JP7801558B2 - Product information generating program, product information generating method and information processing device - Google Patents

Description

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

An information processing system that supports the trading of multiple products may suggest some of those products to a user as trading targets. For example, the information processing system may suggest the purchase of products that match search keywords entered by the user.

An affinity device has been proposed that extracts keywords from the product information of a product selected by a user, uses the keywords to search for other related product names on blog sites, and suggests the searched product names to the user. An information search support device has also been proposed that collects operation information indicating user operations during information searches, detects navigation patterns indicating typical operations from the collected operation information, and suggests the navigation patterns to the user. A product search device has also been proposed that, if there is no product information matching the search keywords entered by the user, suggests reference keywords similar to the search keywords to the user.

Japanese Patent Application Laid-Open No. 2008-139928 JP 2009-64191 A JP 2009-151734 A

Users sometimes trade two or more products together to achieve a certain transaction goal. For example, a user may purchase two or more ingredients together to make a certain dish. However, general product search technology determines whether each product individually matches the user's input, and does not output an appropriate product combination.

In response to this, it is conceivable that information processing systems could extract combinations of products that are often traded together from past transaction data. However, the same product may be traded for different transaction purposes, and transaction data relating to different transaction purposes may be mixed together in multiple transaction data. For this reason, simply extracting product combinations that co-occur frequently may result in combinations that are not very useful. Therefore, in one aspect, the present invention aims to enable the search for product combinations.

In one aspect, a product information generation program is provided that causes a computer to execute the following processes: Acquires a plurality of transaction data each indicating transactions of at least some of a plurality of products, and a plurality of dialogue data corresponding to the plurality of transaction data, each indicating communication related to the transactions using natural language; Determines a product group from the plurality of products based on co-occurrence relationships between products in two or more of the plurality of transaction data corresponding to two or more dialogue data that satisfy a specific condition among the plurality of dialogue data; Associates at least some of the keywords included in the plurality of dialogue data with the determined product group.

In one aspect, a computer-implemented product information generation method is provided.In another aspect, an information processing device having a memory unit and a processing unit is provided.

One aspect is that it allows you to search for product combinations.

FIG. 1 is a diagram illustrating an information processing apparatus according to a first embodiment. FIG. 1 illustrates an example of an information processing system according to a second embodiment. FIG. 2 is a block diagram illustrating an example of hardware of an analysis server. FIG. 2 is a block diagram illustrating an example of hardware of a client. FIG. 10 is a diagram illustrating an example of a search screen for searching a product list. FIG. 10 is a diagram illustrating an example of a search screen for searching for an individual product. FIG. 10 is a diagram showing an example of a combination of products for different purchasing purposes. FIG. 10 is a diagram illustrating an example of association between a purchase record and a dialogue log. FIG. 10 is a diagram illustrating an example of generating a keyword list from a dialogue log. FIG. 10 is a diagram illustrating an example of association between a purchase record and a keyword list. FIG. 10 is a diagram showing an example of a count of pairs of product names and keywords. FIG. 10 is a diagram illustrating an example of the execution of the Apriori algorithm. FIG. 10 is a diagram illustrating an example of extraction of keywords corresponding to a product list. FIG. 10 is a diagram illustrating an example of association between a product list and a tag. FIG. 2 is a block diagram illustrating an example of functions of the information processing system. 10 is a flowchart illustrating an example of a procedure for generating a product list.

The present embodiment will be described below with reference to the drawings.
[First embodiment]
A first embodiment will be described.

FIG. 1 is a diagram illustrating an information processing apparatus according to a first embodiment.
The information processing device 10 according to the first embodiment analyzes past product transactions to generate product information for suggesting suitable product combinations to users. The information processing device 10 may be a client device or a server device. The information processing device 10 may also be referred to as a computer, an analysis device, or a product information generation device.

The information processing device 10 has a memory unit 11 and a processing unit 12. The memory unit 11 may be a volatile semiconductor memory such as RAM (Random Access Memory), or a non-volatile storage such as an HDD (Hard Disk Drive) or flash memory. The processing unit 12 is a processor such as a CPU (Central Processing Unit), GPU (Graphics Processing Unit), or DSP (Digital Signal Processor). However, the processing unit 12 may also include electronic circuits such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array). The processor executes a program stored in memory such as RAM (which may be the memory unit 11). A collection of processors may be called a multiprocessor or simply a "processor."

The memory unit 11 stores multiple pieces of transaction data including transaction data 13a, 13b, and 13c, and multiple pieces of dialogue data including dialogue data 14a, 14b, and 14c. Transaction data 13a, 13b, and 13c each represent a transaction of at least some of multiple products including products A, B, C, D, and E. Dialogue data 14a, 14b, and 14c each represent communication related to the transaction using natural language. Dialogue data 14a, 14b, and 14c correspond one-to-one to transaction data 13a, 13b, and 13c.

Here, a "product" is any object of trade in which multiple units of the same type can be distributed. Therefore, a product may be a tangible item such as food or a book, or an intangible item such as content or a service. Furthermore, a "transaction" may be any contract in which rights related to a product are transferred. Therefore, a transaction may be a paid contract in which a fee is paid, or a free contract in which no fee is paid. Furthermore, a transaction may be a transfer contract in which the product is not returned, or a loan contract in which the product is returned at a later date. Furthermore, a transaction may be conducted face-to-face in a store or online. An example of a product transaction is the purchase of groceries at a supermarket.

Transaction data 13a, 13b, and 13c each indicate one or more products that were the subject of a single past transaction. The multiple transactions indicated by transaction data 13a, 13b, and 13c may have taken place at different times, in different locations, or by different users. If the transaction is a product purchase, transaction data 13a, 13b, and 13c may be referred to as a purchase record. Transaction data 13a, 13b, and 13c may be obtained from a sales management system such as a POS (Point of Sale) system.

The communication represented by dialogue data 14a, 14b, and 14c may be a dialogue between a user and a computer such as AI (Artificial Intelligence), or a dialogue between a human operator and a user. The communication may be performed by voice or by text input. Dialogue data 14a, 14b, and 14c may include recorded voice data, may include character strings converted from voice data using voice recognition, or may include character strings input during the dialogue. Dialogue data 14a, 14b, and 14c may include only the user's speech, or may include both the user's speech and the other party's speech.

