Information Processing Apparatus, Information Processing Method, and Program

Abstract

To allow a user to easily find a desired target text, or the like, through a chat and obtain an appropriate answer based on the target text, or the like.SOLUTION: A question sentence candidate extraction unit 103 extracts one or more question sentence candidates including, in at least a part thereof, a keyword accepted by a keyword or the like acceptance unit 102. A question sentence candidate presentation unit 104 presents the extracted one or more question sentence candidates to a user. A related text or the like presentation unit 106 extracts, from a question sentence dictionary DB 181, a target text or the like associated with one selected from among the one or more question sentence candidates, and presents the extracted target text or the like to the user. A chat type UI control unit 109 controls transfer of information between the keyword or the like acceptance unit 102 to the related text or the like presentation unit 106 and a chatbot, thereby causing the chatbot to generate an answer based on the related text or the like to be presented, to a text input by the user.SELECTED DRAWING: Figure 8

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] As an additional service for assisting users who receive products or services, there is a technology related to a so-called FAQ search service that can search for information regarding a combination of typical questions and answers (for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in conventional FAQ search services including the technology of Patent Document 1, even if a general user executes a search with a keyword that comes to mind, it often does not hit the desired search results and does not reach the desired target text or the like. Also, it is often the case that too many results are hit and the desired target text or the like cannot be reached. Specifically, for example, in the case of a FAQ search service for insurance products, when a search is executed with the keyword "update", a huge amount of search results are displayed, and it is difficult to reach the desired target text or the like. On the other hand, even if a conventional chatbot is simply utilized, it is difficult for a user to reach the desired target text or the like.

[0005] The present invention has been made in view of such a situation, and an object thereof is for a user to easily reach a desired target text or the like through a chat and obtain an appropriate answer based on the target text or the like.

Means for Solving the Problems

[0006] To achieve the above object, an information processing apparatus according to one aspect of the present invention includes: reception means for receiving words related to a target sentence or the like, which is a predetermined word, phrase, or sentence presented to the user to achieve a predetermined purpose of the user, as the target sentence or the like; extraction means for extracting one or more candidate guiding sentences including at least a part of the word or a similar word therefrom, from a guiding sentence dictionary in which a plurality of candidate guiding sentences intended by the user to reach the target sentence or the like are registered in advance in association with the target sentence or the like; first presentation means for presenting the one or more candidate guiding sentences extracted by the extraction means to the user; second presentation means for extracting the target sentence or the like associated with the selected candidate guiding sentence from the guiding sentence dictionary among the one or more candidate guiding sentences presented to the user by the first presentation means and presenting it to the user; chat control means for controlling the exchange of information between the reception means to the second presentation means and the chatbot while the chatbot executes control for chatting with the user, so that the chatbot generates an answer based on the target sentence or the like to be presented by the second presentation means for the sentence input by the user; and includes.

[0007] An information processing method and a program according to one aspect of the present invention are a method and a program corresponding to the information processing apparatus according to one aspect of the present invention, respectively.

Advantages of the Invention

[0008] According to the present invention, the user can easily reach a desired target sentence or the like through chatting and obtain an appropriate answer based on the target sentence or the like.

Brief Description of the Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Embodiment for Carrying Out the Invention

[0010] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0011] First, with reference to FIGS. 1 to 5, an overview of a service (hereinafter referred to as "this service") that can be realized by an information processing system (see FIG. 6 described later) to which a server according to an embodiment of the information processing apparatus of the present invention is applied will be described.

[0012] FIGS. 1 to 5 are diagrams showing an overview of a service (hereinafter referred to as "this service") that can be realized by an information processing system to which a server according to an embodiment of the information processing apparatus of the present invention is applied.

[0013] This service supports a user to easily reach a predetermined word, phrase, or sentence presented to the user for a predetermined purpose as a target sentence or the like by using a chat-type UI (User Interface). The target sentence or the like may have any content (content) as long as it is presented to the user to achieve a predetermined purpose of the user. However, in the following description, the target sentence or the like will be described as being an FAQ and its answer sentence (article). Also, such a target sentence or the like will be appropriately referred to as a related sentence or the like. That is, this service will be described as a service that provides a search function of an FAQ search site or the like.

