the explanatory text. It may happen.

In this case, when the user accesses the commercial transaction site and browses the introduction page, it may be difficult to determine what kind of characteristics the property has. This can reduce the usability of the entire commercial trading site.

In addition, if the content described on the property introduction page is entrusted to the business operator, the characteristics of the property may not be fully appealed. For example, the property accessed by the user may be a property with a relatively low rent or pets can be bred in the set specified by the user as a search target (for example, a group of properties having a common nearest station). Suppose there is an advantage of However, since it is difficult for the business operator to determine whether such information can be an appealing point compared to other properties, there is no particular emphasis on the introduction page (for example, the description of the property). In some cases. Moreover, even if the business operator describes the characteristics, the user who browses the page of the property may not notice it.

Here, at the request of the commercial transaction site side, it is possible to oblige the business operator to give an appropriate explanation, but for the business operator who uploads a large number of properties, it is possible to make each property individually. It is a heavy work load to give an appropriate explanation. Therefore, the business operator may hesitate to upload to the commercial transaction site. In this case, the commercial transaction site side loses a large business operator, which is not desirable for the operation of the commercial transaction site. In addition, the number of uploads is enormous because the commercial transaction site side artificially gives an appropriate description (for example, a description that clearly states the appeal point) to all the uploaded properties. In some cases it is not realistic.

Therefore, the extraction device 100 according to the embodiment extracts information indicating the characteristics of the property from the product information associated with the property by the business operator by the method described below. Further, the extraction device 100 generates an appeal page to which the feature information (appeal point) for promoting the property is given, which is a page introducing the property based on the extracted features. By such processing, the extraction device 100 can generate a page that appeals the characteristics of the property to the user without imposing a burden on the business operator or the commercial transaction site side. As a result, the extraction device 100 can more flexibly appeal the features related to the product, so that the effect of appealing to the user can be improved. Hereinafter, an example of the extraction process performed by the extraction device 100 will be described along the flow with reference to FIG.

As shown in FIG. 1, the extraction device 100 performs learning for performing a process of extracting feature information from product information as a learning phase. First, the extraction device 100 acquires product information as learning data to be used in the learning process (step S11).

Specifically, the extraction device 100 acquires the existing property introduction pages B01, B02, B03, etc. posted on the commercial transaction site. The product B01 and the like are, for example, information on properties uploaded by the business operator, and include product information which is detailed information indicating the specifications of each property. The learning data may be an introduction page for product B01 or the like generated based on the information uploaded by the business operator.

For example, the extraction device 100 acquires information such as the product B01 by crawling on the network using a known method. Alternatively, the extraction device 100 acquires information such as the product B01 from a predetermined web server or the like that provides the real estate information site. The extraction device 100 does not necessarily have to acquire the product B01 or the like from the same commercial transaction site. That is, the extraction device 100 may acquire the learning data from any commercial transaction site as long as the property and the product information associated with the property can be acquired.

Subsequently, the extraction device 100 creates a database of items that can be characteristic of the product information based on the acquired product information (step S12). For example, when the product is real estate such as a property, the extraction device 100 permits the rent of the property, the nearest station, the distance to the nearest station, the area (occupied area), the age of the building, and pet breeding. Whether or not to do so is judged as a characteristic item in the product information. The extraction device 100 may accept in advance the setting of items that can be artificially characterized in such a determination process. For example, the extraction device 100 receives in advance the setting of items that can be characteristic for each product category (for example, real estate). By such processing, for example, when the extraction device 100 acquires the product to be processed, the extraction device 100 can quickly extract the information corresponding to the item that can be the feature information from the product information.

In addition, the extraction device 100 learns about a portion of the acquired product information that can be characteristic information (appeal point) of the product in the explanatory text (text data) associated with the product (step S13). ). That is, the extraction device 100 generates a predetermined model (learner) for extracting features included in the text data.

Although the details will be described later, the extraction device 100 is not limited to one method, and a model may be generated by using various methods. For example, the extraction device 100 uses a summary sentence that manually summarizes the description sentence as learning data together with the description sentence. That is, the extraction device 100 performs machine learning using the abstract sentence in which the explanatory text is artificially summarized as correct answer data and the words and phrases extracted in the process as the feature information in the explanatory text. In this case, the extraction device 100 can learn the characteristics of the words and phrases extracted during the manual summarization process by using a sufficient sample for learning. Therefore, when a new explanatory text is input to the model, the extraction device 100 can extract words and phrases that are assumed to be important from the explanatory text by simulating the process of manually creating the summary text. This extracted information can be characteristic information of the description (in other words, the property associated with the description).

Further, the extraction device 100 may generate a model by learning by statistical processing. For example, the extraction device 100 performs learning to extract words and phrases included in a certain explanatory text and characterizing the explanatory text. Such processing is performed, for example, when the description is regarded as one document, the importance of each word in the document is calculated, and a rule for extracting words from the document based on the importance is generated. .. For example, when a certain explanatory text (document) is input based on an index value such as tf-idf (Term Frequency-Inverse Document Frequency), the extraction device 100 uses the phrase that characterizes the explanatory text. You may generate a model that extracts (outputs).

Further, the extraction device 100 may generate a model by using, for example, a method of deep learning. That is, the extraction device 100 uses a deep learning method to perform learning for extracting a portion that can be a feature in the explanatory text. For example, the model generated by the extraction device 100 is a model in which a plurality of nodes that output calculation results for the input data are connected in multiple layers, and is a model for learning the features included in the text data. For example, the model is a neural network in which layers having a plurality of nodes are connected in multiple stages, and is a DNN (Deep Neural Network) realized by a so-called deep learning technique.

For example, the model is generated by the following learning method by the deep learning technique. The connection factor between each node is initialized in the model, and text with various phrases is input. Then, the model is generated by processing such as backpropagation (Backpropagation) that corrects parameters (connection coefficients) so that the error between the output and the input in the model is reduced. For example, the model is trained so as to output the feature information (for example, words and phrases that characterize the description) included in the description when the text data constituting the description is input. In the learning process, adjustments may be made to the output of the model, such as manually returning appropriate feedback.

The model learning method and the generated model are not limited to the above-mentioned methods, and any known technique can be applied. That is, the extraction device 100 can use any model as long as it can output the feature information indicating the abstracted feature from the text data. The extraction device 100 stores the model generated in the learning phase in a predetermined storage unit. The above is the learning phase according to the embodiment.

After that, as the generation phase, the extraction device 100 extracts points that are particularly appealing points from the explanations associated with the property when the user uses the real estate information site or the like, and based on the extracted information. Performs the process of generating an appeal page.

For example, the user specifies the product (property) to be browsed by inputting the conditions related to the product on the real estate information site (step S21). For example, the user narrows down the properties that he / she wants to browse by specifying conditions such as the area where the property is located and the nearest station. Then, in the example of FIG. 1, the user browses the predetermined property XXX among the narrowed down properties. For example, it is assumed that the user clicks the link for browsing the predetermined property XXX from the property group presented after specifying the conditions.

At this time, the extraction device 100 specifies a set formed based on the contents specified by the user. Here, the set refers to a plurality of properties that match the contents specified by the user among the properties that the extraction device 100 has created in the database. For example, the set is common to the property group narrowed down by the conditions specified by the user.

Then, the extraction device 100 extracts, as feature information, an item that is superior to the property included in the set from the product information included in the property that the user intends to browse (step S22). For example, the extraction device 100 extracts an item that is superior when compared with the product information of a plurality of properties included in the set from the product information associated with the product XXX.

For example, the extraction device 100 refers to a database constructed in advance, and among the product information associated with the property XXX, the rent is relatively cheap compared to the set (for example, in the order of the cheapest rent in the set). To determine the superiority of (belonging to the top 20%, etc.). Further, the extraction device 100 has a relatively short distance from the nearest station as compared with the set among the product information associated with the property XXX (for example, in the set, the distance from the nearest station is higher in the order of the closest distance). Judge the superiority of (belonging to 20%, etc.).

