Inquiry answering program and inquiry answering method

Abstract

To provide a program and method for answering inquiries from customers.SOLUTION: An inquiry answering system 1 includes: an inquiry receiving unit that receives data including contents of an inquiry from a user of a product; an inquiry text generating unit that generates an inquiry text based on the data; an answer candidate text generating unit that uses the inquiry text as input and generates an answer candidate text using artificial intelligence; a specific term usage count calculating unit that calculates the number of uses of a term included in a predetermined term group for each of the inquiry text and the answer candidate text; an additional term selecting unit that selects an additional term to be introduced into an answer to the user when a contents matching rate, which is a ratio between the number of uses and the number of uses of the term in the answer candidate text, is below a predetermined threshold; an answer text generating unit that generates an answer text using artificial intelligence based on the answer candidate text, the term included in the predetermined term group included in the answer candidate text, and the additional term; and an answer transmitting unit that transmits the answer text to the user.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a program for answering inquiries from customers and a method thereof.

Background Art

[0002] Industrial products, particularly electronic components and mechanical parts, are specified by various factors such as material, dimensions, performance, and usage, and are managed by complex part numbers and model numbers. For manufacturers of such products, there is a customer support service of answering inquiries about the products from customers and users of the products using specialized knowledge as the manufacturer and seller of the products.

[0003] Inquiries from customers may be made by phone, but due to the recent spread of the Internet, complex part numbers and model numbers can be accurately conveyed, and inquiries can be made even outside the business hours of the manufacturer such as at night and on holidays. Therefore, inquiries by email or text message transmission equivalent thereto (for example, a text transmission form on the manufacturer's website) are increasing. Furthermore, the content of such inquiries has also become more complex, and the human resources for appropriately and promptly answering the inquiries are increasing.

[0004] Furthermore, the answers by the customer support department in the manufacturer to these inquiries may vary in quality because the knowledge and skills of the persons in charge of answering are different. On the other hand, if multiple persons in charge check the answers to ensure the quality of the answers, there will be a problem that the answers cannot be given promptly.

[0005] The manufacturer's website may have FAQs (Frequently Asked Questions) that customers can access through the Internet. However, FAQs are assumed questions and answers, and the number of questions for which answers can be prepared in advance is limited. Also, FAQs are limited to relatively abstract questions, and answers to individual and specific questions that customers face are often not directly described. On the other hand, if the number of FAQs is increased to be able to answer a wide range of customer questions, it is necessary to search for appropriate questions among a large number of FAQs, which requires a cumbersome procedure for customers and results in the problem that answers cannot be obtained quickly.

[0006] In contrast, a chatbot program having an automatic dialogue program that can communicate with a user and generate a long response message has been proposed (Patent Document 1).

Prior Art Documents

Patent Documents

[0007]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0008] However, in the chatbot system disclosed in Patent Document 1, the response message to the question text received from the user is limited to what can be generated by a knowledge database that stores the question text and its answer in advance. Therefore, for inquiries regarding a wide variety of industrial products and their usage methods, it may not be sufficient to generate appropriate answers.

[0009] An object of the present invention is to provide a program that can quickly and automatically answer inquiries from a user who is a customer with accurate and high-quality answers.

Means for Solving the Problems

[0010] To solve the above problems, the response program of the present invention, described with reference numerals of embodiments, a computer that receives inquiries from users and sends responses inquiry receiving means for receiving data including the content of an inquiry from the user of the product to the manufacturer of the product; inquiry text generation means for generating inquiry text 211 based on the data including the content of the inquiry; response candidate text generation means for generating response candidate text 221 by artificial intelligence (AI) 400 using the inquiry text as an input; specific term usage number calculation means for calculating the number of uses 231 of terms included in a specific public standard and / or a predetermined group of terms 320 based on materials related to the product for each of the inquiry text 211 and the response candidate text 221; content matching rate determination means for comparing the content matching rate, which is the ratio of the number of uses of the terms in the inquiry text 211 to the number of uses of the terms in the response candidate text 221, with a predetermined threshold 241; additional term selection means for selecting additional terms 251 to be introduced into the response to the user when the content matching rate is less than the predetermined threshold 241; response text generation means for generating response text 261 using the artificial intelligence 400 based on the response candidate text 221, the terms included in the predetermined group of terms 320 included in the response candidate text 221, and, when the content matching rate is less than the predetermined threshold 241, further the additional terms 251; inquiry response transmission means for transmitting the response text 261 to the user; is made to function as.

