JP7759630B2 - Information processing device and program - Google Patents

Description

This disclosure relates to an information processing device and a program.

Currently, e-commerce sites (online sales) for apparel and other products are becoming commonplace, and it is expected that in the future there will be more opportunities for virtual stores to open in shopping malls in virtual spaces (metaverses) on the Internet. In this environment, just like in brick-and-mortar stores, it is important from the perspective of sales promotion for e-commerce sites to have staff interact with users to recommend products, uncover user needs and help them realize the products they are looking for, and provide one-on-one customer service to respond to various inquiries.

Generally, the number of users visiting an e-commerce site is far greater than the number of customers visiting a physical store. Therefore, it is not realistic for real staff to provide individual customer service to every user who visits an e-commerce site. Therefore, it is envisioned that virtual staff (also known as digital staff) will be created on a computer, and these digital staff will provide one-on-one customer service to users in place of real staff.

While various characters are set for digital staff, it is not easy to create a character with customer service skills. Therefore, characters are sometimes set to resemble real staff members.

However, no matter how closely the virtual staff characters resemble those of real staff members, they often differ clearly from real staff members in many ways, including the content of their statements and the way they speak.

The goal is to enable virtual users, acting as surrogates for real users, to interact with others as naturally as if they were real users.

The information processing device according to this embodiment includes:
An information processing device that is connected to a user terminal, at least one server selected from an SNS server, a blog site server, a word-of-mouth site server, and an EC site server, and a statement generation device (generation AI) via a communication line, and that allows a virtual user acting as a representative of a real user to interact with other real users,
a first receiving means for receiving data of comments from the other users directly from the user terminal or indirectly via at least one server selected from the SNS server, the blog site server, the word-of-mouth site server, and the EC site server;
a first transmission means for transmitting to the message generation device a request for generating a message from the virtual user in response to the message from the other user, the request including data of the message from the other user;
a second receiving means for receiving data of the message from the message generation device;
a second transmission means for transmitting the received data of the comment of the virtual user as a message of the virtual user to the user terminal or to at least one server selected from the SNS server, the blog site server, the word-of-mouth site server, and the EC site server;
means for extracting character strings of nouns by performing natural language analysis processing on each of comments posted by other users and on at least one of the SNS server, the blog site server, the review site server, and the EC site server;
a feature calculation means for calculating a feature amount that quantifies content features of each of the statements and comments of the other users based on the extracted nouns;
and means for selecting, from the plurality of comments, at least one specific comment having content characteristics closest to the comment of the other user based on the feature amount;
The first transmission means transmits data of the selected specific comment, in addition to data of the other user's comments, to the message generation device by attaching the data to the generation request as reference information.

FIG. 1 is a configuration diagram of an entire system including an information processing apparatus according to this embodiment. FIG. 2 is a diagram showing the physical configuration of the information processing apparatus shown in FIG. FIG. 3 is a diagram showing the functional configuration of the information processing apparatus of FIG. FIG. 4 is a diagram showing the data flow of the entire system of FIG. 1 and the processing of the information processing device. FIG. 5 is a diagram showing an example of the profile of FIG. FIG. 6 is a diagram showing an example of the posted comments and their feature amounts shown in FIG. FIG. 7 is a diagram showing an example of the user message of FIG. 4 and its feature quantities. FIG. 8 is a diagram showing an example of the distance between a user message and a posted comment in an N-dimensional space showing the feature amounts shown in FIGS. 6 and 7. In FIG. FIG. 9 is a diagram schematically showing the flow of processing by the information processing apparatus of FIG.

Embodiments of the present invention will now be described with reference to the drawings. This embodiment uses comments (referred to as posts, articles, reviews, reviews, messages, etc.) actually created and posted by real users on social networking services (hereinafter referred to as "SNS"), which are community-based services that promote and support connections between people; blog sites where users post articles about everyday matters and thoughts and display them chronologically; review sites where users can write and view reviews of products and services; and e-commerce sites that sell products and services online. The AI, which functions as a statement generation device, generates statements in response to statements made by other users, thereby ensuring that the content of the statements is consistent with the statements made by other users, and that the content of the virtual user's statements sounds as if they were made by the real user they represent, using language that is natural and would be used by the real user.