[0014] As shown in FIG. 1, when a user operates an information processing apparatus 2 such as a smartphone (hereinafter referred to as "user terminal 2") to access a website provided by this service and wants to ask a question about the website and obtain an answer, a chat-type UI screen C is popped up and displayed. The user can input a question in the input field B1 of the chat-type UI screen C. The input field B1 allows the input of any text (natural language). However, for the user, it is often difficult to determine what text to input, so they often first input a word In1 related to the content they want to ask. For example, assume that the website in Figure 1 is related to a service that mails a predetermined product (e.g., business card or printed postcard) to the user by a predetermined delivery date, and the user accesses the website today on Thursday and desires the arrival of the delivered item on Saturday, and wants to know if the product will arrive on Saturday with the current order. In this case, assume that the user inputs the word "delivery date" as the word In1 that comes to mind in the input field B1 of the chat-type UI screen C. In this service, through incremental search using intention prediction search, one or more predicted question text candidates QK1, even from a short word In1 like "delivery date", are quickly displayed on the chat-type screen C. In the example of Figure 1, the answer to the user is assumed to be made from the answers in the FAQ. For this reason, as the question text candidate QK1, the FAQ predicted from the word In1 "delivery date" is presented. Also, below the input field B1, one or more keywords related to the input in the input field B1 (the word In1 "delivery date" in the example of Figure 1) are displayed to assist the user in inputting a more appropriate question.

[0015] Next, as shown in Figure 2, assume that the user continues to input the text In2 "Will it meet the delivery date?" in the input field B1 of the chat-type UI screen C. In this case, in this service, more appropriate question text candidates QK2, specifically several question text candidates QK2 such as "What business days after shipment should I choose to meet the delivery date?" are quickly displayed on the chat-type screen C. Although not shown in the figure, when the user clicks on a candidate question sentence QK2 from among several candidate question sentences QK2, such as "What shipping date after how many business days will meet the delivery date?", a page (hereinafter referred to as the "FAQ page") on which the answer to the FAQ corresponding to the clicked candidate question sentence QK2 is posted will open. At this time, in this service, the questions and answers of the FAQ regarding the FAQ page are temporarily memorized and can be reflected in the answers to future users. In other words, it becomes possible to maintain the context in the chat-type UI screen C with the user.

[0016] Next, it is assumed that the user further continues to input "Will it meet the delivery date? It's Saturday" and presses the enter key. As a result, the input in the chat-type UI is finalized, and as shown in FIG. 3, the user's input field B1 is displayed as the chat history, and the answer A1 from the chatbot is displayed. This answer A1 is generated by a chatbot to which an algorithm that enhances the conventional intention prediction search using machine learning is applied. Specifically, it is as follows. That is, "It's a question about whether Saturday is the delivery date and whether it can be delivered in time. In ○○ printing, the goods are shipped from the factory on the shortest possible day, the day after receipt, and the delivery date is the next day to 3 days after shipment. Since today is Thursday, if you confirm the receipt today, it may meet the delivery on Saturday. Please refer to the FAQ for details." The answer A1 is displayed. Also, below that "This answer was generated from the following FAQ [Regarding delivery date and shipping]" The reference information F1 is displayed. Here, the reference information F1 is the source of the FAQ page that was the information source. Thus, according to the response A1 using the chat-based UI screen C of this service, not only a collection of links to relevant articles (pages on which content related to the answers to the user's questions is published) is simply presented, but also a summary of the answer compiled in natural language is appended. When creating such a response A1, the chatbot creates a text that is as conscious as possible of the "parroting" technique, which is a method of intention prediction search, in order to reduce the burden of visually checking the validity of the search results. In this example, since the user is asked "Will I make it on Saturday?", an expression like "You might make it on Saturday" that parrots the question is included at the end of the answer summary of response A1.

[0017] Next, assume that the user enters "By what time do I need to submit my manuscript?" in the input field B of the chat-based UI screen C in Fig. 3 and presses the enter key. As a result, the input in the chat-based UI is finalized, and as shown in Fig. 4, the input field B2 in which "By what time do I need to submit my manuscript?" is entered is displayed as part of the chat history. Then, a response A2 from the chatbot is displayed. Since the previous questions and answers (see Figs. 1 to 3) are remembered, the chatbot creates response A2 using an algorithm that enhances conventional intention prediction search using machine learning based on the premise that "it is necessary to complete the acceptance within today". Specifically, "Your question is about by what time you need to submit your manuscript to complete the acceptance within today. In the case of speed-check submission, it is possible within 24 hours. In the case of operator-check submission, it is until 10:00 p.m. on weekdays." Response A2 as described above is displayed. Also, below that "This answer was generated from the following FAQ [Regarding delivery dates and shipping]" Reference information F2 as described above is displayed. Here, reference information F2 is the source of the FAQ page that was the information source.