Further, the extraction device 100 acquires the explanatory text F10 uploaded from the business operator as the product information associated with the property XXX. Then, the extraction device 100 extracts the feature information, which is the information indicating the feature of the property, from the acquired explanatory text F10. Specifically, the extraction device 100 inputs the acquired explanatory text F10 into the model and outputs the feature information in the explanatory text F10.

For example, the extraction device 100 uses a model generated corresponding to the product category "real estate", and as feature information in the explanation F10, "rent", "corner room", "from the station", etc. "Age" etc. shall be extracted. Further, the extraction device 100 also extracts the vicinity of the extracted feature information (for example, words and phrases included in a predetermined number before and after the explanatory text F10 when morphological analysis is performed) as words and phrases related to the feature information. For such processing, an existing processing method such as extraction of important sentences may be applied.

Then, the extraction device 100 further identifies the feature in which the property XXX is superior in the set among the extracted feature information. In the example of FIG. 1, in the property XXX, the items corresponding to "rent" and "time (distance) from the nearest station" are advantageous features in the set. Therefore, the extraction device 100 determines that these extracted items are the feature information in the property XXX, and generates an appeal page that appeals the extracted feature information (step S23).

Specifically, the extraction device 100 highlights the words and phrases corresponding to the items of "rent" and "time (distance) from the nearest station" in the explanatory text F10 associated with the property XXX by the business operator. Generate an appeal page of the above-mentioned aspect. More specifically, the extraction device 100 generates an appeal page that highlights words and phrases indicating the extracted features.

For example, as shown in FIG. 1, the extraction device 100 has high points in the explanation F10 where "the rent is a reasonable price of 65,000 yen" and "a 5-minute walk from the station". A light-displayed appeal page W10 is generated (in FIG. 1, for illustration purposes, the relevant portion is represented by a frame). The extraction device 100 does not newly generate the appeal page W10, but gives an instruction to highlight the introduction page of the original property (for example, an instruction to highlight the part corresponding to the feature information). The appeal page W10 may be generated by adding a script) in which is described. In the example of FIG. 1, the appeal page W10 is a content in which "rent", "distance from the station", etc. are highlighted with respect to the explanation page F10 and the introduction page including the image P10 of the property XXX.

Then, the extraction device 100 provides the generated appeal page W10 to the user (step S24). By browsing the appeal page W10, the user can see not only the information described in the description F10 of the property XXX associated with the original business operator, but also what excellent features the property XXX has for the set. It is possible to grasp whether or not it has. Specifically, the user browses the description F10 of the property XXX and finds that the rent of the property XXX is relatively low and the distance from the station is short among the properties that satisfy the conditions desired by the user. Can be grasped. That is, when the user browses the page where the property is introduced, the feature of the property XXX is made efficient by reading the feature highlighted by the extraction device 100 without reading all the redundant explanations F10. I can grasp it well.

In this way, the extraction device 100 according to the embodiment acquires the product information associated with the product that is the target of the commercial transaction on the network. Then, the extraction device 100 includes information (for example,) that satisfies a predetermined condition in the set among the product information of the predetermined product in the predetermined product included in the set formed based on the content specified by the user. , Items with specifications belonging to the top 20%, etc.) Extract feature information. Further, the extraction device 100 generates an appeal page for a predetermined product based on the extracted feature information.

Specifically, the extraction device 100 extracts the characteristics of the property from the product information set by various businesses on the property introduction page or the like on the real estate information site, and appeal page W10 based on the extracted characteristics. To generate. As a result, the extraction device 100 generates the information indicating that the property has particularly excellent characteristics in the set and the appeal page W10 adopted while maintaining the information such as the explanatory text F10 given by the business operator. be able to. As a result, the extraction device 100 can have the effect of improving the transmissibility of information on the commercial transaction site and more flexibly appealing the characteristics of the product.

The extraction device 100 may use various information to supplement the learning process. For example, in the learning phase described above, in the extraction device 100, the user has converted (for example, inquired about the property or applied for a rental contract) via the appeal page W10 generated by the extraction device 100. ) May be obtained as a result information (user reaction). Then, the extraction device 100 may feed back the feature information given to the appeal page W10, which has a good user response, such as a high conversion rate, as correct answer data to the learning process. As a result, the extraction device 100 can proceed with learning so as to optimize the process of extracting the points appealing to the user from the explanation. Hereinafter, the configuration of the extraction device 100 that performs such processing and the extraction processing system 1 including the extraction device 100 will be described in detail.

[2. Extraction processing system configuration]
Next, the configuration of the extraction processing system 1 including the extraction device 100 according to the embodiment will be described with reference to FIG. FIG. 2 is a diagram showing a configuration example of the extraction processing system 1 according to the embodiment. As illustrated in FIG. 2, the extraction processing system 1 according to the embodiment includes a user terminal 10, a business operator terminal 20, a web server 30, and an extraction device 100. These various devices are communicably connected via network N by wire or wirelessly. The extraction processing system 1 shown in FIG. 2 may include a plurality of user terminals 10, a plurality of business terminal 20s, and a plurality of web servers 30.

The user terminal 10 is, for example, an information processing device such as a smartphone, a desktop PC (Personal Computer), a notebook PC, a tablet terminal, a mobile phone, a PDA (Personal Digital Assistant), or a wearable device (Wearable Device). Is. The user terminal 10 accesses the commercial transaction site according to the operation by the user, and performs a process for the user to browse the information or apply for a contract for the property on the commercial transaction site.

The business terminal 20 is an information processing terminal used by a business that uploads a property on a commercial transaction site. The business terminal 20 is, for example, an information processing device such as a smartphone, a desktop PC, a notebook PC, a tablet terminal, a mobile phone, a PDA, or a wearable device.

The business operator terminal 20 executes various processes for uploading the property to the commercial transaction site according to the operation by the business operator. For example, the business terminal 20 uploads text data such as a property title and a description to a web server 30 that provides a commercial transaction site. In addition, the business terminal 20 uploads product information including information on property specifications (rent, etc.) to the web server 30.

The web server 30 is a server device that provides content (for example, a web page) when accessed from the user terminal 10. In the embodiment, the web server 30 provides a predetermined commercial transaction site (for example, a real estate information site), but also a news site, a weather forecast site, a shopping site, a finance (stock price) site, a route search site, and a map providing site. , Travel sites, restaurant introduction sites, web blogs, etc. may be provided.

The web page provided by the web server 30 may include an advertising space that is a display area for displaying advertisements and a recommendation frame that is a display area for displaying recommendations related to products. Although the details will be described later, the extraction device 100 may generate the feature information of the product displayed in the advertisement frame or the recommendation frame.

The extraction device 100 acquires the product information associated with the product, and also obtains the product information of the predetermined product in the predetermined product included in the set formed based on the contents specified by the user. Among them, feature information which is information satisfying a predetermined condition in the set is extracted. Further, the extraction device 100 generates an appeal page for a predetermined product based on the extracted feature information.

Then, when the extraction device 100 receives access from the user, the extraction device 100 provides the generated appeal page to the user terminal 10. The extraction device 100 according to the embodiment may also have a configuration as a web server 30 that operates and manages a commercial transaction site and manages information uploaded by a business operator. That is, the extraction device 100 and the web server 30 may be separate devices, or may be realized by a device having both functions.

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

(About communication unit 110)
The communication unit 110 is realized by, for example, a NIC (Network Interface Card) or the like. The communication unit 110 is connected to the network N by wire or wirelessly, and transmits / receives information to / from the user terminal 10, the business terminal 20, and the web server 30 via the network N.

(About storage unit 120)
The storage unit 120 is realized by, for example, a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory (Flash Memory), or a storage device such as a hard disk or an optical disk. The storage unit 120 includes a product information storage unit 121, a learning information storage unit 122, and a provision information storage unit 123.