For example, typical examples of social networking sites include Instagram (registered trademark), Facebook (registered trademark), and Twitter (currently X (registered trademark)). Typical examples of review sites include Tabelog (registered trademark) and Kakaku.com (registered trademark). There are also many blog sites, such as FC2 (registered trademark) blogs.

The following explanation will take as an example a scenario in which clothing and other items are sold via an EC (electronic commerce) site that offers services for selling products and services on online websites via networks such as the Internet. In this scenario, a virtual user (hereinafter referred to as "virtual staff") acting on behalf of a real user (hereinafter referred to as "real staff") interacts in a chat room with a real user (other user) visiting the EC site, and attempts to sell clothing and other items by serving the other user.

As shown in FIG. 1, the information processing device 1 according to this embodiment is connected via a communication network, typically the Internet network 6, to a comment generation device 2 that functions as a generation AI, an SNS server 3-1 that provides an SNS, a blog site server 3-2 that provides blog services, a review site server 3-3 that operates a review site, an EC site server 4 that operates an EC site, and a user terminal 5 used by an actual user.

As shown in FIG. 2, the information processing device 1 as an interactive device has a processor 11 connected to RAM 12, ROM 13, storage unit 14, input device 15, display 16, and communication unit 17 via a system bus 10. The processor 11 is composed of, for example, a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The processor 11 executes programs loaded from the storage unit 14 and ROM 13 to RAM 12 to perform interactive processing that realizes a dialogue with a real user. RAM 12 functions as the main memory, work area, etc. of the processor 11. The ROM 13 or storage unit 14 stores the BIOS (Basic Input Output System), operating system program (OS), the interactive processing program according to this embodiment, other programs for realizing various functions, and various data required for these processes, all of which are executed by the processor 11.

The input device 15 consists of a keyboard (KB), a mouse, a touch panel, or other pointing device. The display 19 is typically implemented as an LCD (Liquid Crystal Display). In addition to the dialogue processing program, the memory unit 14 stores data related to comments required for dialogue processing, their respective features, profiles of real staff members, and messages exchanged between real users and virtual staff members, as well as their order (dialogue history).

As shown in FIG. 3, by executing the dialogue processing program, the processor 11 functions as a control unit 20, a profile acquisition unit 21, a posted comment collection unit 22, a natural language analysis processing unit 23, a feature calculation processing unit 24, a user message receiving unit 25, a posted comment selection processing unit 26, a message generation request unit 27, a message receiving unit 28, and a message sending unit 29.

The profile acquisition unit 21 sends a profile transmission request along with each account of the actual staff member to each of the SNS server 3-1, blog site server 3-2, review site server 3-3, and EC site server 4, and receives profile data from each of the SNS server 3-1, blog site server 3-2, review site server 3-3, and EC site server 4. These profiles are compiled into a single profile, associated with the identification number of the virtual staff member representing the actual staff member, and stored in the storage unit 14.

The posted comment collection unit 22 sends requests to the SNS server 3-1, blog site server 3-2, review site server 3-3, and EC site server 4 to send comments (such as posts, articles, reviews, ratings, and messages) posted by real staff members along with their respective accounts, and receives data on comments actually posted by real staff members from the SNS server 3-1, blog site server 3-2, and review site server 3-3. These posted comments are stored in the storage unit 14 in association with the identification number (ID) of the virtual staff member representing the real staff member.

The natural language analysis processor 23 performs natural language analysis on posted comments and messages, breaking them down into parts of speech and extracting noun strings. The feature calculation processor 24 calculates feature values that quantify the content characteristics of each posted comment based on the extracted nouns. The feature values are typically coordinates in an N-dimensional space. Multiple types of divisions corresponding to theme categories are associated with the N axes that make up the N-dimensional space. Multiple nouns related to each division are assigned to each axis along with their coordinate values. For example, multiple divisions are predefined for each category, such as apparel, music, health foods, books, food, and home appliances. For the apparel category, as shown in Figure 8, each axis corresponds to a generic, higher-level conceptual division such as time, inventory, size, design, item, color, price range, era, and event. Furthermore, each division is assigned a number of nouns (strings) representing specific, lower-level conceptual matters encompassed by that division. The multiple nouns assigned to each axis are ranked according to semantic proximity, and each is assigned a numerical value (coordinate value) in advance. The noun strings extracted from the comment (text) are compared with the nouns on the N-axis of the N-dimensional space, and if a corresponding noun string exists, the coordinate value corresponding to that noun string is given. Of course, if a corresponding noun string does not exist, the coordinate value on that coordinate axis is given a value of zero. Comments with similar features are similar in content, and can be said to have similar content features.