The dialogue data 14a, 14b, and 14c may also be acquired from a purchasing support system that supports users in selecting products through communication. The purchasing support system may use AI technology to output appropriate responses to questions received from users. Users may use a purchasing support application installed on a terminal device such as a smartphone to communicate with the purchasing support system and search for products. The dialogue data 14a, 14b, and 14c may also be referred to as a dialogue log.

Dialogue data 14a, 14b, and 14c may each include one or more keywords. Keyword candidates may be fixedly determined in advance, or may be determined relatively by analyzing multiple dialogue data including dialogue data 14a, 14b, and 14c. In the latter case, words whose indicators, such as IDF (Inverse Document Frequency), are above a threshold may be determined as keyword candidates. Keywords are preferably characteristic words that represent the purpose, background, context, etc. of the transaction. For example, keywords could be the name of a dish if the product is food ingredients, or the proper noun of the subject of research if the product is a book.

The correspondence between transaction data 13a, 13b, 13c and dialogue data 14a, 14b, 14c may be determined based on the identity of the user or the degree of time proximity. For example, transaction data and dialogue data for the same user and on the same date are associated. Also, for example, transaction data and dialogue data for the same user with a time difference less than a threshold are associated. It is preferable that the communication indicated by the dialogue data occurs before the transaction indicated by the transaction data.

As an example, transaction data 13a indicates products A, B, C, and D, transaction data 13b indicates products A, B, C, and E, and transaction data 13c indicates products A, B, and D. Dialogue data 14a and 14b each contain keyword X, and dialogue data 14c contains keyword Y. Transaction data 13a corresponds to dialogue data 14a, transaction data 13b corresponds to dialogue data 14b, and transaction data 13c corresponds to dialogue data 14c.

The processing unit 12 analyzes the transaction data 13a, 13b, and 13c and the dialogue data 14a, 14b, and 14c to determine a product group 15 from among multiple products that can be traded. At this time, the processing unit 12 determines co-occurrence relationships between products, where different products are included in the same transaction data, and determines the product group 15 based on the co-occurrence relationships. For example, the processing unit 12 determines, as a product group 15, a combination of products whose co-occurrence frequency is equal to or greater than a threshold.

Product combinations with high co-occurrence frequencies may be determined using algorithms such as the apriori algorithm, k-means, or AGM (Apriori-based Graph Mining). Furthermore, if there are multiple product combinations with co-occurrence frequencies above a threshold, product group 15 may be the product combination with the largest number of products. For example, if the co-occurrence frequencies of products A and B and products A, B, and C are both above a threshold, processing unit 12 may preferentially select products A, B, and C, which have the largest number of products, because the latter combination encompasses the former combination.

However, the processing unit 12 determines the co-occurrence relationship between products among two or more transaction data corresponding to two or more dialogue data that satisfy a specific condition. The two or more dialogue data that satisfy a specific condition may be some of the dialogue data 14a, 14b, and 14c, and the two or more transaction data for determining the co-occurrence relationship may be some of the transaction data 13a, 13b, and 13c. The specific condition may be that the same keywords are included, or that the similarity is greater than or equal to a threshold. The specific condition may be referred to as a similarity condition.

When the product group 15 is determined, the processing unit 12 associates keywords 16 with the product group 15. The associated keywords 16 may be used as tag information for searching the product group 15. The keywords 16 are keywords included in the dialogue data 14a, 14b, and 14c. The keywords 16 may also be keywords included in dialogue data corresponding to transaction data that includes all of the products in the product group 15. In this case, the processing unit 12 may select keywords from the relevant dialogue data whose frequency of appearance is equal to or greater than a threshold.

For example, the processing unit 12 determines the co-occurrence relationship between products by limiting the transaction data 13a, 13b, and 13c to transaction data 13a and 13b corresponding to dialogue data 14a and 14b containing keyword X. The processing unit 12 detects products A, B, and C from the transaction data 13a and 13b as the combination with the largest number of products among combinations whose co-occurrence frequency is equal to or greater than a threshold, and generates a product group 15 containing products A, B, and C. The processing unit 12 selects keyword X included in dialogue data 14a and 14b corresponding to transaction data 13a and 13b containing product group 15, and associates keyword X with product group 15 as keyword 16.

Here, if transaction data 13c is used in addition to transaction data 13a and 13b to determine the co-occurrence relationship, the combination of products A and B has the highest co-occurrence frequency, and therefore products A and B may be determined to be part of product group 15. However, there is a risk that the transaction purpose represented by keyword 16 will not be achieved using products A and B alone, and an inappropriate product group 15 may be generated. On the other hand, by determining the co-occurrence relationship using only transaction data 13a and 13b, an appropriate product group 15 that can achieve the transaction purpose represented by keyword 16 is generated. For example, all the ingredients used for a certain dish may be listed in their entirety.

The processing unit 12 associates the product group 15 with the keyword 16 and outputs it. The processing unit 12 may store the product group 15 and the keyword 16 in non-volatile storage, display it on a display device, or transmit it to another information processing device. The product group 15 and the keyword 16 may be used by the transaction support system. For example, the transaction support system may suggest the product group 15 to the user as a candidate for a product combination. Also, for example, when the user inputs a keyword 16, the transaction support system may search for the product group 15 associated with the keyword 16 and provide a response to the user.

As described above, the information processing device 10 of the first embodiment acquires multiple transaction data and multiple dialogue data. The information processing device 10 determines a product group 15 based on the co-occurrence relationship between products in two or more transaction data corresponding to two or more dialogue data that satisfy specific conditions, and associates keywords 16 in the dialogue data with the product group 15.

This automatically generates product information from past transaction data that indicates optimal product combinations for achieving a certain transaction objective. Furthermore, this product information can be searched. This allows the transaction support system to suggest highly useful product information to users, improving user convenience. Furthermore, the burden of creating product information is reduced compared to manually extracting optimal product combinations. Furthermore, because differences in transaction objectives are taken into account, the risk of extracting inappropriate product combinations is reduced and the accuracy of product information is improved compared to determining co-occurrence relationships between products without using dialogue data.

The specific condition may be that the same keywords are included. This allows for determining co-occurrence relationships between products within transaction data with the same transaction purpose, improving the accuracy of product combinations. The information processing device 10 may also generate correspondence information that associates multiple products with multiple keywords and multiple identifiers that identify transactions, and may determine the above-mentioned co-occurrence relationships based on the correspondence information. This allows the information processing device 10 to efficiently utilize co-occurrence determination algorithms such as the Apriori algorithm.