(About the product information storage unit 121)
The product information storage unit 121 stores information about the product uploaded to the commercial transaction site. Here, FIG. 4 shows an example of the product information storage unit 121 according to the embodiment. FIG. 4 is a diagram showing an example of the product information storage unit 121 according to the embodiment. In the example shown in FIG. 4, the product information storage unit 121 has items such as "product ID" and "product information". In addition, "product information" includes "category", "region", "floor plan", "rent", "time from the nearest station", "size", "age", "pet", and "description". It has a small item such as ".

The "product ID" indicates identification information for identifying the product. In the embodiment, the identification information such as the product ID is the same as the reference code used in the description. For example, a product whose product ID is "B01" may be described as "product B01".

"Product information" indicates information associated with a product. For example, product information is composed of items indicating the specifications of the product, values for each item, and the like. In addition, the product information includes information such as a description set by the business operator who is the provider of the product and the title of the product.

The product information includes information different for each product category. For example, when the product category is "real estate", the product information includes items such as "rent" and "size". Alternatively, when the product category is "computer", the product information includes items such as "CPU" and "memory capacity".

"Category" indicates the category to which the product belongs on the commercial transaction site. “Region” indicates the region where the product (property in the example of FIG. 4) is located. “Floor plan” indicates the floor plan of the property. "Rent" indicates the rent of the property. "Time from the nearest station" indicates the distance from the nearest station, the travel time from the nearest station, and the like. "Area" indicates the occupied area of the property. "Age" indicates the number of years that have passed since the property was built. “Pet” indicates information such as whether or not there is a property where pets can be bred.

The “description” indicates a description associated with the property. In the example shown in FIG. 4, the "explanatory text" is shown by a concept such as "F01", but in reality, the text data constituting the explanatory text is stored in the item of the explanatory text.

That is, in an example of the data shown in FIG. 4, the category of the product in the product B01 identified by the product ID "B01" is "real estate", and the product information of the product B01 is based on the area where the product is located. It is "AAA Ward", the floor plan is "1K", the rent is "65,000" yen, the time from the nearest station is "5 minutes on foot", the area is "22" square meters, and pet breeding. Is "OK" and the description is "F01".

(About learning information storage unit 122)
The learning information storage unit 122 stores information related to learning processing based on product information. Here, FIG. 5 shows an example of the learning information storage unit 122 according to the embodiment. FIG. 5 is a diagram showing an example of the learning information storage unit 122 according to the embodiment. In the example shown in FIG. 5, the learning information storage unit 122 has items such as "learning data ID", "learning data", "category", and "extracted item".

The "learning data ID" indicates identification information for identifying the learning data. The “learning data” indicates the content of the learning data actually used in the learning process. For example, the learning data according to the embodiment is product information associated with the product.

"Category" indicates the category to which the product belongs. For example, the extraction device 100 performs learning for each category of products and generates a model for each category. “Extracted items” indicate items learned to be extracted as feature information in each category.

That is, as an example of the data shown in FIG. 5, the learning data C01 identified by the learning data ID “C01” learns the product “B01”, the product “B02”, the product “B03”, etc. as learning data. The category is "real estate", and the items extracted as characteristic information are, for example, "rent", "size", "time from the nearest station", and "age". It shows that.

The learning information storage unit 122 may store the model itself learned based on the learning data. Further, the extraction device 100 may update the learning data stored in the learning information storage unit 122 as appropriate, and may continue learning the generated model.

Further, in the example shown in FIG. 5, "rent" and the like are shown as extraction items, but it is assumed that the extraction device 100 does not necessarily extract only the phrase "rent" but indicates "rent". Synonyms and the like may be stored as extraction items. For example, the extraction device 100 estimates that terms such as "rent" and "monthly payment amount" are synonymous by manual setting or learning processing. In this case, the extraction device 100 uses the phrase "rent" when the phrase "monthly payment is ..." appears even if the phrase "rent" does not appear in the description. Can be extracted as feature information corresponding to.

Further, although not shown in FIG. 5, the learning data items include not only the product information used for learning, but also, for example, information manually set as an item to be extracted as feature information and explanatory text. Correct answer data such as a summary sentence manually created based on the above may be included. Further, when a deep learning method is used in learning, it is not always necessary to use only learning data for which a correct example (correct answer data) is clearly known. For example, the extraction device 100 does not necessarily require correct answer data as long as it can extract the features of the text data (document) included in the product information and can learn to recognize the feature information included in the text data.

(About the provided information storage unit 123)
The provided information storage unit 123 stores information about a page (appeal page) that is provided to the user and is generated by the extraction device 100. Here, FIG. 6 shows an example of the provided information storage unit 123 according to the embodiment. FIG. 6 is a diagram showing an example of the provided information storage unit 123 according to the embodiment. In the example shown in FIG. 6, the provided information storage unit 123 has the "appeal page ID", "original product ID", "aggregate information", "extracted item", "appeal point", "effect measurement information", It has items such as "review information".

The "appeal page ID" indicates identification information for identifying the appeal page generated by the extraction device 100. The "original product ID" indicates the identification information of the product that is the source of the appeal page.

The "set information" indicates information about the formed set specified by the user. In the example shown in FIG. 6, "set information" is shown by a concept such as "F01", but in reality, various specific information is stored. For example, the set information includes a condition specified by the user, identification information of a plurality of products included in the set formed based on the condition specified by the user, and the like. The set information is dynamic information. That is, since the set changes every time the condition or the like specified by the user is changed, the set information also changes every time the condition or the like specified by the user is changed. In addition, as the set changes, the dominant information in the set also changes, so the items (feature information) highlighted on the appeal page also change.

The “extracted item” corresponds to the same item shown in FIG. The "appeal point" indicates an item that is superior in the set among the items extracted as information that characterizes the product. For example, if the rent price is superior to other products among a plurality of products included in the set, the appeal point will include the item of "rent". As described above, the condition such as which item is "superior" may be artificially set by, for example, the administrator of the extraction device 100.

The “effect measurement information” indicates the result information indicating that the appeal page is effective among the information regarding the user's reaction to the appeal page. The "review information" indicates the information regarding the review posted by the user to the appeal page among the information regarding the user's reaction to the appeal page. In the example shown in FIG. 6, "effect measurement information" and "review information" are shown by concepts such as "G01" and "H01", but in reality, various specific information is stored. To.

For example, the “effect measurement information” may store the number of times the appeal page is accessed by the user, and the number and rate of users who have converted on the appeal page. Examples of conversions include applying for a preview of the property on the appeal page and accessing the property introduction page (for example, the web page of the company that provides the property). Further, in the effect measurement information, for example, the number (the number of clicks or touches, etc.) or the rate of properties selected by the user may be stored even if a specific action such as a preview is not reached. In addition, the effect measurement information may store the frequency with which the user accesses the appeal page, the time spent on the appeal page, and the like. That is, any information may be stored in the effect measurement information as long as it is information on the user's reaction to the appeal page.

In addition, the "review information" stores information regarding user reviews for the appeal page. For example, when the appeal page employs a system in which the user evaluates the property on a five-point scale, the review information stores the points evaluated by the user. For example, in the review information, the average value of the numerical values from "0" to "5" transmitted from the user is stored. Further, the review information may store the number and rate of user reviews sent from the user, specific user review comments (text data), and the like. That is, any information may be stored in the review information as long as it is information related to the user review for the appeal page.

That is, an example of the data shown in FIG. 6 shows that the appeal page W01 identified by the appeal page ID "W01" is a page related to the product B01 identified by the original product ID "B01". There is. In addition, the set information associated with the appeal page W01 is "F01", and the extracted items are "rent", "size", "time from the nearest station", "age", etc., and further. Among them, the items that are the appealing points that are superior to the set are "rent" and "time from the nearest station". Further, it is shown that the effect measurement information of the appeal page W01 is "G01" and the review information is "H01".