In addition, the feature calculation process may use comments and messages as input data and use a pre-trained model that is trained to output feature data that quantifies the content features of the comments and messages.

The user message receiving unit 25 receives messages (comments) from real users from the user terminal 5 either directly from the user terminal 5 or indirectly via the EC site server 4. In practice, messages from real users are received indirectly via at least one of the SNS server 3-1, blog site server 3-2, word-of-mouth site server 3-3, and EC site server 4, which serve as platforms where real users interact with virtual staff. These messages are subjected to natural language analysis by the natural language analysis processing unit 23, and their features are calculated by the feature calculation processing unit 24.

The posted comment selection processing unit 26 selects one comment that is closest to the characteristics of the real user's message, or selects a predetermined number of comments that are closest to the characteristics of the real user's message. The selected comments have content characteristics that are similar to the real user's message (statement).

The message generation request unit 27 transmits data on the real user's message (message) to the message generation device 2 along with a request to generate a message from a virtual user in response to the real user's message. In addition to the message, the message generation request unit 27 transmits the profile of the real staff member to the message generation device 2. Furthermore, in addition to the real user's message and the real staff member's profile, the message generation request unit 27 also attaches at least one comment that the real staff member actually posted on SNS or the like and that has similar content characteristics to the real user's message as reference information to the generation request and transmits this to the message generation device 2.

Here, the following requirements are necessary for a normal dialogue to take place between the real user and the virtual staff member, and for the real user to perceive the dialogue with the virtual staff member as natural, even if it is a dialogue with a real staff member represented by the virtual staff member.

1) The comments made by the virtual staff are consistent in content, responding to the comments (messages) made by real users.

2) The statements made by the virtual staff are similar to what real staff would say.

3) The language used by virtual staff reflects the language used by real staff.

These three requirements are met by sending comments that real staff have actually posted on social media etc. and that are similar in content to the real user's message, in addition to the real user's comments (messages) and the real staff's profiles, as reference information to the comment generation device 2, and the comment generation device 2 then uses this reference information to generate a comment from a virtual staff member in response to the real user's message.

The message receiving unit 28 receives message data generated by the message generation device 2 from the message generation device 2. The message sending unit 29 sends the message received from the message generation device 2 as is or after appropriate modification to the user terminal 5 or the EC site server 4 as a message from the virtual staff member to the real user.

Figure 4 shows the dialogue processing procedure, along with the data flow, centered on the information processing device 1 as the dialogue device according to this embodiment. First, as a preliminary process, account data for the actual staff member who is intended to be represented by the virtual staff member for each of the SNS, blog service, and review site is provided from the actual staff member's terminal to the information processing device 1, along with a virtual staff member dialogue processing request. The information processing device 1 then issues an identification number (ID) that identifies the virtual staff member (S11).

The profile acquisition unit 21 of the information processing device 1 sends a profile transmission request to the SNS server 3-1, blog site server 3-2, review site server 3-3, and EC site server 4, along with their respective account information. The SNS server 3-1, blog site server 3-2, and review site server 3-3 return profile data of the actual staff registered by the actual staff when using their respective services to the information processing device 1. As shown in the example in Figure 5, these profiles are compiled into a single profile. In addition to basic information such as name, gender, and age, the profile includes, for example, current occupation, characteristics of the company brand, and characteristics such as the actual staff member's personality and preferences. The compiled profile is associated with a virtual staff member ID and stored in the memory unit 14 (S12).

Next, the posted comment collection unit 22 sends a request to send posted comments along with each account of the real staff member to each of the SNS server 3-1, blog site server 3-2, review site server 3-3, and EC site server 4, and data on comments (posted text, etc.) posted by the real staff member is received from each of the SNS server 3-1, blog site server 3-2, review site server 3-3, and EC site server 4. These posted comments are associated with the ID of the virtual staff member representing the real staff member and stored in the storage unit 14 (S13).

As shown in FIG. 6, each comment is subjected to natural language analysis processing by the natural language analysis processing unit 23, broken down into parts of speech, and nouns are extracted (S14). Then, based on the extracted nouns, feature amounts that quantify the content features of the comment are calculated by the feature amount calculation processing unit 24 (S15). Feature amount calculation is repeated for all comments.