The information processing device 10 may also select keywords 16 to be associated with the product group 15 from keywords included in the dialogue data corresponding to the transaction data including the product group 15. This associates appropriate keywords that reflect the transaction purpose of the product group 15 with the product group 15, improving the accuracy of keyword searches. Furthermore, the information processing device 10 may detect product combinations whose co-occurrence frequency is equal to or greater than a first threshold when determining the product group 15, while extracting keywords whose occurrence frequency is equal to or greater than a second threshold that is smaller than the first threshold when associating keywords 16. This associates multifaceted keywords with the product group 15, improving the accuracy of keyword searches.

The information processing device 10 may also acquire transaction data from a first system that records product purchases and acquire dialogue data from a second system that supports product selection. This allows data for determining the product group 15 to be collected efficiently. The keywords 16 may also be tag information used to search the product group 15. This allows the product group 15 to be used in keyword searches, improving the usefulness of the product group 15.

Second Embodiment
Next, a second embodiment will be described.
FIG. 2 illustrates an example of an information processing system according to the second embodiment.

The information processing system of the second embodiment supports the purchase of products at a supermarket. In the second embodiment, the products are mainly assumed to be food products. The information processing system includes a POS server 31, a web server 32, an analysis server 33, and clients 34 and 35, all connected to a network 30. The network 30 may include a wide area network such as the Internet, or a local area network (LAN). The analysis server 33 corresponds to the information processing device 10 of the first embodiment.

The POS server 31 is a server computer that manages product sales and payments at supermarket stores. In addition to the POS server 31, the POS system also includes POS terminals and other devices. The POS server 31 saves a purchase record each time a user purchases a product at the store. The purchase record includes a user ID that identifies the user, the time of purchase, and the product names of one or more products purchased. The user ID is used to determine the identity of the user when linking the purchase record with the interaction log of the web server 32, which will be described later.

The user ID may be the electronic payment card number used for payment. The user ID may also be the membership number presented by the user at the time of purchase. Users who are registered as supermarket members are assigned a membership number. The user may present a membership card on which the membership number is recorded. The user may also install a membership application on their mobile terminal device in advance and present the mobile terminal device with the membership number displayed on the screen.

Web server 32 is a server computer that provides services, such as product search services, to help users smoothly purchase products at supermarket stores. Users who use the services of Web server 32 are registered as members in advance and identified by a user ID. Web server 32 accepts search keywords from the user. Web server 32 then proposes to the user a product list showing individual products that match the search keywords, or combinations of products that match the search keywords.

The web server 32 also uses AI technology and natural language processing technology to provide a dialogue service that allows it to converse with the user in natural language. The web server 32 accepts questions from the user expressed as voice or text. The web server 32 then references product information for products sold in the supermarket and replies with an appropriate answer to the question.

For example, the Web server 32 may respond with today's recommended products in response to a user request. For example, the Web server 32 may also respond with a list of ingredients for preparing a dish specified by the user. For example, the Web server 32 may also respond with the location within the store where the product specified by the user is displayed. The Web server 32 saves a dialogue log for each dialogue with the user. The dialogue log includes the user ID, the time of the dialogue, and a string indicating the content of the dialogue. The content of the dialogue includes both utterances from the user and utterances from the system.

The analysis server 33 is a server computer that generates product information used in the services of the web server 32. The analysis server 33 collects multiple purchase records from the POS server 31 and multiple interaction logs from the web server 32. The analysis server 33 associates purchase records and interaction logs related to the same purchasing behavior based on the user ID and time.

The analysis server 33 analyzes pairs of purchase records and conversation logs to automatically generate a product list showing suitable product combinations and assigns search tags to the product list. An example of a product list is a list of ingredients for making a certain dish. An example of a tag is a keyword indicating the name of the dish or the situation in which the dish is recommended. The analysis server 33 transmits the tagged product list to the web server 32. The web server 32 suggests a product list to the user based on the search keywords entered by the user.

The POS server 31, web server 32, and analysis server 33 may be computing resources of an on-premise system owned by the retailer, or may be computing resources of a data center or cloud system under contract with the retailer.

Clients 34 and 35 are client computers used by different users. Clients 34 and 35 may be mobile terminal devices such as smartphones or tablet devices, or personal computers installed in the users' homes. Clients 34 and 35 communicate with web server 32 via network 30 in response to user operations. Communication with web server 32 may take place at the user's home. Furthermore, if clients 34 and 35 are mobile terminal devices, communication with web server 32 may take place outside the user's home, such as inside a supermarket.

Clients 34 and 35 download the application program in advance from an application distribution server (not shown). Clients 34 and 35 communicate as follows in accordance with the application program.

Clients 34 and 35 accept search keywords from the user. Clients 34 and 35 then send the search keywords to Web server 32, receive search results from Web server 32, and display the search results on a display device. Clients 34 and 35 also accept questions from the user. Clients 34 and 35 then send the questions to Web server 32, receive answers from Web server 32, and output the answers. The answers may be displayed on a display device or read aloud.

The user may input search keywords or questions into the client 34, 35 by voice, or may input them as a string of characters into the client 34, 35. In the former case, the client 34, 35 may convert the voice into a string of characters using voice recognition and send it to the web server 32, or may send the voice data to the web server 32.

FIG. 3 is a block diagram illustrating an example of hardware of the analysis server.
The analysis server 33 has a CPU 100, RAM 101, HDD 102, GPU 103, input interface 104, media reader 105, and communication interface 106 connected via a bus. The POS server 31 and web server 32 may also have the same hardware as the analysis server 33. The CPU 100 corresponds to the processing unit 12 in the first embodiment. The RAM 101 or HDD 102 corresponds to the storage unit 11 in the first embodiment.

CPU 100 is a processor that executes program instructions. CPU 100 loads at least a portion of the programs and data stored in HDD 102 into RAM 101 and executes the programs. Analysis server 33 may have multiple processors. A collection of processors may be called a multiprocessor or simply a "processor."

RAM 101 is a volatile semiconductor memory that temporarily stores programs executed by CPU 100 and data used in calculations by CPU 100. The analysis server 33 may also have other types of volatile memory besides RAM.

The HDD 102 is non-volatile storage that stores software programs such as the OS (Operating System), middleware, and application software, as well as data. The analysis server 33 may also have other types of non-volatile storage, such as flash memory or an SSD (Solid State Drive).