(About control unit 130)
The control unit 130 is, for example, a controller, and various programs (as an example of an extraction program) stored in a storage device inside the extraction device 100 by a CPU (Central Processing Unit), an MPU (Micro Processing Unit), or the like. (Equivalent) is realized by executing RAM as a work area. Further, the control unit 130 is, for example, a controller, and is realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

As shown in FIG. 3, the control unit 130 includes an acquisition unit 131, a learning unit 132, a reception unit 133, an extraction unit 134, a generation unit 135, and a provision unit 136, and the information described below. Realize or execute the function or action of processing. The internal configuration of the control unit 130 is not limited to the configuration shown in FIG. 3, and may be another configuration as long as it is a configuration for performing information processing described later. Further, the connection relationship of each processing unit included in the control unit 130 is not limited to the connection relationship shown in FIG. 3, and may be another connection relationship.

(About acquisition unit 131)
The acquisition unit 131 acquires various types of information. For example, the acquisition unit 131 acquires the product information associated with the product that is the target of the commercial transaction on the network.

For example, the acquisition unit 131 acquires an item indicating the specifications of the product and a value (numerical value or content) corresponding to the item. The acquisition unit 131 acquires different product information for each product category. As an example, if the product category is real estate, the acquisition unit 131 acquires the product information such as rent and area as the product information and the product information shown in FIG. The acquisition unit 131 may acquire any information as product information as long as it is information such as specifications that characterize the product.

Then, the acquisition unit 131 associates the product with the product information and stores the product in the product information storage unit 121 as shown in FIG. That is, the acquisition unit 131 builds a database related to products on the commercial transaction site. In such a database, for example, it is possible to manage data by sorting properties according to rent, size, etc., and extracting the top 20% when sorted.

In addition, the acquisition unit 131 acquires text data associated with the product as product information. For example, the acquisition unit 131 acquires the text data corresponding to the explanatory text for explaining the product on the page introducing the product. The page introducing the product includes a page on which an application for the product can be made and a page on which a transaction such as purchasing or selling the product can be performed.

For example, when the product is a property, the acquisition unit 131 morphologically analyzes the explanatory text associated with the property and acquires the words included in the explanatory text. Further, the acquisition unit 131 acquires a word indicating the characteristics of the property. More specifically, the acquisition unit 131 acquires, for example, words indicating rent, size, age, distance, nearest station, etc. as words indicating the characteristics of the property. The acquisition unit 131 may acquire any information as long as it is text data associated with a product such as a title, not limited to an explanatory text.

When the title or description is morphologically analyzed, the acquisition unit 131 determines the frequency of appearance of the word or a plurality of words based on the number of occurrences of words (phrases) included in the document (here, the title or description). You may calculate the importance of each word in the description of. For example, the acquisition unit 131 may calculate the DF value of the word related to the acquired explanation. Further, the acquisition unit 131 may acquire not only the DF value but also an index value such as tf-idf. Then, the acquisition unit 131 may acquire only the words that exceed a predetermined threshold value. As a result, the acquisition unit 131 can acquire only the words that are supposed to characterize the property without acquiring general words and phrases contained in various documents.

In addition, the acquisition unit 131 may acquire information regarding the reaction to the appeal page from the user who actually browsed the appeal page. For example, the acquisition unit 131 acquires at least one of the number or rate of products selected by the user or the number or rate of conversions related to the product as the user's reaction to the appeal page or product. May be good. In addition, the acquisition unit 131 may acquire information regarding the review transmitted from the user. Then, the learning unit 132, the extraction unit 134, and the generation unit 135, which will be described later, may appropriately tune the generated model and the feature information to be extracted based on the information regarding the user's reaction acquired by the acquisition unit 131. Good.

The acquisition unit 131 may acquire information about a user who has viewed the appeal page generated by the generation unit 135. The information about the user is, for example, user attribute information (gender, age, place of residence, occupation, etc.) and user behavior information (past browsing history, purchase history, etc. on the network). For example, the acquisition unit 131 identifies a user based on a cookie transmitted from the user terminal 10 and acquires information about the identified user.

The acquisition unit 131 appropriately stores the acquired information in each unit of the storage unit 120. Further, the acquisition unit 131 may appropriately acquire information used for processing by each processing unit, which will be described later, from each unit of the storage unit 120.

(About learning unit 132)
The learning unit 132 learns to extract feature information, which is information indicating the characteristics of the product, from the product information acquired by the acquisition unit 131.

More specifically, the learning unit 132 learns to extract characteristic information indicating the characteristics of the product from the text data (explanatory text) preset by the business operator for the product. ..

For example, the learning unit 132 performs learning for extracting feature information based on learning processing (for example, supervised machine learning) using correct answer data set in advance as items to be extracted from product information. For example, the learning unit 132 acquires information obtained by artificially extracting items and phrases that are expected to indicate the characteristics of the product as correct answer data. Then, the learning unit 132 learns the feature information to be extracted from the product information by learning these correct answer data (samples).

The learning unit 132 may perform learning for extracting feature information based on predetermined conditions set in advance for each product category. Specifically, when the product is "real estate", the learning unit 132 accepts items such as "rent" and "size" as correct answer data as the feature information to be extracted. Then, the learning unit 132 learns which part of the explanatory text corresponds to the item such as "rent" or "size" based on the correct answer data. In this way, the learning unit 132 can efficiently learn which information in the explanatory text indicates the characteristics of the product by using the correct answer data for each category as a sample.

Further, the learning unit 132 may extract the feature information by using a model (learner) learned to extract the information that characterizes the explanatory text associated with the product from the product information. Good. For example, the learning unit 132 can generate a model for extracting information (feature amount) that characterizes each explanatory sentence even if it does not have clear correct answer data by using a method such as deep learning.

The learning unit 132 may generate a model by using any of the above learning processes. Further, the learning unit 132 may appropriately use a known learning process other than those exemplified above. That is, if the learning unit 132 can perform learning for realizing the extraction process of the feature information by the extraction unit 134 described later and the generation process of the appeal page by the generation unit 135, the learning method is limited. Absent.

The learning unit 132 may perform a process of adjusting the feature information to be extracted based on the information regarding the reaction of the user. For example, the learning unit 132 extracts features extracted on an appeal page in which the response from the user exceeds a predetermined threshold value (for example, a page in which the conversion exceeds a predetermined number or a page in which the user review score exceeds a predetermined score). The model is updated by judging the information as correct data. That is, the learning unit 132 treats the appeal page, which has a relatively good response from the user, as correct answer data in model generation, assuming that the page is a page from which feature information can be appropriately extracted. As a result, the learning unit 132 can optimize the model.

Further, the learning unit 132 may generate a model that extracts different feature information for each user. For example, among the feature information appealed on the appeal page, there is a possibility that the feature information whose appeal effect changes depending on the user's attribute may exist. For example, on an appeal page where a feature is highlighted, the conversion rate may be high for female users, but low for male users. In such a case, the learning unit 132 may perform learning such as adjusting the items to be extracted from the product information according to the user attribute. Specifically, the learning unit 132 may perform learning such as adjusting items extracted as feature information so that the appeal page for female users and the appeal page for men are displayed in different modes. Good. As a result, the generation unit 135, which will be described later, generates a different appeal page in which each feature information that is expected to have a high appeal effect to each user is emphasized even if the page appeals the same product. can do.

The learning unit 132 appropriately stores the learning data used for learning and the model corresponding to the learning data in the learning information storage unit 122.

(About reception desk 133)
The reception unit 133 receives various requests regarding commercial transactions. For example, the reception unit 133 receives a request from the user terminal 10 to search for a product that meets the conditions desired by the user on the commercial transaction site.