Coordinates in N-dimensional space are determined as feature quantities. As shown in Figure 8, multiple categories are associated with the N axes of the N-dimensional space. For example, in the apparel category, multiple categories such as season, inventory, size, design, item, color tone, price range, era, and event are associated with multiple axes. Each category is associated with multiple specific nouns (character strings) contained within it. Multiple specific nouns contained in the same category are ranked according to the similarity of their content meanings, and each is assigned a numerical value (coordinate value). Nouns extracted from comments are compared with specific nouns on each axis of the N-dimensional space, and if a corresponding character string is found, the axis number and coordinate value corresponding to that character string are given.

For example, for comment No. 2, "Tomorrow I plan to wear glen plaid pants and a blue jacket to work," the following nouns are extracted: "tomorrow," "glen plaid," "pants," "blue," "jacket," "work," and "plan." For example, on the axis associated with time periods, specific nouns related to many periods, such as yesterday, today, tomorrow, January, February, next month, the month after next, spring, and autumn, are associated and assigned coordinate values, respectively. The noun "tomorrow" extracted from the comment is given the coordinate value assigned to the specific noun "tomorrow" on the axis associated with the time period. Similarly, on the axis associated with item classes, specific nouns related to clothing, such as pants, skirts, blouses, shirts, dresses, suits, jackets, blazers, and coats, are associated and assigned coordinate values, respectively. The noun "pants" extracted from the comment is given the coordinate value of the corresponding specific noun on the item class axis. On the event class axis, nouns related to events, such as Father's Day, Mother's Day, sports day, birthdays, and dates, are associated and assigned coordinate values, respectively.

The calculated feature values are associated with the comment and stored in the memory unit 14 (S16).

New comments are collected at a predetermined interval, such as once a day, and feature values for the new comments are calculated and stored. This completes the preparatory process.

When a conversation begins, for example, a request to open a chat room (chatbot) for one-on-one conversation is sent from the user terminal 5 to the EC site server 4 that operates the EC site. Of course, the conversation format is not limited to chat rooms. The information processing device 1 selects a virtual staff member to converse with the user, and opens a chat room between the user and the virtual staff member (S17). For example, a virtual staff member who has previously conversed with the same real user is selected, or if a real staff member has served the real user in a real store, a virtual staff member who will represent that real staff member is selected. The method for selecting a virtual staff member is not limited to these methods.

For example, a message (statement) from a real user input from the user terminal 5 shown in FIG. 7 is received by the user message receiving unit 25 from the user terminal 5 via the EC site server 4. This message is subjected to natural language analysis by the natural language analysis processing unit 23 (S18), and its features are calculated by the feature calculation processing unit 24 (S19). The natural language analysis processing and feature calculation processing for the message are the same as those for the comment.

Next, the posted comment selection processing unit 26 selects one or a predetermined number of comments from the comments of the real staff member (S20). Specifically, one comment having feature quantities closest to those of the message from the real user, or a predetermined number of comments having feature quantities close to those of the message from the real user, are selected. More specifically, since feature quantities are coordinates in N-dimensional space, the distance between the feature quantities (coordinates) of the message from the real user and the feature quantities (coordinates) of the comment from the real staff member is calculated, and the comment with the shortest distance is selected, or a predetermined number of comments are selected in order of shortest distance. Since feature quantities quantify content-based features, the selected posted comment is similar in content to the real user's message. For example, as illustrated in Figure 9, if the message from the real user asks for recommendations on a jacket to wear on a trip, comments related to jackets and holidays are selected.

The message is sent from the message generation request unit 27 to the message generation device 2 along with a message generation request. The message generation request is accompanied by a message from the real user or a history of conversations between the real user and real staff. The message generation request is also accompanied by a profile of the real staff member represented by the virtual staff member. Furthermore, the message generation request is accompanied by reference information of comments similar in content to the real user's message, selected from comments actually posted by real staff members on social media, etc.

The message generation request unit 27 generates a message from a virtual staff member responding to a real user's message by referencing the real staff member's profile and comments similar in content to the real user's message, selected from comments actually posted by the real staff member on social media, etc. (S21). When generating the virtual staff member's message, comments similar in content to the real user's message are referenced, so the virtual staff member's message is generated with content similar to that of a real staff member's likely response. For example, when a real user asks for a jacket recommendation, a response recommending a "blue tweed jacket" is generated from comments previously posted by the real staff member. The language used by the virtual staff member is also based not only on the real staff member's profile, but also on the language used in comments previously posted by the real staff member.