The GPU 103 works in conjunction with the CPU 100 to perform image processing and output images to a display 107 connected to the analysis server 33. The display 107 is, for example, a CRT (Cathode Ray Tube) display, a liquid crystal display, or an organic EL (Electro Luminescence) display. Note that other types of output devices, such as a projector or printer, may also be connected to the analysis server 33.

The GPU 103 may also be used as a GPGPU (General Purpose Computing on Graphics Processing Unit). The GPU 103 may execute programs in response to instructions from the CPU 100. The analysis server 33 may have volatile semiconductor memory other than the RAM 101 as GPU memory used by the GPU 103.

The input interface 104 receives input signals from an input device 108 connected to the analysis server 33. The input device 108 is, for example, a mouse, a touch panel, or a keyboard. Multiple input devices may be connected to the analysis server 33.

The media reader 105 is a reading device that reads programs and data recorded on the recording medium 109. The recording medium 109 is, for example, a magnetic disk, an optical disk, or a semiconductor memory. Magnetic disks include flexible disks (FDs) and HDDs. Optical disks include compact discs (CDs) and digital versatile discs (DVDs). The media reader 105 copies the programs and data read from the recording medium 109 to other recording media such as the RAM 101 or the HDD 102. The read programs may be executed by the CPU 100.

The recording medium 109 may be a portable recording medium. The recording medium 109 may be used to distribute programs and data. The recording medium 109 and the HDD 102 may also be referred to as computer-readable recording media.

The communication interface 106 communicates with the POS server 31 and the web server 32 via the network 30. The communication interface 106 may be a wired communication interface connected to a wired communication device such as a switch or router, or a wireless communication interface connected to a wireless communication device such as a base station or access point.

FIG. 4 is a block diagram showing an example of hardware of a client.
The client 34 has a CPU 110, a RAM 111, a flash memory 112, a display 113, a touch panel 114, a microphone 115, a speaker 116, and a communication interface 117, all connected to a bus. The client 35 may have the same hardware as the client 34.

The CPU 110 is a processor that executes program instructions. The RAM 111 is a volatile semiconductor memory that temporarily stores at least a portion of the programs and data. The client 34 may also have other types of volatile memory. The flash memory 112 is non-volatile storage that stores the programs and data. The client 34 may also have other types of non-volatile storage.

The display 113 is an output device that displays images. The display 113 is, for example, a liquid crystal display or an organic EL display. The touch panel 114 is an input device that is placed on top of the display 113. The touch panel 114 detects touch operations by the user and notifies the CPU 110 of the touch position. The microphone 115 is an input device that converts sound into an electrical signal. The speaker 116 is an output device that reproduces sound from the electrical signal. The communication interface 117 communicates with the web server 32 via the network 30. The communication interface 117 is, for example, a wireless communication interface that is connected to a base station or an access point.

Next, the product search service provided by the web server 32 will be described.
FIG. 5 is a diagram showing an example of a search screen for searching a product list.
When the client 34 accesses the web server 32 using an application, a screen 41 is displayed on the display 113 of the client 34. The screen 41 includes an input field for inputting search keywords. The user inputs one or more search keywords into the input field on the screen 41 by voice or text input. In this example, the user wants to immediately purchase groceries for one person's dinner at a supermarket. Therefore, the user inputs the search keywords "dinner" and "living alone." Two or more search keywords are evaluated, for example, as a logical sum (OR).

Client 34 sends the search keywords to Web server 32. Web server 32 searches for product lists that match the received search keywords and sends them to client 34. Screen 42 is then displayed on display 113 of client 34. Screen 42 includes multiple product lists, such as product lists 161 and 162. Product list 161 is a list of ingredients for making curry rice, and includes the product names of four ingredients: curry roux, carrots, potatoes, and pork belly. Product list 162 also includes four product names.

By selecting one of the product lists on screen 42, the user can register all of the products included in the selected product list in "My Shopping List." "My Shopping List" is a memo that lists the product names of the products the user wishes to purchase. By referring to "My Shopping List," the user can efficiently search for the products they want within the supermarket.

In this way, by preparing a tagged product list in advance, the web server 32 can suggest a suitable combination of products that suits the user's purchasing objectives. However, it is a heavy burden for supermarket employees to manually create tagged product lists. Therefore, the analysis server 33 automatically generates the product list. However, typical technologies for automatically generating product lists have the following problems:

FIG. 6 is a diagram showing an example of a search screen for searching for individual products.
A typical product search technology compares a search keyword entered by a user with the product information of each product and suggests multiple products in descending order of suitability. However, such product search technology does not suggest a combination of products to achieve a certain purchasing purpose. For example, when "curry" is entered as a search keyword on screen 43, screen 44 showing the search results is displayed. Only the product name of curry roux, which has a relatively high suitability to the search keyword, is listed at the top of the search results, and the product names of other ingredients, which have a relatively low suitability to the search keyword, are not listed.

As a result, it is difficult to identify the exact number of ingredients needed to make curry rice from the search results on screen 44. Also, while ingredients may not go well together, the search results on screen 44 do not reveal any combinations of ingredients that go well together.

Another possible method is for the search server to reference recipe information listing the ingredients for a dish and obtain the names of ingredients that correspond to the name of the dish specified by the user. However, because the ingredient names listed in the recipe information are abstract and not linked to product names, it is difficult for the search server to suggest specific product combinations. It is also difficult for the search server to search for products using keywords that indicate the purchasing background, such as "dinner" or "living alone."

FIG. 7 shows examples of product combinations for different purchasing purposes.
Simple product list generation techniques extract two or more products that are often purchased in combination from past purchase records. Products that are often purchased in combination can also be said to have a high co-occurrence frequency. However, a large number of purchase records contain a mixture of purchase records with different user purchase purposes. Because the same product may be used for different purposes, simply extracting combinations of products with a high co-occurrence frequency may result in extracting a small set of products that commonly appear in purchase records with different purchase purposes. As a result, a product list with unclear purchase purposes may be generated, which is of little use. The products listed in such a product list alone may be insufficient to achieve a specific purchase purpose.

For example, purchase record 131 indicates products purchased by user a on September 10th, and is assigned a purchase ID of 001 to identify the purchase record. Purchase record 131 includes the product names of curry roux, carrots, potatoes, pork belly, yogurt, and snack foods. User a intends to make curry rice using the products included in purchase record 131. Purchase record 134 indicates products purchased by user c on September 15th, and is assigned a purchase ID of 004. Purchase record 134 includes the product names of carrots, potatoes, pork belly, mirin, and snack foods. User c intends to make meat and potato stew using the products included in purchase record 134.