Specifically, the reception unit 133 receives a search query for searching for a property from the user terminal 10 that has accessed the commercial transaction site. In addition, the reception unit 133 receives a request for browsing (acquiring) a page related to a product desired by the user from a user who browses the search result page corresponding to the search query. That is, the reception unit 133 receives an access request to the appeal page desired by the user. The reception unit 133 may not accept various requests directly from the user terminal 10 but may accept them via the web server 30.

(About extraction unit 134)
The extraction unit 134 is information that satisfies a predetermined condition in the set among the product information of the predetermined product in the predetermined product included in the set formed based on the contents specified by the user. Extract feature information.

Specifically, the extraction unit 134 extracts the product information as feature information when the product information of the predetermined product included in the set is information showing superiority in the set as a predetermined condition. To do.

For example, the extraction unit 134 extracts feature information based on predetermined conditions set in advance for each product category. Specifically, the extraction unit 134 first extracts characteristic items from the product information using the model generated by the learning unit 132. Then, the extraction unit 134 extracts the extracted items that meet the predetermined conditions set in advance for each category as the feature information as the information indicating the superiority in the set.

As an example, if the category is real estate, the extraction unit 134 states that "a set of product information items of a plurality of properties (that is, a set formed based on the conditions specified by the user) that match the conditions searched by the user). We accept in advance conditions such as "extract information belonging to the top 20% as feature information." Then, the extraction unit 134 extracts the extracted items that meet the above conditions as feature information. In the example of FIG. 1, the extraction unit 134 states that, of the product information of the property XXX that the user tried to browse, two items, "rent" and "time to the nearest station", met the above conditions. judge. Then, the extraction unit 134 extracts these items as feature information (appeal points).

It should be noted that the information having superiority in the set does not necessarily have to be the information to be ranked. For example, the information that is superior in the set may be information that indicates the advantages that other products do not necessarily have in the set. Specifically, when the product is a property, it can be said that the information that "pets can be bred" indicates the advantage that other products do not necessarily have. In this case, as a predetermined condition, for example, "Among the items of product information of a plurality of properties that match the conditions searched by the user, a predetermined specification that more than half of the products do not have is extracted as feature information. When the setting such as "to do" is made, the extraction unit 134 may extract the information that "pets can be bred" as feature information.

The extraction unit 134 determines conformity with the above conditions by referring to a database constructed from the product information acquired by the acquisition unit 131.

Here, the extraction unit 134 extracts the feature information from the description of the product (text data) from the product information associated with the product. Such processing is realized by using the model generated by the learning unit 132 described above.

That is, the extraction unit 134 extracts the feature information based on the learning process using the correct answer data set for the text data associated with the product among the product information. In this case, the extraction unit 134 inputs the text data into the model learned by being artificially given the correct answer data in advance, and any part of the text data corresponds to the "item to be extracted". Can be determined. Then, the extraction unit 134 extracts the items that satisfy the predetermined conditions as described above from the items extracted from the text data as the feature information of the product.

Further, the extraction unit 134 may extract the feature information by using a model (learner) learned to extract the information that characterizes the text data associated with the product among the product information. Such processing is realized by using a model learned by using a method such as deep learning.

That is, the extraction unit 134 extracts words and sentences that are determined to characterize the text data associated with the product from the product information. Then, the extraction unit 134 associates the extracted words and sentences with each item included in the database of the product, and any part of the text data corresponds to the "item to be extracted". Is determined. It should be noted that the process of associating the extracted words and sentences with each item included in the database of products is also realized by using a model learned by using the above-mentioned method such as deep learning. To. Then, the extraction unit 134 extracts the items that satisfy the predetermined conditions as described above from the items extracted from the text data as the feature information of the product.

For example, the extraction unit 134 extracts words and phrases that characterize the product as feature information. That is, the extraction unit 134 extracts words and phrases that characterize the product and that show superiority in the set. The generation unit 135, which will be described later, generates an appeal page by highlighting the words (appeal points) extracted by the extraction unit 134.

The feature information (words) extracted by the extraction unit 134 is not limited to one word such as "rent" or "size". For example, the extraction unit 134 may extract a sentence that is assumed to describe "rent" as feature information. In addition, the extraction unit 134 may extract a phrase that is supposed to describe "rent" and a predetermined number of words before and after the phrase as feature information. In addition, such a process may be realized by using a known important sentence extraction process or the like. As a result, the extraction unit 134 extracts a phrase that is supposed to describe "rent" and a predetermined number of words before and after it, so that it does not become unnatural for the user. Can be selected and extracted.

The extraction unit 134 may perform a process of adjusting the feature information to be extracted based on the information regarding the reaction of the user. For example, the extraction unit 134 is a feature extracted on an appeal page in which the response from the user exceeds a predetermined threshold value (for example, a page in which the conversion exceeds a predetermined number or a page in which the user review score exceeds a predetermined score). The information is determined to be correct answer data, and the result is fed back to the learning unit 132. That is, the extraction unit 134 treats the appeal page, which has a relatively good response from the user, as the correct answer data in the model generation, assuming that the page is the page from which the feature information can be appropriately extracted. As a result, the extraction unit 134 can optimize the extraction process.

Further, the extraction unit 134 may extract different feature information for each user. As explained in the learning unit 132, the appealing effect of the appeal point in the product may change depending on the attribute of the user. Therefore, the extraction unit 134 may adjust the feature information to be extracted for the product, for example, based on the correlation between the user's attribute and the user's reaction.

(About generator 135)
The generation unit 135 generates a page related to a predetermined product based on the feature information extracted by the extraction unit 134.

For example, the generation unit 135 generates a page in which the words (feature information) extracted by the extraction unit 134 are highlighted in the text data associated with the product. Specifically, the generation unit 135 generates an appeal page in which the portion of the text data corresponding to the feature information is highlighted and appealed. As a result, the generation unit 135 can immediately appeal to the user information indicating specifications such as how excellent the characteristics of the set are in the product displayed on the user terminal 10.

In addition, the generation unit 135 may generate an appeal page based on the reaction from the user. That is, like the extraction unit 134 described above, the generation unit 135 also learns the generation process based on the reaction of the user, using the highlighting range, the display mode of the feature information, etc. as correct answer data (or incorrect answer data). , You may optimize the generated appeal page.

As a specific example, a case where the property is converted will be described. On the appeal page, if the highlighting of the information presented as feature information (for example, the text corresponding to the description of the rent) is not appropriate, or if the highlighting method is not appropriate, in other words. If the appeal point of the product is not well communicated to the user, the user is expected to hesitate to apply for the property. On the other hand, when the property is converted by the user, it is assumed that the feature information and display method appealed on the appeal page were correct.

Therefore, the generation unit 135 estimates the correctness of the mode of the appeal page and the feature information based on the reaction of such a user. For example, the generation unit 135 sets a predetermined threshold value in advance. For example, the generation unit 135 sets a conversion rate threshold value for a certain property. Then, the generation unit 135 determines whether or not the conversion rate exceeding the threshold value has been achieved within a predetermined period or during a certain number of browsing.

When the conversion rate exceeds the threshold value, the generation unit 135 feeds back the appeal page as one of the correct examples, assuming that the appeal page related to the property is the correct answer. Further, when the conversion rate does not exceed the threshold value, the generation unit 135 feeds back the appeal page as one of the negative examples, assuming that the appeal page related to the property is incorrect. In this way, by continuing learning using the user's reaction, the generation unit 135 can use the method of appealing the feature information, the color usage of the page, and the highlighting process (for example, the size of the font and the font). Type, layout of the entire page, etc.) can be improved.

As a method for obtaining such a user reaction, the extraction device 100 may perform a manual determination using, for example, crowdsourcing. By performing learning using human hands, the generation unit 135 can generate an appeal page that does not look unnatural when viewed by a person.

In addition, the generation unit 135 may generate a different appeal page for each user based on the information about the user. That is, like the extraction unit 134 described above, the generation unit 135 may also generate an appeal page that is expected to be more effective for the user based on the user's attribute information, action history, and the like.