As described above, when generating comments from virtual staff members in response to messages from real users, in addition to the profiles of the real staff members, comments that are similar in content to the real user's message and selected from comments that the real staff members have actually posted on social media, etc., can be referenced. This makes it possible to meet the following three requirements with a high degree of accuracy.

1) The comments made by the virtual staff are consistent in content with the comments (messages) made by real users.

2) The statements made by the virtual staff are the same as or similar to what a real staff member would say.

3) The language used by virtual staff mirrors that of real staff.

While several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments may be embodied in a variety of other forms, and various omissions, substitutions, and modifications may be made without departing from the spirit of the invention. These embodiments and their variations are within the scope of the invention and its equivalents as defined in the claims, as well as the scope and spirit of the invention.

1... Information processing device (interactive device), 2... Sentence generation device (generation AI), 3-1... SNS server, 3-2... Blog site server, 3-3... Review site server, 4... EC site server, 5... User terminal.

Claims (4)

An information processing device that is connected to a user terminal, at least one server selected from an SNS server, a blog site server, a word-of-mouth site server, and an EC site server, and a statement generation device (generation AI) via a communication line, and that allows a virtual user acting as a representative of a real user to interact with other real users,
a first receiving means for receiving data of comments from the other users directly from the user terminal or indirectly via at least one server selected from the SNS server, the blog site server, the word-of-mouth site server, and the EC site server;
a first transmission means for transmitting to the message generation device a request for generating a message from the virtual user in response to the message from the other user, the request including data of the message from the other user;
a second receiving means for receiving data of the message from the message generation device;
a second transmission means for transmitting the received data of the message of the virtual user to the user terminal or to at least one server selected from the SNS server, the blog site server, the word-of-mouth site server, and the EC site server;
means for extracting character strings of nouns by performing natural language analysis processing on each of the statements of the other users and a plurality of comments posted to at least one of the SNS server, the blog site server, the review site server, and the EC site server;
a feature calculation means for calculating a feature amount that quantifies content features of each of the statements and comments of the other users based on the extracted nouns;
and means for selecting, from the plurality of comments, at least one specific comment having content characteristics closest to the comment of the other user based on the feature amount;
The first sending means sends data of the selected specific comment as reference information to the generation request in addition to data of the other user's comments to the message generation device. The information processing device of claim 1, wherein the feature calculation means associates the extracted nouns with an N-dimensional space in which multiple types of categories represented by the nouns are respectively associated with N axes, and multiple nouns related to each category are assigned to each axis along with coordinate values, and assigns coordinates in the N-dimensional space as the feature. The information processing device described in claim 1, wherein the first transmission means transmits to the statement generation device, as reference information, a profile registered by the actual user on at least one of the SNS server, the blog site server, the review site server, and the EC site server, along with data on the other user's statements and data on the specific comment. an information processing device that is connected to a user terminal via a communication network, at least one server selected from an SNS server, a blog site server, a word-of-mouth site server, and an EC site server, and a statement generation device (generation AI) via a communication line, and that allows a virtual user acting as a proxy for a real user to interact with other real users;
a first receiving means for receiving data of comments from the other users directly from the user terminal or indirectly via at least one server selected from the SNS server, the blog site server, the word-of-mouth site server, and the EC site server;
a first transmission means for transmitting to the message generation device a request for generating a message from the virtual user in response to the message from the other user, the request including data of the message from the other user;
a second receiving means for receiving data of the message from the message generation device;
a program that functions as a second transmission means that transmits the received data of the message of the virtual user to the user terminal or to at least one server selected from the SNS server, the blog site server, the word-of-mouth site server, and the EC site server,
means for extracting character strings of nouns by performing natural language analysis processing on each of the statements of the other users and a plurality of comments posted to at least one of the SNS server, the blog site server, the review site server, and the EC site server;
a feature calculation means for calculating a feature amount that quantifies content features of each of the statements and comments of the other users based on the extracted nouns;
and a means for selecting, from the plurality of comments, at least one specific comment having content characteristics closest to the comment of the other user based on the feature amount;
The first sending means is a program that sends data of the selected specific comment as reference information to the generation request in addition to data of the other user's comments to the comment generation device.