[0011] According to this configuration, for an inquiry from a user who is a customer of a product, an accurate and high-quality response by artificial intelligence can be quickly and automatically provided, so that cost reduction by automating customer support operations and improvement of customer satisfaction by a highly accurate response by a computer can be achieved.

[0012] In the program of the present invention, the artificial intelligence 400 may be constituted by a machine learning model 410 trained using the specific public standard and / or the materials related to the product as teacher data.

[0013] According to this configuration, in response to an inquiry from a user who is a customer of the product, an accurate and high-quality response sentence can be generated by the artificial intelligence 400 having specialized knowledge about the product. Therefore, cost reduction by automating the customer support service and improvement of customer satisfaction by accurate computer-generated responses can be achieved.

[0014] In the program of the present invention, the machine learning model may be a large language model (LLM) that outputs natural language sentences.

[0015] According to this configuration, in response to an inquiry from a user who is a customer of the product, a response sentence using fluent natural language can be generated using the machine learning model 410 which is a large language model. Therefore, improvement of customer satisfaction can be achieved by accurate computer-generated responses.

[0016] In the program of the present invention, the specific public standard may be the Japanese Industrial Standard (JIS), and / or the materials related to the product may be the product catalog related to the product issued by the manufacturer.

[0017] According to this configuration, in response to an inquiry from a user who is a customer of the product, the machine learning model 410 constituting the artificial intelligence 400 can accurately and easily learn specific and sufficient specialized knowledge about the product.

[0018] In the program of the present invention, the machine learning model may be trained to process the synonyms and similar terms of each term included in the predetermined group of terms as the corresponding term.

[0019] According to this configuration, even when the words included in the inquiry text do not exactly match the terms included in the official standards and the materials regarding the manufacturer's products, the response text can be generated based on the corresponding synonyms or similar words. Therefore, even when there are variations in the wording of the inquiry from the user, an accurate and high-quality response text can be generated for the inquiry.

Brief Description of the Drawings

[0020]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Modes for Carrying Out the Invention

[0021] [First Embodiment] Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a diagram showing the configuration of an inquiry response system 1 including a computer in which the program according to the present embodiment is executed.

[0022] <Inquiry Response System> The inquiry response system 1 is composed of an information processing device 10 and a user terminal 20 that is communicably connected to the information processing device 10 via a network 30. The information processing device 10 and the user terminal 20 are realized by a computer, and the program according to the present embodiment is executed by the computer that realizes the information processing device 10.

[0023] <Information Processing Apparatus> The information processing apparatus 10 includes a communication unit 10 that communicates with the outside to transmit and receive data, a processing unit 200 that performs processing based on the received data, a storage unit 300 that stores data necessary for the processing, and an artificial intelligence 400 that performs processing in cooperation with the processing unit 200 based on a command from the processing unit 200. The internal configuration of the information processing apparatus 10 is not limited to the configuration shown in FIG. 1, and may be any other configuration as long as it is a configuration for performing the information processing described later.

[0024] <Communication Unit> The communication unit 100 is composed of an inquiry receiving unit 110 and a response transmitting unit 120. The inquiry receiving unit 110 receives an inquiry from a user who is a customer from the user terminal 20 via the network 30, and the response transmitting unit 120 transmits a response to the inquiry to the user terminal 20 via the network 30.

[0025] The inquiry receiving unit 110 functions as an input interface for receiving a response from the outside in the program of the present invention. The inquiry from the user received by the inquiry receiving unit 110 may be an inquiry by e-mail, or an inquiry by input from a text input form on a website or a chatbot. Also, the inquiry from the user may be transmitted by means capable of text transmission such as a messaging application or an SNS (Social Network Service). These applications or services may operate on either a fixed terminal or a mobile terminal, and may operate on either a general-purpose terminal or a dedicated terminal. Furthermore, the inquiry from the user may be an inquiry by voice, such as a phone call using a public switched telephone network or an IP phone using a wide area network such as the Internet.

[0026] In the program of the present invention, the response transmission unit 120 functions as an output interface for transmitting a response externally. The response transmission unit 120 transmits, via the network 30, a response to a query from the user, which is generated by the query response system 1, to the user terminal 20. The response to be transmitted is based on the response text 261 generated by the response text generation unit 260 of the processing unit 200 described later.

[0027] When the query from the user is by e-mail, the response transmission unit 120 transmits an e-mail with the text data serving as the response described in the body or attached as an attachment file to the user's e-mail address. Thereby, the response to the query is transmitted to the user terminal via the e-mail server.