(About the provider 136)
The providing unit 136 provides various information. For example, when a request for viewing a product is received by the reception unit 133, the providing unit 136 provides (transmits) an appeal page to the user who has sent the request. Further, when the reception unit 133 receives the search query for searching the product, the providing unit 136 provides the user terminal 10 with the search result corresponding to the received search query.

[4. Web server configuration]
Next, the configuration of the web server 30 according to the embodiment will be described with reference to FIG. 7. FIG. 7 is a diagram showing a configuration example of the web server 30 according to the embodiment. As shown in FIG. 7, the web server 30 has a communication unit 31, a content storage unit 32, and a control unit 33.

The communication unit 31 is realized by, for example, a NIC or the like. Then, the communication unit 31 is connected to the network N by wire or wirelessly, and transmits / receives information to / from the user terminal 10, the business terminal 20, and the extraction device 100.

The content storage unit 32 is realized by, for example, a semiconductor memory element such as a RAM or a flash memory, or a storage device such as a hard disk or an optical disk. Then, the content storage unit 32 stores a web page which is an example of the content. For example, the content storage unit 32 stores an HTML file forming a web page, and a still image or a moving image displayed on the web page. The web page stored in the content storage unit 32 may include an advertisement acquisition command for acquiring the advertisement content to be displayed on the web page.

The control unit 33 is a controller, and is realized by, for example, a CPU, an MPU, or the like executing various programs stored in a storage device inside the web server 30 using the RAM as a work area. Further, the control unit 33 is a controller, and is realized by, for example, an integrated circuit such as an ASIC or FPGA.

As shown in FIG. 7, the control unit 33 has a reception unit 34 and a distribution unit 35, and realizes or executes an information processing function or operation described below. The internal configuration of the control unit 33 is not limited to the configuration shown in FIG. 7, and may be another configuration as long as it is a configuration for performing information processing described later. Further, the connection relationship of each processing unit included in the control unit 33 is not limited to the connection relationship shown in FIG. 7, and may be another connection relationship.

The reception unit 34 receives a web page acquisition request from the user terminal 10. For example, the reception unit 34 accepts an HTTP request as a request for acquiring a web page. In addition, the reception unit 34 receives information about the product from the business terminal 20. The reception unit 34 stores information about the product received from the business operator terminal 20 in the content storage unit 32.

The distribution unit 35 distributes the web page to the user terminal 10 when the reception unit 34 receives the request for acquiring the web page. Specifically, the distribution unit 35 acquires the acquisition request target web page from the content storage unit 32, and distributes the acquired web page to the user terminal 10. As described above, the web page stored in the content storage unit 32 includes an advertisement acquisition command. That is, when displaying the acquired web page, the user terminal 10 transmits an advertisement content distribution request to the extraction device 100 in accordance with the advertisement acquisition command included in the web page.

In the above embodiment, an example in which the appeal page is provided by the providing unit 136 related to the extraction device 100 is shown, but the web page such as the appeal page is distributed to the user terminal 10 by the distribution unit 35 related to the web server 30. May be done. In this case, the web server 30 cooperates with the extraction device 100 to transmit the information required for generating the appeal page to the extraction device 100 and to acquire the appeal page from the extraction device 100. ..

[5. Processing procedure]
Next, the procedure of the processing by the extraction device 100 according to the embodiment will be described with reference to FIGS. 8, 9 and 10. First, a processing procedure related to model generation will be described with reference to FIG. FIG. 8 is a flowchart (1) showing a processing procedure according to the embodiment.

As shown in FIG. 8, the extraction device 100 acquires product information associated with the extraction device product as learning data for model generation (step S101).

Then, the extraction device 100 generates a database based on the acquired product information (step S102). Further, the extraction device 100 generates a learning model based on the acquired product information (step S103). Extraction device The learning phase by the extraction device 100 ends according to the above flow.

Next, a processing procedure relating to the extraction process using the model will be described with reference to FIG. FIG. 9 is a flowchart (2) showing a processing procedure according to the embodiment.

As shown in FIG. 9, the extraction device 100 determines whether or not the designation regarding the product has been received from the user terminal 10 (step S201). When the extraction device 100 has not received the designation regarding the product (step S201; No), the extraction device 100 waits until it receives the designation.

On the other hand, when the designation regarding the product is received (step S201; Yes), the extraction device 100 specifies the set corresponding to the received designation (condition) (step S202). Then, the extraction device 100 extracts the feature information having an advantage over the set in the predetermined product included in the set (step S203).

Subsequently, the extraction device 100 generates an appeal page based on the extracted feature information (step S204). Then, the extraction device 100 provides an appeal page to the user who has specified the product (step S205). According to the above flow, the generation phase by the extraction device 100 ends.

Next, the procedure of feedback regarding the extraction process and the generation process will be described with reference to FIG. FIG. 10 is a flowchart (3) showing a processing procedure according to the embodiment.

As shown in FIG. 10, the extraction device 100 determines whether or not the reaction to the appeal page has been received from the user terminal 10 (step S301). When the extraction device 100 does not accept the reaction (step S301; No), the extraction device 100 waits until it accepts the reaction.

On the other hand, when the reaction is received (step S301; Yes), the extraction device 100 acquires the reaction from the user (step S302). Then, the extraction device 100 determines whether or not the reaction from the user exceeds a predetermined threshold value (step S303).

When the reaction from the user exceeds a predetermined threshold value (step S303; Yes), the extraction device 100 determines that the feature information (or the appeal page itself) on the appeal page provided to the user is the correct answer (step). S304). On the other hand, when the response from the user does not exceed a predetermined threshold value (step S303; No), the extraction device 100 determines that the feature information is incorrect (step S305).

Then, the extraction device 100 updates the model by using the learning data using the data obtained in steps S304 and S305 as a positive example or a negative example (step S306). In other words, the extraction device 100 learns the model more strongly based on the reaction of the user. By repeating the above flow, the extraction device 100 optimizes the model so that the feature information can be extracted more appropriately.

[6. Modification example]
The extraction device 100 described above may be implemented in various different forms other than the above embodiment. Therefore, another embodiment of the extraction device 100 will be described below.

[6-1. Types of feature information]
In the above embodiment, the extraction device 100 shows an example of generating an appeal page in a mode of highlighting a portion corresponding to the feature information in the description of the product. Here, the extraction device 100 may generate an appeal page in a mode different from the mode in which the feature information is highlighted.

For example, the extraction device 100 may generate a summary sentence of the explanatory text associated with the product based on the feature information, and may also generate an appeal page to which the summary sentence is added. Specifically, the extraction device 100 generates an appeal page in which the generated summary text is arranged prior to the explanatory text previously associated with the product by the business operator. That is, the extraction device 100 can convey to the user a summary sentence that concisely conveys only the features of the product such as excellent points as compared with the set, before the explanatory sentence. As a result, the user can accurately grasp the characteristics of the product without reading the explanation itself, for example, even if the explanation is redundant.

[6-2. Use for search results]
The extraction device 100 may not only generate an appeal page using the generated feature information, but may also use the feature information for various processes. For example, the extraction device 100 may display the feature information together with the search result in the search for the product performed on the commercial transaction site or the like.

That is, the extraction device 100 may generate a search result page including the feature information of the product when the product is displayed as a result of the predetermined search process. In this case, on the search result page, for example, feature information (for example, a summary sentence) related to the product is displayed in the vicinity of the product searched based on the search query. Therefore, when the user searches for a product, the user can grasp the characteristics of the product without displaying the appeal page. In this way, the extraction device 100 can improve the convenience of the user who uses the commercial transaction site.

[6-3. Use for recommendations]
The extraction device 100 displays the feature information not only in the commercial transaction site but also in the recommendation of the product displayed together with the web page or the like (for example, the information displayed in the advertisement frame or the recommendation frame of the web page). You may.