Here, if product combinations with high co-occurrence frequencies are extracted from purchase records 131 and 134 intended for different dishes, a combination of carrots, potatoes, and pork belly may be extracted. However, this product combination only includes the common ingredients for curry rice and meat and potato stew, and therefore does not achieve the purchasing purpose of cooking. Therefore, a product list showing this combination is not very useful and is inappropriate as a suggestion to the user.

The analysis server 33 therefore analyzes the purchase records collected from the POS server 31 as well as the interaction logs collected from the web server 32. As a result, the analysis server 33 automatically generates an appropriate product list that matches the specific purchasing purpose. The steps taken by the analysis server 33 to generate the product list are described below.

FIG. 8 is a diagram showing an example of the association between a purchase record and a dialogue log.
The analysis server 33 acquires a plurality of purchase records, including the purchase record 131, from the POS server 31. The analysis server 33 also acquires a plurality of dialogue logs, including the dialogue log 141, from the web server 32. The analysis server 33 associates the purchase records and dialogue logs relating to the same purchasing behavior based on the user ID and time.

It is preferable that associated purchase records and dialogue logs satisfy the relationship that the purchase record contains a product selected by the user through the communication indicated by the dialogue log. For example, the analysis server 33 associates a purchase record and a dialogue log that have the same user ID and the same date. The analysis server 33 may employ a constraint that the time of the purchase record is later than the time of the dialogue log. The analysis server 33 may also employ a constraint that the time of the purchase record is within a predetermined time period from the time of the dialogue log.

Here, the analysis server 33 associates the purchase record 131 with the dialogue log 141. In the dialogue log 141, user A asks about dinner options, and the system suggests curry and stir-fried meat and vegetables. User A also asks where carrots are sold, and the system replies that carrots are sold in the vegetable section on the first floor.

FIG. 9 is a diagram showing an example of generating a keyword list from a dialogue log.
The analysis server 33 generates a keyword list from each of the multiple dialogue logs associated with the purchase record. The keyword list is a list of keywords included in the dialogue logs. First, the analysis server 33 extracts words that appear in the dialogue logs and calculates the IDF of each word. The IDF of a word is the reciprocal of the number of dialogue logs that include that word. Words with a large IDF are characteristic words that are only included in a small number of dialogue logs, while words with a small IDF are general words that are included in a large number of dialogue logs.

The analysis server 33 selects words whose IDF is equal to or greater than the threshold value v as keywords. However, the analysis server 33 may select keywords using other indicators that indicate the importance of words. Alternatively, keyword candidates may be manually defined in advance. The analysis server 33 extracts keywords for each dialogue log and generates a keyword list.

Here, the analysis server 33 generates keyword list 151 from dialogue log 141, generates keyword list 152 from dialogue log 142, and generates keyword list 153 from dialogue log 143. Keyword list 151 includes the keywords "dinner," "pork," "curry," "stir-fried meat and vegetables," and "vegetables." Keyword list 152 includes the keywords "curry" and "recipe." Keyword list 153 includes the keywords "beer," "snacks," and "potato crackers."

FIG. 10 is a diagram showing an example of the association between a purchase record and a keyword list.
The analysis server 33 associates a purchase record with a keyword list generated from the interaction log corresponding to the purchase record. Here, the analysis server 33 associates the purchase record 131 with the keyword list 151, and associates the purchase record 132 with the keyword list 152. The analysis server 33 also associates the purchase record 133 with the keyword list 153, and associates the purchase record 134 with the keyword list 154.

Purchase record 132 indicates products purchased by user b on September 12th, and is assigned a purchase ID of 002. Purchase record 132 includes the product names of curry roux, carrots, potatoes, pork belly, beer, and salad. Purchase record 133 indicates products purchased by user a on September 12th, and is assigned a purchase ID of 003. Purchase record 133 includes the product names of pork belly, beer, and snacks. Keyword list 154 includes the keywords "nikujaga" (meat and potatoes) and "living alone."

FIG. 11 is a diagram showing an example of a collection of pairs of product names and keywords.
The analysis server 33 breaks down purchase records with attached keyword lists into pairs of product names included in the purchase records and keywords included in the keyword list. From a purchase record containing n product names and a keyword list containing m keywords, n x m pairs of product names and keywords are comprehensively generated. The analysis server 33 aggregates the product name and keyword pairs generated from multiple purchase records to generate table 45. Table 45 includes multiple records that associate product names, keywords, and one or more purchase IDs. The purchase IDs in table 45 are identifiers that identify the purchase records from which product name and keyword pairs were extracted.

For example, the analysis server 33 extracts pairs such as (curry roux, curry), (carrot, curry), (potato, curry), and (pork belly, curry) from the purchase record 131. The analysis server 33 also extracts pairs such as (curry roux, curry), (carrot, curry), (potato, curry), and (pork belly, curry) from the purchase record 132. The analysis server 33 also extracts pairs such as (carrot, nikujaga), (potato, nikujaga), and (pork belly, nikujaga) from the purchase record 134.

In this way, (curry roux, curry), (carrot, curry), (potato, curry), and (pork belly, curry) are extracted from purchasing records 131 and 132, but not from purchasing record 134. (carrot, meat and potato stew), (potato, meat and potato stew), and (pork belly, meat and potato stew) are extracted from purchasing record 134, but not from purchasing records 131 and 132. This is because the dishes purchased are different.

The analysis server 33 divides the multiple records contained in the table 45 into groups of records with the same keyword. For example, the analysis server 33 divides the records into a group of "dinner," a group of "curry," a group of "appetizers," and a group of "meat and potato stew." This can also be said to group the correspondence between products and purchase records by purchase purpose.

Analysis server 33 determines the largest product group whose co-occurrence frequency is equal to or greater than threshold k1 among records with the same keyword. Here, the co-occurrence frequency of a product group may be the number of purchase records that include that product group, or the proportion of purchase records that include that product group among purchase records related to the keyword.

When there are multiple product groups with co-occurrence frequencies above threshold k1, the product group that is not a subset of other product groups is the maximal product group. For example, if the co-occurrence frequencies of products A and B and products A, B, and C are both above threshold k1, products A and B are not maximal product groups because they are included in products A, B, and C. By determining which product groups are maximal, small product groups that are insufficient to achieve the purchasing objective are eliminated.