[0028] When the query from the user is an input from the text transmission form of the website, a command including the text data serving as the response is transmitted to the web server having the function of displaying the website. By the web server receiving and processing this command, the response to the query is displayed on the page of the website that the user terminal 20 can view. When the query from the user is an input from the chatbot on the website, the response is displayed as a statement by the chatbot.

[0029] When the query from the user is transmitted from a general-purpose or dedicated application such as a messaging application, the response is displayed in the form of a reply in the application on the user terminal where the application is operating.

[0030] <Processing Unit> The processing unit 200 is composed of a query text generation unit 210, a response candidate text generation unit 220, a specific term usage count calculation unit 230, a content match rate determination unit 240, an additional term selection unit 250, and a response text generation unit 260, which are obtained from the data received by the query reception unit. Each configuration will be described below.

[0031] <Inquiry Text Generation Unit> The inquiry text generation unit 210 receives inquiry data including an inquiry from a user who is a customer from the inquiry reception unit 110, and generates an inquiry text 211 representing the content of the inquiry in a method according to the inquiry method and the inquiry data.

[0032] When the user makes an inquiry from a text input form including a chatbot or the like on a website, the inquiry text 211 is generated based on data transmitted from a web server that displays the website. When the user makes an inquiry using an email, the inquiry text 211 is generated by extracting text representing the inquiry content from the body or attached file of the email received at a predetermined email address for receiving inquiries.

[0033] When the user makes an inquiry using a phone, the inquiry text 211 is generated by converting the voice on the phone into text using voice recognition technology. By obtaining the inquiry content as text-formatted data, the same processing as that for text inquiries can be applied to voice inquiries in subsequent steps.

[0034] <Answer Candidate Text Generation Unit> The answer candidate text generation unit 220 generates an answer candidate text 221 for the inquiry text 211. The answer candidate text 221 is obtained as text data by using the artificial intelligence 400 described later.

[0035] <Specific Term Usage Count Calculation Unit> The specific term usage count calculation unit 230 has a function of inputting a text and a word group consisting of one or more words and outputting the number of words in the word group included in the text. Specifically, it performs morphological analysis on the input text to extract nouns in the text, and determines whether each noun is a word included in the word group, thereby calculating the number of words in the word group used in the text. In this embodiment, as the input text, the inquiry text 211 and the answer candidate text 221 are used, and for each text, the number of words included in the specific term group data 320 described later that are used in the text is calculated as the specific term usage count 231.

[0036] <Content matching rate determination unit> The content matching rate determination unit 240 calculates the content matching rate based on the specific term usage counts 231 in two texts, and determines the degree of content matching between the two texts by comparing with a predetermined threshold 241. In this embodiment, the specific term usage count M for the inquiry text 211 A and the specific term usage count M for the answer candidate text 221 B are used to calculate the content matching rate α as the ratio, and it is determined whether it is equal to or greater than the predetermined threshold 241.

[0037] <Additional term selection unit> The additional term selection unit 250 selects additional terms 251 using the artificial intelligence 400 based on the inquiry text 211 and the answer candidate text 221. Specifically, the additional term selection unit 250 commands the artificial intelligence 400 to select, as the additional terms 251, the terms that are lacking in the answer represented by the answer candidate text 221 for the inquiry represented by the inquiry text 211 and that are included in the specific term group data 320 described later.

[0038] <Answer text generation unit> The response text generation unit 260 generates a response text 261 by the artificial intelligence 400 based on the response candidate text 221, the terms of the specific term group data 320 included in the response candidate text 221, and the selected additional terms 251. Specifically, the response text generation unit 260 instructs the artificial intelligence 400 so that the response text 261 generated by the artificial intelligence 400 is a text that improves the response candidate text 221 and includes the terms of the specific term group data 320 and the additional terms 251 included in the response candidate text 221. The generated response text 261 is a text that serves as a response to an inquiry from a customer in a series of processes according to the program of the present invention.

[0039] <Memory unit> The memory unit 300 includes product expertise data 310 and specific term group data 320. The product expertise data 310 is data representing information included in specific public standards and / or materials related to the product, and is used as learning data for the artificial intelligence 400 described later. The specific term group data 320 is data representing a term group composed of a plurality of terms in the technical field of the product included in the product expertise data 310.