[0018] Although not shown in the figure, if the user is not satisfied with the answer from the chatbot, the user can send it as an inquiry as it is. That is, when it seems that a certain amount of conversation has been had and a solution has not been reached, a "Consultation" button is displayed on the chat-type UI screen C simultaneously with the response from the chatbot. When the "Consultation" button is pressed, the human customer support reads the conversation log with the chatbot and then the human customer support responds.

[0019] Also, by embedding the chat-type UI screen C into a predetermined website, the chatbot can also grasp the content of the open page as context and create an appropriate answer by an algorithm that enhances the conventional intention prediction search using machine learning. For example, in the example of FIG. 5, it is assumed that the user has opened a page titled "Price List for A4-Size Flyers and Leaflets" and entered "Where should I enter the data?" in the input field B4 of the chat-type UI screen C and pressed the enter key. As a result, the input in the chat-type UI is finalized, and as shown in FIG. 5, an answer A4 from the chatbot is displayed such that the input field B4 is displayed as part of the chat history. Also, below that, reference information F4 indicating the source of the FAQ page that was the information source is also displayed. This answer A4 was created by the chatbot in accordance with the context of "wanting to print flyers" based on the page titled "Price List for A4-Size Flyers and Leaflets". Note that what the chatbot can use as context is not limited to only the currently open page, and pages that the user has viewed in the past can also be used in the same way.

[0020] In this way, the chat-type UI using the chat-type UI screen C of this service is a combination of the "keyword search-type FAQ" developed by the applicant and the AI chatbot. Here, the "keyword search-type FAQ" will be described.

[0021] The "keyword search type FAQ" is a method that enables a user to quickly reach the information (FAQ and its answers) they desire, even if the keyword entered in the search window is highly general. Here, the "generality" of a keyword refers to an index indicating the breadth of the range of meanings that can be evoked from the keyword itself. A keyword with high generality has a wide range of meanings that can be evoked from it, while a keyword with low generality has a narrow range of meanings that can be evoked from it.

[0022] As a specific example, for instance, assume that overtime extends into the late night and a problem occurs where the in-house terminal of a user who is the only one remaining in the company cannot connect to the Internet. In such a case, the user shall use the "keyword search type FAQ" to input possible keywords into the search window and execute a search in order to quickly reach the FAQ for solving the problem.

[0023] For example, assume that the user inputs the keyword "not connecting" into the search window and executes a search. From the keyword "not connecting", it is possible to evoke a state where something that should be connected in some manner is not connected. However, with only the keyword "not connecting", it is not possible to evoke specifically what and what are not connected in what manner. That is, the keyword "not connecting" is an example of a keyword with high generality.

[0024] Here, assume that the user searches for the keyword "not connecting" using a conventional FAQ search service. In this case, as search results, a large number of FAQ candidates partially including the keyword "not connecting" are displayed. However, since this search result also includes many FAQs not related to Internet connection, it is difficult for the user to reach the desired FAQ.

[0025] For example, assume that a user enters the keyword "Internet" into a search window and executes a search. From the keyword "Internet", the so-called concept of the Internet can be recalled. However, with only the keyword "Internet", specific details such as what the Internet has done cannot be recalled. That is, the keyword "Internet" is an example of a highly general keyword. Here, assume that the user searches for the keyword "Internet" using a conventional FAQ search service. In this case, as search results, a number of FAQ candidates that partially include the keyword "Internet" are displayed. However, since this search result includes many FAQs that are not related to Internet connection, it is difficult for the user to reach the desired FAQ.

[0026] On the other hand, when a "keyword search type FAQ" is used, when a highly general keyword is entered into the search window B, first, one or more FAQ (question text) candidates (hereinafter referred to as "question text candidates") that include at least a part of the keyword are displayed. The question text candidates are generated in advance in a plurality and are stored and managed in a predetermined database (for example, the question text dictionary DB181 in FIG. 8 described later) in a retrievable and extractable manner.

[0027] Specifically, for example, although not shown in the figure, when a highly general keyword "cannot connect" is entered into the search window, a question text candidate "cannot connect to the in-house wireless network" is displayed. Also, for example, when a highly general keyword "Internet" is entered into the search window, a question text candidate "how can I connect to the in-house Internet?" is displayed. In this example, only one question text candidate is displayed in each case, but this is for the convenience of making the explanation easier, and of course, two or more question text candidates may be displayed.