The largest product groups with high co-occurrence frequencies are identified within the same keyword, but are not identified across different keywords. This reduces the possibility of extracting insufficient product combinations that are merely common to multiple purchasing purposes.

For example, in the example of Figure 11, let's assume that the threshold k1 is 2 times. For the keyword "nikujaga" (meat and potato stew), the combination of carrots, potatoes, and pork belly is not extracted as a product group because it is only included in purchase record 134. For the keyword "curry," the combination of curry roux, carrots, potatoes, and pork belly is extracted as a maximum product group because it is included in purchase records 131 and 132. Therefore, the insufficient product combination of carrots, potatoes, and pork belly is not extracted.

The analysis server 33 uses an algorithm such as the Apriori algorithm to determine the largest product groups with the highest co-occurrence frequency. The Apriori algorithm is also described in the following non-patent document: Rakesh Agrawal and Ramakrishnan Srikant, "Fast Algorithms for Mining Association Rules," Proc. of the 20th International Conference on Very Large Data Bases (VLDB '94), pp. 487-499, September 12, 1994.

FIG. 12 is a diagram showing an example of the execution of the Apriori algorithm.
In the Apriori algorithm, the analysis server 33 first calculates the appearance frequency of each product and generates a set of products whose appearance frequency is equal to or greater than a threshold k1 as a list _L1 . The analysis server 33 repeatedly generates lists _L1+1 from lists _L1 by the following procedure.

The analysis server 33 extracts size i+1 combinations in which i+1 different products are selected from the products appearing in list L _i , and all of the size i combinations, which are a subset of the size i combinations, are included in list L _i . The analysis server 33 generates list C _i+1 by comprehensively extracting size i+1 combinations that satisfy this condition. The analysis server 33 extracts product combinations from list C _i+1 whose appearance frequency is equal to or greater than threshold k1, and generates list L _i+1 . If list L _i+1 is an empty set, the analysis server 33 terminates processing, and if list L _i+1 is not an empty set, it increments i by 1 and continues processing.

As an example, data 50 includes purchase records #1 to #7. Purchase record #1 includes apples, mandarins, and watermelon. Purchase record #2 includes mandarins and grapes. Purchase record #3 includes apples, mandarins, grapes, and bananas. Purchase record #4 includes apples, bananas, and watermelon. Purchase record #5 includes apples, mandarins, and bananas. Purchase record #6 includes grapes and bananas. Purchase record #7 includes mandarins.

The analysis server 33 calculates the frequency of appearance of each product. The frequency of appearance of mandarins is 5, the frequency of appearance of apples is 4, the frequency of appearance of bananas is 4, the frequency of appearance of grapes is 3, and the frequency of appearance of watermelons is 2. Here, the threshold k1 is assumed to be 3 times. Then, the analysis server 33 generates a list 51 as _L1 . The list 51 includes mandarins, apples, bananas, and grapes.

The analysis server 33 generates a list 52 as _C2 based on the list 51. The list 52 includes the combinations (apple, mandarin orange), (apple, banana), (mandarin orange, grape), (mandarin orange, banana), (grape, banana), and (apple, grape).

The analysis server 33 calculates the frequency of occurrence of each combination included in the list 52. The frequency of occurrence of (apple, mandarin orange) is 3, the frequency of occurrence of (apple, banana) is 3, the frequency of occurrence of (mandarin orange, grape) is 2, the frequency of occurrence of (mandarin orange, banana) is 2, the frequency of occurrence of (grapes, banana) is 2, and the frequency of occurrence of (apple, grape) is 1. The analysis server 33 extracts from the list 52 combinations whose frequency of occurrence is equal to or greater than the threshold k1, and generates a list 53 as _L2 . The list 53 includes (apple, mandarin orange) and (apple, banana).

The analysis server 33 generates a list 54 as C3 based on the list _53. The list 54 includes the combination (apple, mandarin orange, banana). The analysis server 33 calculates the frequency of occurrence of each combination included in the list 54. The frequency of occurrence of (apple, mandarin orange, banana) is 2. The analysis server 33 extracts combinations from the list 54 whose frequency of occurrence is equal to or greater than the threshold k1, and generates a list 55 as _L3 . The list 55 is an empty set. This completes the Apriori algorithm for the data 50. In this example, the product groups with the highest co-occurrence frequency of 3 or more are (apple, mandarin orange) and (apple, banana).

FIG. 13 is a diagram showing an example of extraction of keywords corresponding to a product list.
Once the product group is determined, the analysis server 33 generates a product list that lists the product names of the products included in the product group.The analysis server 33 then refers to the purchase record with the keyword list and associates search tags with the product list.

In this example, the analysis server 33 generates a product list 161 that includes the product names of curry roux, carrots, potatoes, and pork belly. The analysis server 33 searches for purchase records 131, 132, 135, etc. that include all of the products listed in the product list 161, and reads out keyword lists 151, 152, 155, etc. that correspond to the searched purchase records 131, 132, 135, etc. The analysis server 33 aggregates the keywords included in the read keyword lists 151, 152, 155, etc. to generate table 46.

Table 46 associates keywords with their frequency of appearance. The frequency of appearance of a keyword may be the number of keyword lists that include that keyword, or the proportion of keyword lists that include that keyword among the keyword lists being aggregated. The analysis server 33 associates keywords whose frequency of appearance is equal to or greater than threshold k2 with the product list 161 as tags.

Keywords with an appearance frequency equal to or greater than threshold k2 typically include keywords used as criteria for determining product groups. For example, the four products included in product list 161 are grouped under the keyword "curry." Therefore, "curry" is included among the tags associated with product list 161.

On the other hand, keywords with an appearance frequency equal to or greater than the threshold k2 may include keywords other than those used as criteria when determining product groups. Such keywords may not represent a direct purpose, such as the name of a dish, but may represent an indirect purchasing background, such as the time of day the meal is served or the number of people present. Keywords that represent an indirect purchasing background are also useful when users search product lists. By setting the threshold k2 used to assign keywords to a value smaller than the threshold k1 used to determine product groups, the analysis server 33 can extract many such useful keywords.

FIG. 14 is a diagram showing an example of association between a product list and a tag.
For example, the analysis server 33 associates tags 171 and 172 with the product list 161. The tag 171 represents the keyword "curry," and the tag 172 represents the keyword "dinner." The analysis server 33 also associates tags 173, 174, and 175 with the product list 162. The tag 173 represents the keyword "prepared food," the tag 174 represents the keyword "time saving," and the tag 175 represents the keyword "living alone."