[0040] <Product expertise data> The product expertise data 310 is data representing information included in specific public standards and materials related to the product. In the present embodiment, the product expertise data 310 is prepared based on Japanese Industrial Standards (JIS) and the manufacturer's catalog, and hereinafter, for bearings and parts related to bearings, which are products represented by F16C of Japanese Industrial Standards (JIS), it contains detailed and extensive information about the meaning of terms in the technical field of the product, data on dimensions, shapes, performance, uses, etc. of various products of the manufacturer, and the relevance between products.

[0041] By having the artificial intelligence 400 learn the information included in the product expertise data 310, the artificial intelligence 400 can generate a response even to an inquiry that requires specialized knowledge about the product.

[0042] <Specific term group data> The specific term group data 320 is data consisting of terms included in the product expertise data 310. That is, the term group in the specific term group data 320 includes terms extracted from public standards and materials related to the product, particularly catalogs related to the product issued by the product manufacturer. In this embodiment, the term group included in the specific term group data 320 is data extracted from the product, the bearing, and standard documents represented by JIS B 15XX, etc. related thereto, and the product catalog of the bearing which is the product of the manufacturer.

[0043] <Artificial intelligence> Artificial intelligence (AI; Artificial Intelligence) is a program that receives a command and outputs a response to the command. The artificial intelligence 400 can receive commands from the response candidate text generation unit 220, the additional term selection unit 250, and the response text generation unit 260 in the processing unit 200 and output a response that satisfies the requirements included in the command.

[0044] The artificial intelligence 400 is realized by the machine learning model 410. The machine learning model 410 acts as a large language model (LLM; Large Language Models) that realizes natural language generation and understanding like a human by learning a large amount of text data, and performs deep learning (deep learning) using a neural network that constitutes the model with the product expertise data 310 as teacher data.

[0045] Thereby, the artificial intelligence 400 can respond to an inquiry sentence in text described in natural language with text described in natural language, and can generate a response text even for an inquiry that requires specialized knowledge about the product by learning the information included in the product expertise data 310.

[0046] 〔Regarding answer generation by artificial intelligence (AI)〕 Dialogue programs using computers range from those called artificial idiots that simply output fixed phrases according to the words contained in the question sentence, to large language models constructed by deep learning using neural networks based on a large amount of learning data, and can generate new content not limited to fixed phrases, including Generative Artificial Intelligence. Pre-trained models that have undergone pre-training using an enormous amount of text data have been made publicly available for large language models and can be incorporated into dialogue programs for use.

[0047] While dialogue programs using large language models can generate response sentences in fluent natural language with appropriate content for question sentences in a wide range of fields, there is a problem that appropriate response sentences are not generated for question sentences that require knowledge or non-public information in a specific specialized field and are not included in the data used for training the model. For example, inappropriate responses may be generated for inquiries from users regarding specific products.

[0048] To address this problem, in addition to presenting specialized knowledge or references to specialized knowledge in the instruction message, which is the input to the artificial intelligence called a prompt, to reflect the specialized knowledge in the response sentence, a model specialized in a specific specialized field can be generated by fine-tuning the large language model using additional learning data for training. The program of this embodiment aims to generate a response sentence to an inquiry to the manufacturer regarding the bearing, which is the manufacturer's product. The large language model incorporated into the program of this embodiment is fine-tuned by product specialized knowledge data 310 containing information such as the meaning of terms in the technical field of bearings, dimensions, shapes, performance, uses, etc. of the manufacturer's specific products, and the relevance between products, which is prepared based on the manufacturer's catalog and Japanese Industrial Standards (JIS).

[0049] FIG. 5 is a cross-sectional view showing an outline of a bearing. A bearing is a mechanical element used to support a rotating shaft in various machines and devices. The bearing 900 shown in FIG. 5 includes an inner ring 901, an outer ring 902 coaxially provided radially outward of the inner ring 901, a plurality of balls 903, 903 rotatably held between the inner ring 901 and the outer ring 902, and a cage 904 that holds the plurality of balls 903, 903 at a predetermined interval, and can rotatably support a shaft 905 fitted into the inner ring 901. As the main dimensions characterizing the bearing 901, there are a bearing inner diameter d, a bearing outer diameter D, a bearing width B, and the like. In addition to the main dimensions, there are various types of bearings due to differences in components, overall shape, structure, material, etc. Information on these various bearings can be obtained from materials such as public standard documents and catalogs related to products issued by manufacturers. For example, the bearing 900 has a catalog number determined by the main dimensions, tolerance grade, type of bearing, etc. In public standards, in addition to the definition of terms related to bearings, these catalog numbers are standardized. Also, the values indicating the performance of the bearing, such as the basic dynamic load rating, basic static load rating, fatigue limit load, allowable rotational speed, etc., defined by public standards are determined by the manufacturer of the bearing for each bearing having a respective catalog number. Various information regarding these bearings is described in materials such as public standard documents and catalogs related to products issued by manufacturers. Furthermore, information on the bearing selection method, handling method, product model numbers of the manufacturer, etc. can be obtained from the manufacturer's materials.