[0028] When a candidate question sentence desired by the user is selected from the one or more presented candidate question sentences, the screen transitions. Then, one or more sentences (hereinafter referred to as "related sentences, etc.") including FAQs having a predetermined relevance to the keyword input by the user and their answers are displayed. Note that there are no particular limitations on how to define "relevance" in the "keyword search type FAQ" (including the chat type UI of this service), and the service provider can arbitrarily define it. Then, keywords, candidate question sentences, and related sentences, etc. whose mutual relevance is recognized by the service provider are associated with each other to generate a question sentence dictionary. That is, the related sentences, etc. associated with and managed by keywords in the question sentence dictionary are treated as having "relevance" to those keywords. The question sentence dictionary is stored and managed in a predetermined database (for example, the question sentence dictionary DB181 in FIG. 5 described later) in a retrievable and extractable manner.

[0029] Specifically, for example, although not shown in the drawings, when a candidate question sentence "not connected to the in-house wireless network" is selected by the user, the following display is made as related sentences, etc. having relevance to the keyword "not connected" input by the user. That is, an FAQ "Connect to the in-house Wi-Fi", "You can connect to the in-house Wi-Fi network with the following ID. · SSID: XXXX · Password: XXXX Do not disclose to people outside the company." The answer to the said FAQ, "When there is a problem with the connection, please contact up to the following. [Contact person] Information Systems Department Kaneko XX" The supplementary information, and related sentences, etc. composed of are displayed. Thereby, the user can quickly solve the problem of not being connected to the Internet by referring to the displayed related sentences, etc.

[0030] For example, although not shown in the figure, when a user selects a candidate question sentence such as "How can I connect to the company's Internet?", the same display is made as relevant articles or the like having relevance to the keyword "Internet" input by the user. That is, when paying attention to the relationship between the keyword and relevant articles or the like, even if the user inputs a keyword such as "not connected" or inputs a keyword such as "Internet", the same relevant articles or the like are immediately presented, and the problem that the user cannot connect to the Internet can be quickly solved.

[0031] As described above, in the "keyword search type FAQ", when a keyword is input into the search window, one or more candidate question sentences containing at least a part of the keyword are displayed. Then, when one candidate question sentence is selected from them, relevant articles or the like including the FAQ related to the keyword and its answer are displayed. Thereby, the user can easily reach the desired FAQ and its answer.

[0032] Here, there are two points worthy of note regarding the "keyword search type FAQ". The first point is that the relevant articles or the like do not include the keyword input for the search. That is, the user can easily reach the desired FAQ by simply inputting a highly common keyword that is easy to come to mind into the search window and selecting the displayed candidate question sentence.

[0033] The second point worthy of note is that even if different keywords are input into the search window, the user can reach the same desired relevant articles or the like. In other words, a plurality of candidate question sentences are associated with one FAQ, and different one or more keywords are associated with each of these plurality of candidate question sentences. That is, when the user wants to obtain a desired FAQ and its answer, the user only needs to input a highly common keyword that comes to mind on the spur of the moment and select a candidate question sentence that is displayed. That is, when a problem such as "cannot connect to the Internet" occurs as in the above example, highly common keywords such as "not connected" or "Internet" may be input.

[0034] In addition, in the "keyword search type FAQ", in related articles and the like, in addition to the FAQ and its answer desired by the user, as described above as additional information, the person in charge related to the answer and the organization to which the person in charge belongs can be displayed. Specifically, for example, in the above example, as additional information for the FAQ and its answer, as the [contact information] for the answer, the person in charge (Kaneko XX) and the organization to which the person in charge belongs (Information Systems Department) are shown. Here, each of the notation of the person in charge and the notation of the organization to which the person in charge belongs may be simply represented by text, but it is preferable that both are displayed in the form of buttons that can access a predetermined web page on which detailed information is posted. In this case, for example, when an operation of pressing (for example, tapping) the display of the person in charge (Kaneko XX) is performed, a predetermined web page (hereinafter referred to as the "detail page") on which detailed information such as department, telephone number, and email address is posted is displayed. Thereby, when the problem cannot be solved only by the related articles and the like (only the presented FAQ and its answer), the user can quickly identify the person in charge and make an inquiry, so that the problem can be solved at an early stage.

[0035] As described above, in the "keyword search type FAQ", a question sentence dictionary is managed in a predetermined database. Thereby, when a candidate question sentence is selected, related articles and the like including the FAQ and its answer can be displayed in the above-described manner. Here, the related text etc. associated with a predetermined keyword does not necessarily have to include the FAQ and its answers, and a predetermined word, phrase, or sentence may be sufficient as long as it is the target text etc. For example, the related text etc. can be displayed in the following manner.