For example, a product listed on a commercial transaction site may be displayed as a recommendation on a web page or the like other than the product. In this case, the extraction device 100 displays the feature information together with the recommended product. As a result, the user can grasp the characteristics of the product without accessing the appeal page of the recommended product (without transitioning the display from the currently viewed page to the appeal page).

The extraction device 100 may acquire information as to whether the recommendation of the product has been clicked or whether the user has converted. For example, if the feature information displayed as a recommendation is inappropriate or the appeal effect is low, it is assumed that the user does not respond to clicking or converting. On the other hand, when the user returns a reaction of clicking or converting, it is assumed that the feature information displayed together with the recommendation is information that appropriately indicates the features of the product (correct answer).

Then, as described in the above embodiment, the extraction device 100 determines whether the feature information or the processing of the feature information is correct or not according to the reaction to the recommendation. As a result, the extraction device 100 can improve (learn) the accuracy of the extraction process and the extraction process by utilizing the reaction from the user other than the appeal page.

[6-4. commodity〕
In the above embodiment, real estate rental properties, personal computers, and the like are given as examples of products, but the products are not limited to these. That is, the extraction device 100 performs the extraction process according to the embodiment for any product such as a product or a service as long as it is a product handled on the commercial transaction site and the product information can be acquired. be able to.

[6-5. content〕
In the above embodiment, a web page is given as an example of the content, but the content is not limited to this example. For example, the content may be an advertisement content, an information page provided via an application, or content related to a recommendation. That is, the content generated by the extraction device 100 may be any content as long as it is information content provided via the network.

For example, when the extraction device 100 extracts the feature information of a predetermined product, the extracted feature information may be reflected in the advertising content for promoting the product. For example, the extraction device 100 extracts characteristic information about a product advertised by an advertising content submitted by an advertiser (product provider). Then, the extraction device 100 adds advertising information (for example, a summary sentence) generated from the extracted feature information to the advertisement content to generate the advertisement content. Specifically, the extraction device 100 adds a summary sentence generated based on the feature information to a place where text data for promoting the product is inserted in the advertisement content. Thereby, the extraction device 100 can generate advertising content that appeals more superior features to, for example, many competing products (that is, a set of products).

[7. Hardware configuration]
The extraction device 100, the user terminal 10, the business terminal 20, and the web server 30 according to the above-described embodiment are realized by, for example, a computer 1000 having a configuration as shown in FIG. Hereinafter, the extraction device 100 will be described as an example. FIG. 11 is a hardware configuration diagram showing an example of a computer 1000 that realizes the functions of the extraction device 100. The computer 1000 has a CPU 1100, a RAM 1200, a ROM 1300, an HDD 1400, a communication interface (I / F) 1500, an input / output interface (I / F) 1600, and a media interface (I / F) 1700.

The CPU 1100 operates based on a program stored in the ROM 1300 or the HDD 1400, and controls each part. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 is started, a program that depends on the hardware of the computer 1000, and the like.

The HDD 1400 stores a program executed by the CPU 1100, data used by such a program, and the like. The communication interface 1500 receives data from another device via the communication network 500 (corresponding to the network N shown in FIG. 2) and sends the data to the CPU 1100, and the data generated by the CPU 1100 is transmitted to the other device via the communication network 500. Send to the device.

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

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

For example, when the computer 1000 functions as the extraction device 100 according to the embodiment, the CPU 1100 of the computer 1000 realizes the function of the control unit 130 by executing the program loaded on the RAM 1200. Further, the data in the storage unit 120 is stored in the HDD 1400. The CPU 1100 of the computer 1000 reads these programs from the recording medium 1800 and executes them, but as another example, these programs may be acquired from another device via the communication network 500.

[8. Others]
Further, among the processes described in the above-described embodiment, all or a part of the processes described as being automatically performed can be manually performed, or the processes described as being manually performed can be performed. All or part of it can be done automatically by a known method. In addition, the processing procedure, specific name, and information including various data and parameters shown in the above document and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each figure is not limited to the illustrated information.

Further, each component of each of the illustrated devices is a functional concept, and does not necessarily have to be physically configured as shown in the figure. That is, the specific form of distribution / integration of each device is not limited to the one shown in the figure, and all or part of the device is functionally or physically distributed / physically in arbitrary units according to various loads and usage conditions. It can be integrated and configured. For example, the acquisition unit 131 shown in FIG. 3 and the reception unit 133 may be integrated. Further, for example, the information stored in the storage unit 120 may be stored in a predetermined storage device provided externally via the network N.

Further, in the above embodiment, an example is shown in which the extraction device 100 performs, for example, an acquisition process for acquiring product information, an extraction process for extracting feature information, and a generation process for generating an appeal page. However, the above-mentioned extraction device 100 may be separated into an acquisition device that performs acquisition processing, an extraction device that performs extraction processing, and a generation device that performs generation processing. In this case, the acquisition device has at least the acquisition unit 131. The extraction device has at least an extraction unit 134. The generator has at least a generator 135. Then, the processing by the extraction device 100 is realized by the extraction processing system 1 having each device of the acquisition device, the extraction device, and the generation device.

In addition, the above-described embodiments and modifications can be appropriately combined as long as the processing contents do not contradict each other.

[9. effect〕
As described above, the extraction device 100 according to the embodiment has an acquisition unit 131 and an extraction unit 134. The acquisition unit 131 acquires the product information associated with the product that is the target of the commercial transaction on the network. In the predetermined product included in the set formed based on the contents specified by the user, the extraction unit 134 is the feature information which is the information satisfying the predetermined condition in the set among the product information of the predetermined product. Is extracted.

As described above, the extraction device 100 according to the embodiment extracts the characteristics of the product from the product information set by various business operators, for example, on the product introduction page on the commercial transaction site. That is, the extraction device 100 appeals the dynamic features according to the user's designation by extracting the appeal points that are characteristic when the product is compared with the set (other products) designated by the user. That is, it is possible to more flexibly appeal the characteristics of the product.

Further, the extraction unit 134 extracts the product information as the feature information when the product information of the predetermined product included in the set is the information indicating the superiority in the set.

As described above, since the extraction device 100 according to the embodiment extracts the features having an advantage when compared with the set specified by the user, the features of the product can be clearly conveyed to the user.

In addition, the extraction unit 134 extracts feature information based on predetermined conditions set in advance for each product category.

In this way, the extraction device 100 according to the embodiment sets conditions for each category and extracts feature information. As a result, the extraction device 100 can extract accurate feature information corresponding to various products.

Further, the extraction unit 134 extracts the feature information based on the learning process using the correct answer data set for the text data associated with the product among the product information.

As described above, the extraction device 100 according to the embodiment can extract feature information similar to the features extracted manually by performing learning based on preset correct answer data. That is, the extraction device 100 can accurately extract the feature information indicating the features of the product.

Further, the extraction unit 134 extracts the feature information by using a learning device learned to extract the information that characterizes the text data associated with the product among the product information.

As described above, the extraction device 100 according to the embodiment may perform learning by a method of extracting the feature amount characterizing the text data by a method such as deep learning. By such processing, the extraction device 100 can extract the feature information accurately and efficiently.

In addition, the extraction device 100 according to the embodiment further includes a generation unit 135 that generates a page (an example of content) related to a predetermined product based on the feature information extracted by the extraction unit 134.

As described above, the extraction device 100 according to the embodiment extracts the features of the product on the product introduction page or the like on the commercial transaction site, and generates a page for appealing the features based on the extracted features. As a result, the extraction device 100 can appeal the dynamic features according to the user's designation.

In addition, the acquisition unit 131 acquires the reaction of the user who browses the page generated by the generation unit 135. The generation unit 135 updates the generated page based on the reaction from the user.

In this way, the extraction device 100 according to the embodiment can perform the extraction process by feeding back the reaction of the user. As a result, the extraction device 100 can optimize the page to be generated.