[0050] Since the manufacturer's catalog contains detailed information about the manufacturer's products, in addition to being teacher data and effective learning data for learning, the manufacturer generally stores the data described in the catalog in a data format that is easy to handle, so it can be easily converted into learning data for fine-tuning. Also, in order to use the various parameters of the products represented in tabular form in the catalog as learning data, a framework has been proposed that can use tabular data for the deep learning of language models. In this way, the manufacturer can utilize the product catalog prepared by the manufacturer itself to enable the artificial intelligence to learn knowledge about the domain (specialty field) of the products manufactured by the manufacturer, thereby obtaining a model specialized for a specific domain from a general-purpose large-scale language model. With such a model, for the question text asking about the manufacturer's products, the accurate knowledge possessed by the model regarding the products is reflected, and an answer text described in fluent natural language without grammatical defects can be output.

[0051] <Operation of the program> Hereinafter, a series of processes when the inquiry response system 1 receives an inquiry from a user and gives an answer will be described. FIG. 2 is a flowchart showing the operation of a program that generates an answer to be sent to the user in response to the received inquiry from the user in the inquiry response system 1.

[0052] In step S700, the inquiry reception unit 110 receives inquiry data from the user terminal 20 via the network 30. Next, in step S710, the inquiry text generation unit 210 generates an inquiry text 211 by means of the inquiry text generation means.

[0053] Next, in step S712, the response candidate text generation unit 220 generates a response candidate text 221 based on the inquiry text 211 by the response candidate text generation means. The response candidate text generation means is realized by using the artificial intelligence 400. The machine learning model 410 that constitutes the artificial intelligence 400 is trained using an appropriate and sufficient amount of training data, and can generate a smooth response text in natural language without being limited to only the preset fixed patterns even for a question sentence in natural language that may include ambiguity. Further, the machine learning model 410 is trained using the product expertise data 310 so as to be able to answer a question sentence regarding the expertise of the product. Therefore, according to the artificial intelligence 400, it is possible to generate a response candidate text 221 that is a smooth text in natural language and is based on the expertise of the product.

[0054] Next, in step S714, the specific term usage count 231 and the content matching rate based on the specific term usage count 231 are calculated. The specific term usage count calculation unit 230 calculates the specific term usage count 231, which is the number of terms included in the specific term group data 320 in the input text, with each of the inquiry text 211 and the response candidate text 221 as the input text by the specific term usage count calculation means. Specifically, first, the input text is morphologically analyzed using a morphological analyzer to extract nouns in the text and generate a word group included in the input text. Next, for each word in the generated word group, it is determined whether it is a term included in the specific term group data 320, and the number of included words is accumulated. As a result, the specific term usage count M A for the inquiry text 211 and the specific term usage count M B for the response candidate text 221 are calculated.

[0055] Next, in step S716, the content matching rate determination unit 240 compares the content matching rate α, which indicates the degree of relevance between the inquiry text 211 and the response candidate text 221, with a predetermined threshold value 241 by the content matching rate determination means, and determines whether the content matching rate α is equal to or greater than the threshold value. Here, the content matching rate α is the specific term usage count M of the inquiry text 211A and the number M of specific terms used in the response candidate text 221 B Using these, the following formula α = M A / M B is obtained.

[0056] When the content matching rate α is less than the threshold value (No in step S716 of FIG. 2), the additional term selection unit 250 selects an additional term 251 by the additional term selection means (step S720 of FIG. 2). The additional term 251 is a term determined by the artificial intelligence 400 to be lacking in the response candidate text 221 for the inquiry from the user, and is selected from the terms included in the specific term group data 320.

[0057] Next, the response text generation unit 260 generates a response text 261 corresponding to the final response to the inquiry from the user by the response text generation means. The response text 261 is generated by the artificial intelligence 400 based on the response candidate text 221 and so that the terms of the specific term group included in the response candidate text 221 and the additional term 251 selected by the additional term selection unit 250 are included in the response text 261 (steps S718 and S722 of FIG. 2). When the additional term 251 is not selected, that is, when the content matching rate α is greater than or equal to the threshold value (Yes in step S716 of FIG. 2), the response text 261 may be the same as the response candidate text 221. As described above, according to the artificial intelligence 400, a response text 261 that is a fluent sentence in natural language and is based on specialized knowledge about the product can be obtained.