The analysis server 33 also associates tags 176 and 177 with product list 163. Tag 176 represents the keyword "curry," and tag 177 represents the keyword "family-friendly." The analysis server 33 also associates tags 178 and 179 with product list 164. Tag 178 represents the keyword "curry," and tag 179 represents the keyword "living alone." In this way, even for the same dish, curry rice, different product lists can be suggested depending on the number of people dining. Furthermore, different product lists corresponding to dishes with different cooking methods can be suggested to a user living alone.

Next, the functions and processing procedures of the information processing system will be described.
FIG. 15 is a block diagram illustrating an example of functions of the information processing system.
The POS server 31 has a POS data storage unit 120 and a transaction execution unit 125. The POS data storage unit 120 is implemented, for example, using RAM or a HDD included in the POS server 31. The transaction execution unit 125 is implemented, for example, using a CPU and a program included in the POS server 31.

The POS data storage unit 120 stores the purchase records such as the aforementioned purchase records 131, 132, 133, and 134. The purchase records are sent from the POS server 31 to the analysis server 33.
The transaction execution unit 125 works in conjunction with the POS terminal to execute the purchase and sale of goods at supermarket stores. For each transaction, the transaction execution unit 125 generates a purchase record that lists the goods sold to the user. The transaction execution unit 125 assigns a purchase ID that identifies the purchase record and the current time to the purchase record. The transaction execution unit 125 also assigns a user ID to the purchase record based on an item presented by the user, such as a membership card. The transaction execution unit 125 saves the purchase record in the POS data storage unit 120.

The web server 32 has a dialogue log storage unit 121, a product information storage unit 122, a dialogue execution unit 126, and a product search unit 127. The dialogue log storage unit 121 and the product information storage unit 122 are implemented, for example, using RAM or a HDD provided by the web server 32. The dialogue execution unit 126 and the product search unit 127 are implemented, for example, using a CPU and programs provided by the web server 32.

The dialogue log storage unit 121 stores dialogue logs such as the aforementioned dialogue logs 141, 142, and 143. The dialogue logs are transmitted from the web server 32 to the analysis server 33. The product information storage unit 122 stores product lists such as the aforementioned product lists 161, 162, 163, and 164. The product lists are transmitted from the analysis server 33 to the web server 32.

The dialogue execution unit 126 receives questions from the clients 34 and 35, generates answers corresponding to the questions, and transmits them to the clients 34 and 35. The dialogue execution unit 126 records a series of utterances including both the user's questions and the system's answers as a dialogue log. The dialogue execution unit 126 adds the current time to the dialogue log. The dialogue execution unit 126 also adds a user ID to the dialogue log based on the login information used by the user to access the web server 32. The dialogue execution unit 126 saves the dialogue log in the dialogue log storage unit 121.

The product search unit 127 receives search keywords from the clients 34 and 35 and searches the product information storage unit 122 for product lists with tags that match the search keywords. The product search unit 127 transmits the searched product lists to the clients 34 and 35.

The analysis server 33 has a history memory unit 123, a product information memory unit 124, a communication unit 128, and a product information generation unit 129. The history memory unit 123 and the product information memory unit 124 are implemented, for example, using the RAM 101 or HDD 102 of the analysis server 33. The communication unit 128 and the product information generation unit 129 are implemented, for example, using the CPU 100, communication interface 106, programs, etc. of the analysis server 33.

The history storage unit 123 stores purchase records received from the POS server 31 and conversation logs received from the web server 32. The product information storage unit 124 stores the product list generated by the analysis server 33. This product list is sent to the web server 32.

The communication unit 128 communicates with the POS server 31 and the web server 32. The communication unit 128 receives purchase records from the POS server 31 and stores the received purchase records in the history storage unit 123. The communication unit 128 also receives conversation logs from the web server 32 and stores the received conversation logs in the history storage unit 123.

The product information generation unit 129 reads purchase records and dialogue logs from the history storage unit 123 and associates purchase records and dialogue logs related to the same purchasing behavior based on the user ID and time. The product information generation unit 129 analyzes the associated purchase records and dialogue logs to generate one or more product lists indicating recommended product combinations. The product information generation unit 129 also assigns search tags to the product lists. The product information generation unit 129 saves the generated product lists in the product information storage unit 124.

FIG. 16 is a flowchart showing an example of a procedure for generating a product list.
(S10) The product information generation unit 129 acquires a plurality of purchase records and a plurality of conversation logs.
(S11) The product information generation unit 129 compares the user ID and time included in the purchase record with the user ID and time included in the interaction log, and associates the purchase record and interaction log related to the same purchasing behavior. For example, the product information generation unit 129 associates the purchase record and interaction log with the same user ID and the same date.

(S12) The product information generation unit 129 extracts words from multiple dialogue logs, calculates the importance of each word, and determines words whose importance is equal to or greater than a threshold value v as keywords. For example, the product information generation unit 129 determines words whose IDF is equal to or greater than a threshold value v as keywords.

(S13) For each purchase record, the product information generation unit 129 extracts keywords from the dialogue log corresponding to the purchase record to generate a keyword list, and associates the keyword list with the purchase record.
(S14) The product information generation unit 129 comprehensively lists pairs of one product name included in the purchase record and one keyword included in the keyword list from each pair of associated purchase record and keyword list.

(S15) The product information generation unit 129 tallys up pairs of product names and keywords and determines the correspondence between the product names and purchase records containing those product names for each keyword. The product information generation unit 129 determines product groups whose co-occurrence frequency within the same keyword is equal to or greater than threshold k1. For example, an apriori algorithm is used for this determination.

(S16) The product information generation unit 129 extracts the product group with the largest number of products by removing small product groups that are subsets of other product groups from the product groups determined in step S15.

(S17) The product information generation unit 129 generates one or more product lists indicating the largest product group. For each product list, the product information generation unit 129 searches for purchase records that include all of the products in that product list. The product information generation unit 129 tallies the keywords included in the keyword lists corresponding to the searched purchase records, and calculates the frequency of appearance of each keyword.

(S18) The product information generation unit 129 generates tags indicating keywords whose appearance frequency is equal to or greater than the threshold k2 for each product list, and associates the tags with the product list.
(S19) The product information generation unit 129 outputs the tagged product list. At this time, the product information generation unit 129 stores the tagged product list in the product information storage unit 124. The product information generation unit 129 may display the tagged product list on the display 107.