In addition, the acquisition unit 131 acquires information about the user who has viewed the page generated by the generation unit 135. The generation unit 135 generates a page related to a predetermined product and corresponding to the information about the user based on the information about the user.

As described above, the extraction device 100 according to the embodiment may generate different pages for each user. As a result, the extraction device 100 can provide a page in which the feature information tailored to the characteristics of each user is appealed, so that the appealing effect of the page can be improved.

In addition, the extraction unit 134 extracts words and phrases that characterize the product as feature information. The generation unit 135 generates a page in which the words and phrases extracted by the extraction unit 134 are highlighted in the text data associated with the product.

As described above, the extraction device 100 according to the embodiment may generate a page having a mode in which only the features are highlighted with respect to the original description. As a result, the extraction device 100 can appeal the superior feature of the product to the set without breaking the layout and contents such as the display of the explanation of the original business operator.

In addition, the generation unit 135 generates a summary sentence of the text data associated with the product based on the feature information, and also generates a page to which the summary sentence is added.

As described above, the extraction device 100 according to the embodiment may generate a summary sentence summarizing the features of the product as the feature information. As a result, the extraction device 100 can present the characteristics of the product to the user in a short sentence instead of the redundant explanation, so that it is possible to generate a page capable of efficiently transmitting information.

In addition, the generation unit 135 generates a page in which a summary sentence is arranged prior to the text data associated with the product.

As described above, the extraction device 100 according to the embodiment can convey the characteristics of the product to the user more quickly and accurately by generating a page for presenting the summary sentence prior to the original explanatory sentence. As a result, the extraction device 100 can improve the usability of the user who uses the commercial transaction site.

In addition, the generation unit 135 generates a search result page including the feature information extracted by the extraction unit 134 when the product is displayed as a result of the predetermined search processing.

As described above, the extraction device 100 according to the embodiment may generate a page in which the feature information is presented as the information displayed together with the search result. As a result, the extraction device 100 can convey the features of the product to the user without the user having to bother to access the appeal page.

The embodiments of the present application have been described in detail with reference to the drawings, but this is an example, and various modifications and improvements are made based on the knowledge of those skilled in the art, including the embodiments described in the disclosure column of the invention. It is possible to carry out the present invention in other forms described above.

In addition, the above-mentioned "section, module, unit" can be read as "means" or "circuit". For example, the acquisition unit can be read as an acquisition means or an acquisition circuit.

1 Extraction processing system 10 User terminal 20 Business terminal 30 Web server 100 Extraction device 110 Communication unit 120 Storage unit 121 Product information storage unit 122 Learning information storage unit 123 Provided information storage unit 130 Control unit 131 Acquisition unit 132 Learning unit 133 Reception Part 134 Extraction part 135 Generation part 136 Providing part

Claims (16)

An acquisition unit that acquires product information associated with a product that is the target of commercial transactions on the network,
The reception department that accepts specifications regarding products from users,
When the reception unit receives a designation regarding a product from the user, the predetermined product included in the set formed based on the designation regarding the product is included in the product information of the predetermined product. , An extraction unit that extracts feature information that satisfies a predetermined condition regarding the order of the specifications in the set from the information related to the specifications of the predetermined product.
An extraction device characterized by being equipped with. The extraction unit
When the product information of a predetermined product included in the set is information showing superiority in the set, the product information is extracted as the feature information.
The extraction device according to claim 1. The extraction unit
The feature information is extracted based on the predetermined conditions set in advance for each category of the product.
The extraction device according to claim 1 or 2. The extraction unit
Among the product information, the feature information is extracted based on the learning process using the correct answer data set for the text data associated with the product.
The extraction device according to any one of claims 1 to 3. The extraction unit
Among the product information, the feature information is extracted by using a learning device learned to extract information that characterizes the text data associated with the product.
The extraction device according to any one of claims 1 to 4, wherein the extraction device is characterized. A generation unit that generates content related to the predetermined product based on the feature information extracted by the extraction unit.
The extraction device according to any one of claims 1 to 5, further comprising. An acquisition unit that acquires product information associated with a product that is the target of commercial transactions on the network,
In a predetermined product included in a set formed based on the contents specified by the user, feature information that satisfies a predetermined condition in the set is extracted from the product information of the predetermined product. Extractor and
A generation unit that generates content related to the predetermined product based on the feature information extracted by the extraction unit is provided.
The extraction unit
The feature information is extracted from the text data associated with the product, and the feature information is extracted.
The generator
Based on the feature information, a summary sentence of the text data associated with the product is generated, and the content to which the summary sentence is attached is generated.
An extraction device characterized by that. The acquisition unit
The reaction of the user who viewed the content generated by the generation unit is acquired, and the reaction is obtained.
The generator
Update the generated content based on the user's reaction.
The extraction device according to claim 6 or 7. The acquisition unit
Obtain information about the user who viewed the content generated by the generation unit,
The generator
Based on the information about the user, the content related to the predetermined product and corresponding to the information about the user is generated.
The extraction device according to any one of claims 6 to 8, wherein the extraction device is characterized. The extraction unit
As the feature information, words and phrases that characterize the product are extracted.
The generator
Of the text data associated with the product, the content in which the words and phrases extracted by the extraction unit are highlighted is generated.
The extraction device according to any one of claims 6 to 9, wherein the extraction device is characterized. The generator
Prior to the text data associated with the product, the content in the mode in which the summary sentence is arranged is generated.
The extraction device according to claim 7. The generator
When the predetermined product is displayed as a result of a predetermined search process corresponding to a search query for searching the product having the content specified by the user, the extraction is performed as the content related to the predetermined product. Generate a search result page containing the feature information extracted by the department,
The extraction device according to any one of claims 6 to 11, characterized in that. An extraction method performed by a computer
The acquisition process to acquire the product information associated with the product that is the target of commercial transactions on the network,
The reception process that accepts specifications regarding products from users,
When the reception process receives a designation regarding a product from the user, the predetermined product included in the set formed based on the designation regarding the product is included in the product information of the predetermined product. , An extraction process for extracting feature information that satisfies a predetermined condition regarding the order of the specifications in the set from the information regarding the specifications of the predetermined product.
An extraction method characterized by containing. An extraction method performed by a computer
The acquisition process to acquire the product information associated with the product that is the target of commercial transactions on the network,
In a predetermined product included in a set formed based on the contents specified by the user, feature information that satisfies a predetermined condition in the set is extracted from the product information of the predetermined product. Extraction process and
Includes a generation step of generating content related to the predetermined product based on the feature information extracted by the extraction step.
The extraction step
The feature information is extracted from the text data associated with the product, and the feature information is extracted.
The production step is
Based on the feature information, a summary sentence of the text data associated with the product is generated, and the content to which the summary sentence is attached is generated.
An extraction method characterized by that. The acquisition procedure for acquiring the product information associated with the product that is the target of commercial transactions on the network, and
The reception procedure for accepting specifications regarding products from users,
When the reception procedure receives a designation regarding a product from the user, the predetermined product included in the set formed based on the designation regarding the product is included in the product information of the predetermined product. , An extraction procedure for extracting feature information that satisfies a predetermined condition regarding the order of the specifications in the set from the information regarding the specifications of the predetermined product.
An extraction program characterized by having a computer execute. The acquisition procedure for acquiring the product information associated with the product that is the target of commercial transactions on the network, and
In a predetermined product included in a set formed based on the contents specified by the user, feature information that satisfies a predetermined condition in the set is extracted from the product information of the predetermined product. Extraction procedure and
A computer is made to execute a generation procedure for generating content related to the predetermined product based on the feature information extracted by the extraction procedure.
The extraction procedure
The feature information is extracted from the text data associated with the product, and the feature information is extracted.
The generation procedure is
Based on the feature information, a summary sentence of the text data associated with the product is generated, and the content to which the summary sentence is attached is generated.
An extraction program characterized by that.

Images