[0058] Next, the response transmission means transmits the response text 261, which is the final response generated by the processing unit 200, to the user terminal 20 by the response transmission unit 120 of the communication unit 100 (step S730 of FIG. 2).

[0059] Through the series of processes by the inquiry response program described above, the inquiry response system 1 receives an inquiry from a customer from the user terminal 20, generates an answer to the inquiry, and transmits the answer to the user terminal 20.

[0060] In the present embodiment, the machine learning model 410 constituting the artificial intelligence 400 may be trained to treat synonyms and similar words of each term included in the specific term group data 320 as the term itself.

[0061] Examples of synonyms and similar words for terms in the case where the manufacturer's product is a bearing and parts related to the bearing are shown in Table 1 below. Also, when a term is expressed using variant Chinese characters in the Chinese character notation, it may be treated as a synonym of the corresponding term. The terms in Table 1 are generally used for bearings including the bearing 900 in FIG. 5.

[0062]

Table 1

[0063] <Function and effect> According to the inquiry response program of the present invention, for an inquiry from a user who is a customer of a product, even for an inquiry that requires specialized knowledge about the product, an artificial intelligence having knowledge of the product catalog can generate an answer sentence using accurate, high-quality, and fluent natural language. Therefore, cost reduction by automating customer support operations and improvement of customer satisfaction by highly accurate computer-generated answers can be achieved.

[0064] <Regarding other embodiments> In the following description, parts corresponding to matters described in advance in each embodiment are given the same reference numerals, and redundant descriptions are omitted. When only a part of the configuration is described, the other parts of the configuration are the same as those in the embodiments described in advance unless otherwise specified. The same configuration exhibits the same effects. It is possible not only to combine the parts specifically described in each embodiment, but also to partially combine the embodiments with each other as long as there is no problem with the combination.

[0065] FIG. 3 shows a flowchart showing the flow of processing of a program according to another embodiment of the present invention. In this embodiment, the processing unit 200 of the information processing apparatus 10 further includes a feedback request addition unit, and in the feedback request addition unit, the program causes the computer to function as feedback request addition means. The feedback request addition means adds a feedback request instruction text, which is text for obtaining feedback from the user on the answer, to the answer text 261 generated by the answer text generation unit 260 (step S724 in FIG. 3). The feedback is typically the user's satisfaction with the answer, and the added text may prompt a selection from a plurality of options such as "very satisfied", "satisfied", "insufficient", "dissatisfied", etc. When the answer transmission unit 120 transmits the answer text 261 to a web browser or a messaging application, the added text may include not only text including a plurality of options for the user's satisfaction, but also a script for instructing to display a UI (user interface) for promoting input such as radio buttons and check boxes in the web browser or the messaging application.

[0066] In this embodiment, the processing unit 200 of the information processing apparatus 10 further includes a feedback determination unit and a content matching rate threshold adjustment unit, and the program causes the computer to function as a feedback determination means and a content matching rate threshold adjustment means in the feedback determination unit and the content matching rate threshold adjustment unit, respectively. The feedback determination unit determines the presence or absence of feedback from the user and the user's satisfaction with the answer in the feedback. When feedback from the user is obtained (Yes in step S732 of FIG. 3), the content matching rate threshold adjustment unit adjusts the threshold 241 for the content matching rate α according to the user's satisfaction determined by the feedback determination unit (step S734 of FIG. 3). When feedback with low user satisfaction for the answer is obtained, the threshold 241 is increased. When feedback with high user satisfaction for the answer is obtained, the threshold 241 is decreased. As a result, in subsequent inquiries, the answer candidate text 221 is more often considered insufficient, and the answer text 261 including the additional term 251 is generated, so that the quality of the answer can be improved by the feedback from the user.

[0067] Also, in this embodiment, the program may further have means for analyzing and self-evaluating the answer text 261 by the artificial intelligence 400 before transmitting the answer text 261 to the user.

[0068] In still another further embodiment of the present invention, the answer text generated by the program includes terms included in the answer text generated in the past, that is, the product expertise data 310 includes the inquiry text from the user in the past history and the answer text thereto, and using this as learning data, the machine learning model 410 constituting the artificial intelligence 400 may be further trained. Since the past inquiry history is non-public information and is expected to include a set of question texts with high similarity and high-quality answer texts for expected inquiries, by having the artificial intelligence 400 learn such learning data, the artificial intelligence 400 can generate more appropriate and high-quality answer texts.