As described above, the information processing system of the second embodiment proposes a product list showing suitable combinations of products, rather than individual products, in response to search keywords entered by the user. This improves convenience for users purchasing products. Furthermore, the product list is automatically generated by the analysis server 33, which reduces the burden on retailers that provide product search services. Furthermore, because search tags are automatically assigned to the product list, keyword searches of the product list are possible.

The analysis server 33 also associates the purchase records generated by the POS system with the dialogue logs generated by the dialogue system, and generates a product list by analyzing the purchase records and dialogue logs together. In this case, the analysis server 33 calculates the co-occurrence frequency between products in purchase records corresponding to dialogue logs containing the same keywords, rather than the co-occurrence frequency between products in all purchase records. This reduces noise caused by the mixing of purchase records with different purchasing purposes, and lowers the risk of extracting inappropriate product combinations. In particular, it reduces the risk of generating an insufficient product list in which none of the purchasing purposes are achieved.

The analysis server 33 also assigns keywords from the dialogue log corresponding to purchase records that include all products in the product list as tags to the product list. This allows tags that are useful for keyword searches to be assigned to the product list. The dialogue log also contains not only keywords that directly indicate the purpose of the purchase, such as the name of a dish, but also a variety of keywords that indicate the background to the purchase, such as the time of day the meal is eaten and the number of people present. This allows the information processing system to suggest to the user an appropriate product list based on various search keywords. Furthermore, by lowering the threshold for the frequency of appearance of keywords used as tags, a variety of tags can be assigned to the product list.

REFERENCE SIGNS LIST 10 Information processing device 11 Storage unit 12 Processing unit 13a, 13b, 13c Transaction data 14a, 14b, 14c Dialogue data 15 Product group 16 Keywords

Claims (9)

acquiring a plurality of transaction data each indicating a transaction of at least some of a plurality of products and a plurality of dialogue data corresponding to the plurality of transaction data, each indicating communication related to the transaction using natural language, each including a question sentence and an answer sentence corresponding to the question sentence ;
determining a product group from the plurality of products based on a co-occurrence relationship between products in two or more transaction data among the plurality of transaction data corresponding to two or more dialogue data among the plurality of dialogue data that satisfy a specific condition;
Associating at least some of the keywords included in the plurality of pieces of dialogue data with the determined product group;
A product information generation program that causes a computer to execute processing. The specific condition is that the same keywords are included.
The product information generating program according to claim 1. In determining the product group, correspondence information is generated based on the plurality of transaction data and the plurality of conversation data, in which correspondence information is generated between the plurality of products, the plurality of keywords, and a plurality of identifiers that identify the transactions, and the co-occurrence relationship is determined based on the correspondence information.
The product information generating program according to claim 1. In associating at least some of the keywords, a keyword to be associated with the product group is selected from keywords included in dialogue data corresponding to transaction data including the product group among the plurality of dialogue data.
The product information generating program according to claim 1. acquiring a plurality of transaction data each indicating a transaction of at least some of a plurality of products and a plurality of dialogue data corresponding to the plurality of transaction data each indicating communication related to the transaction using natural language;
determining a product group from the plurality of products based on a co-occurrence relationship between products in two or more transaction data among the plurality of transaction data corresponding to two or more dialogue data among the plurality of dialogue data that satisfy a specific condition;
Associating at least some of the keywords included in the plurality of pieces of dialogue data with the determined product group;
Have the computer execute the process ,
In determining the product group, a combination of products whose co-occurrence frequency is equal to or greater than a first threshold is detected among the two or more pieces of transaction data;
In associating at least some of the keywords, keywords whose appearance frequency is equal to or greater than a second threshold value that is smaller than the first threshold value are extracted from the dialogue data corresponding to the transaction data including the product group.
Product information generation program. acquiring a plurality of transaction data each indicating a transaction of at least some of a plurality of products and a plurality of dialogue data corresponding to the plurality of transaction data each indicating a communication related to the transaction using natural language;
determining a product group from the plurality of products based on a co-occurrence relationship between products in two or more transaction data among the plurality of transaction data corresponding to two or more dialogue data among the plurality of dialogue data that satisfy a specific condition;
Associating at least some of the keywords included in the plurality of pieces of dialogue data with the determined product group;
Have the computer execute the process ,
The plurality of transaction data are acquired from a first system that records product purchases by users, and the plurality of interaction data are acquired from a second system that supports product selection by the users through the communication with the users.
Product information generation program. acquiring a plurality of transaction data each indicating a transaction of at least some of a plurality of products and a plurality of dialogue data corresponding to the plurality of transaction data each indicating communication related to the transaction using natural language;
determining a product group from the plurality of products based on a co-occurrence relationship between products in two or more transaction data among the plurality of transaction data corresponding to two or more dialogue data among the plurality of dialogue data that satisfy a specific condition;
Associating at least some of the keywords included in the plurality of pieces of dialogue data with the determined product group;
Have the computer execute the process ,
the at least some of the keywords associated with the product group are tag information used to search for the product group;
Product information generation program. acquiring a plurality of transaction data each indicating a transaction of at least some of a plurality of products and a plurality of dialogue data corresponding to the plurality of transaction data, each indicating communication related to the transaction using natural language, each including a question sentence and an answer sentence corresponding to the question sentence ;
determining a product group from the plurality of products based on a co-occurrence relationship between products in two or more transaction data among the plurality of transaction data corresponding to two or more dialogue data among the plurality of dialogue data that satisfy a specific condition;
Associating at least some of the keywords included in the plurality of pieces of dialogue data with the determined product group;
A product information generating method in which processing is performed by a computer. a storage unit that stores a plurality of transaction data each indicating a transaction of at least some of a plurality of products, and a plurality of dialogue data corresponding to the plurality of transaction data, each indicating communication related to the transaction using natural language, and each including a question sentence and an answer sentence corresponding to the question sentence ;
a processing unit that determines a product group from among the plurality of products based on a co-occurrence relationship between products in two or more pieces of transaction data among the plurality of transaction data corresponding to two or more pieces of dialogue data that satisfy a specific condition among the plurality of pieces of dialogue data, and associates at least some of a plurality of keywords included in the plurality of pieces of dialogue data with the determined product group;
An information processing device having the above.