[0069] In still another embodiment of the present invention, the processing unit 200 further includes an additional information request sentence output unit, and the program may further cause the computer to function as additional information request sentence output means. The additional information request sentence output unit may output an additional information request sentence for requesting additional information regarding the usage conditions of the bearing and / or the characteristics of the bearing based on the content of the inquiry from the customer of the product by the additional information request sentence output means. The additional information request sentence is transmitted by the answer transmission unit 120 to the user terminal 20 to request the user, who is the customer, for additional information regarding the inquiry.

[0070] In this embodiment, the additional information request sentence may request information regarding the operating time of the bearing, preferably information on whether the operating time of the bearing is less than 1000 hours, less than 5000 hours, less than 10,000 hours, less than 50,000 hours, less than 100,000 hours, or 100,000 hours or more.

[0071] Alternatively, in this embodiment, the additional information request sentence may request information regarding the main rotational speed of the bearing, preferably information on whether the product of the pitch diameter Dpw of the rolling elements of the bearing and the main rotational speed n of the bearing is less than 10,000 [mm / min] or 10,000 [mm / min] or more.

[0072] According to the additional information request sentence output means, it is possible to prompt the user to acquire the information necessary for the answer, and when an answer from the user is obtained for the additional information request sentence, it becomes possible to generate a more appropriate and accurate answer to the user's inquiry.

[0073] In still another embodiment of the present invention, the program further causes the computer to function as inquiry content storage means, and the inquiry content determination means further determines, by analogy with the artificial intelligence 400, whether the content of the inquiry is a complaint and the type of the complaint. When it is determined that it is a complaint, the inquiry content storage means may store the type of the complaint in storage means provided in the computer.

[0074] In this embodiment, the inquiry content determination means may further determine the type of the complaint based on whether the content of the inquiry includes terms related to overheating and whether it includes terms related to noise.

[0075] According to the inquiry content determination means and the inquiry content storage means, it is possible to quickly and appropriately respond to complaints from users that may lead to the loss of customers of the product, and to automatically record a history of the complaints.

[0076] In still another embodiment of the present invention, the inquiry from the user may be related to the International Patent Classification F16C including bearings.

[0077] In the embodiments described above, Japanese is used for the inquiry from the user, but any other natural language or artificial language may be used as long as it is a text-processable language.

[0078] <Hardware Configuration> The information processing apparatus 10 and the user terminal 20 shown in FIG. 1 are realized by, for example, a computer 800 whose internal structure is shown in FIG. 4. The components of the computer 800 are connected by a bus 805 used for data transmission between the components. The input / output device interface 810 enables various input / output devices (e.g., keyboard, mouse, display, printer, speaker, etc.) to be connected to the computer 800. The network interface 830 enables the computer 800 to communicate with various other devices connected to a network (e.g., the network 30 in FIG. 1). The memory 840, which is a volatile storage unit, and the disk storage 860, which is a non-volatile storage unit, store the programs 842, 862 and the data 844, 864 in one embodiment of the present invention. The central processing unit 820 executes computer instructions that make up the program. Regarding the specific hardware configuration of the computer, omissions, substitutions, and additions of components can be appropriately made according to the embodiment. For example, the computer may include a plurality of central processing units.

[0079] As described above, the present invention has been described based on the embodiments. However, the embodiments disclosed this time are illustrative in all respects and not restrictive. The scope of the present invention is indicated by the claims rather than the above description, and it is intended that all modifications within the meaning and scope equivalent to the claims be included.

[0080] The program of the present invention can be recorded on a computer-readable medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, a semiconductor memory, etc. More specifically, the computer-readable medium includes, as a magnetic recording device, a hard disk device, a flexible disk, a magnetic tape, etc.; as an optical disk, a DVD (Digital Versatile Disc), a DVD-RAM (Random Access Memory), a CD-ROM (Compact Disc Read Only Memory), a CD-R (Recordable) / RW (ReWritable), etc.; as a magneto-optical recording medium, an MO (Magneto-Optical disc), etc.; and as a semiconductor memory, an EEP-ROM (Electrically Erasable and Programmable-Read Only Memory), etc.

[0081] In the above-described embodiment, the program executed by a computer has been mainly described, but the form of the present invention is not limited thereto. The scope of the present invention includes a method of executing the program of the present invention by a computer, a computer that executes the program of the present invention and an information processing apparatus including the computer, and an information processing system configured by the information processing apparatus. That is, the following inquiry response system, which is a modification of the present invention, is included.

[0082] <Modification Example> An inquiry response system 1 that receives an inquiry from a user and transmits a response, an inquiry receiving unit 110 that receives data including the content of an inquiry from the user of the product to the manufacturer of the product; an inquiry text generation unit 210 that generates an inquiry text based on the data including the content of the inquiry; a response candidate text generation unit 220 that takes the inquiry text 211 as an input and generates a response candidate text 221 by an artificial intelligence (AI) 400; For each of the inquiry text 211 and the answer candidate text 221, a specific term usage count calculation unit 230 that calculates the number of terms used in a specific public standard and / or a predetermined group of terms based on materials related to the product; A content matching rate determination unit 240 that compares the content matching rate, which is the ratio of the number of terms used in the inquiry text 211 to the number of terms used in the answer candidate text 221, with a predetermined threshold value 241; An additional term selection unit 250 that selects additional terms 251 to be introduced into the answer to the user when the content matching rate is less than the predetermined threshold value 241; An answer text generation unit 260 that generates an answer text 261 using the artificial intelligence 400 based on the answer candidate text 221, the terms included in the predetermined group of terms included in the answer candidate text 221, and, when the content matching rate is less than the predetermined threshold value, further the additional terms; An answer transmission unit 120 that transmits the answer text 261 to the user; An inquiry answering system comprising the above.

[0083] Also, in the above-described embodiment, the processing content is realized as software by executing the program of the present invention on a computer. However, in the system using the program of the present invention, at least a part of these processing contents may be realized hardware-wise.

Explanation of Reference Numerals

[0084] 1... Inquiry answering system, 10... Information processing device, 100... Communication unit, 200... Processing unit, 300... Storage unit, 400... Artificial intelligence, 20... User terminal, 30... Network

Claims

1. A computer that receives inquiries from users and sends responses, Inquiry receiving means for receiving data including the content of an inquiry from the user of the product to the manufacturer of the product, Inquiry text generation means for generating an inquiry text based on the data including the content of the inquiry, Response candidate text generation means for generating a response candidate text by artificial intelligence (AI) using the inquiry text as an input, Specific term usage number calculation means for calculating the number of uses of terms included in a specific public standard and / or a predetermined group of terms based on materials related to the product for each of the inquiry text and the response candidate text, Content matching rate determination means for comparing the content matching rate, which is the ratio of the number of uses of the terms in the inquiry text to the number of uses of the terms in the response candidate text, with a predetermined threshold value, Additional term selection means for selecting additional terms to be introduced into the response to the user when the content matching rate is less than the predetermined threshold value, Response text generation means for generating a response text using the artificial intelligence based on the response candidate text, the terms included in the predetermined group of terms included in the response candidate text, and, when the content matching rate is less than the predetermined threshold value, further the additional terms, Response sending means for sending the response text to the user, A program for causing it to function.

2. The program according to claim 1, wherein the artificial intelligence is constituted by a machine learning model trained using the specific public standard and / or materials related to the product as teacher data.

3. The program according to claim 2, wherein the machine learning model is a large language model (LLM) that outputs natural language sentences.

4. In the program according to claims 1 to 3, the specific public standard, the materials related to the product are the product catalog related to the product issued by the manufacturer, the program.

5. In the program according to claims 1 to 3, the machine learning model is trained to process the synonyms and similar terms of each term included in the predetermined group of terms as the corresponding term, the program.

6. Executed by a computer An inquiry text receiving step of receiving an inquiry text which is data including the content of an inquiry from the user of the product to the manufacturer of the product; An answer candidate text generation step of generating an answer candidate text by artificial intelligence (AI) using the inquiry text as an input; A term usage number calculation step of calculating the number of terms used in the inquiry text and the answer candidate text respectively, which are included in a specific public standard and / or a predetermined group of terms based on materials related to the product; A content matching rate determination step of comparing the content matching rate, which is the ratio of the number of terms used in the inquiry text to the number of terms used in the answer candidate text, with a predetermined threshold value; An additional term selection step of selecting additional terms to be introduced into the answer to the user when the content matching rate is less than the predetermined threshold value; An answer text generation step of generating an answer text using the artificial intelligence based on the answer candidate text, the terms included in the answer candidate text, and when the content matching rate is less than the predetermined threshold value, further the additional terms; An answer sending step of sending the answer text to the user; Having, an inquiry answering method.

Images