[0036] That is, when a keyword is input into the search window, one or more question sentence candidates are displayed. So far, it is the same as the above example, but a list of related text etc. may be displayed as "word hints" below the search window etc. Specifically, for example, assume that the keyword "address" is input. Then, as "word hints" (related text etc.) corresponding to "address", "purchase", "convenience store", "listing", "address", "anonymity", "change", and "return" are displayed. The user performs an operation (for example, a tapping operation) to select a desired "word hint" (related text etc.) from among the one or more displayed "word hints" (related text etc.). Then, although not shown in the figure, related text etc. including the FAQ and its answers desired by the user are displayed.

[0037] Also, the "keyword search type FAQ" is improved as appropriate. As described above, the keyword, the question sentence candidates, and the related text etc. are associated and managed as a question sentence dictionary. Also, the operation history of the user is stored and managed as history information. That is, the keyword input into the search window by the user, the question sentence candidates selected by the user, and the related text etc. reached by the user are managed as history information. The history information includes information regarding question sentence candidates that were not selected by the user even though they were displayed when a keyword was input into the search window. Also, the history information includes information regarding cases where, even though related text etc. were displayed when a question sentence candidate was selected, it was not what the user desired. Specifically, for example, assume that a certain EC (Electronic Commerce) site has adopted a "keyword search type FAQ". And assume that many users enter the name of a specific payment method in the search window. In this case, although one or more question text candidates are displayed on the user terminals 2 of many users, many users may not select the question text candidates. Also, there may be a case where no question text candidates are displayed. In such a case, since it is possible to infer the question text candidates that many users are expected to desire, based on the result of the inference, new question text candidates can be prepared. Thereby, it can be improved to a more substantial "keyword search type FAQ". As a result, the satisfaction of users who use the "keyword search type FAQ" can be improved. Also, since keywords, related articles, etc. included in the question text candidates not selected by the users, and products and services that are not provided although desired by the users when they do not select to reach them are identified, it can contribute to the creation of new products and new services desired by the users.

[0038] In this way, the "keyword search type FAQ" gives the user a feeling that it is very fast until reaching the target information (related articles, etc.). However, when the "keyword search type FAQ" is applied alone, since the starting point (user input) is a "keyword" after all, and the end point (output to the user) is the presentation of pre-prepared related articles, etc., it also has the weakness of being weak in text. Also, since the content before the input "keyword" in terms of time is irrelevant, for the user, even if there is a series of processes where a "predetermined input" was made earlier and a keyword related to the predetermined input is entered, and even if the user says it is related to the "earlier", on the "keyword search type FAQ" side, that "earlier" does not connect (presents the target article, etc. without considering that "earlier"), which is also a weakness.

[0039] To compensate for such weaknesses, this service provides the chat-type UI shown in FIGS. 1 to 5 above, which combines an AI chatbot with a "keyword search-type FAQ".

[0040] When compared with conventional AI chatbots, since conventional AI chatbots form conversations (chats), they can predict the intention and answer in response to the content input by the user, and can consider the history of previous conversations when making such intention predictions. That is, AI chatbots can eliminate the weaknesses of "keyword search-type FAQs". On the other hand, conventional AI chatbots have the weaknesses of not knowing whether correct results can be obtained until they are sent, and the problem of hallucination, that is, the problem that incorrect answers or answers unrelated to the question occur. Since the "keyword search-type FAQ" can present relevant articles, etc. as correct answers as long as the "keyword" of the input is determined, it can eliminate the weaknesses of these conventional AI chatbots.

[0041] Therefore, this service combines an AI chatbot with a "keyword search-type FAQ" to eliminate their mutual weaknesses. As a result, similar to the "keyword search-type FAQ", it is possible to give the user the feeling that it is very fast to reach the target information (related articles, etc.), and to make the user experience the feeling that the chat-type UI of this service is intelligent.

[0042] Next, with reference to FIG. 6, the configuration of the information processing system for realizing the provision of the above-described service, that is, the information processing system to which the server according to an embodiment of the information processing apparatus of the present invention is applied, will be described. FIG. 6 is a diagram showing an example of the configuration of an information processing system to which a server according to an embodiment of the information processing apparatus of the present invention is applied.

[0043] The information processing system shown in FIG. 6 is configured to include a server 1 and a user terminal 2. The server 1 and the user terminal 2 are interconnected via a predetermined network NW such as the Internet.

[0044] The server 1 is an information processing device managed by a service provider. The server 1 executes various processes for realizing this service while appropriately communicating with the user terminal 2.

[0045] The user terminal 2 is an information processing device operated by a user. The user terminal 2 is composed of a smartphone, a tablet, a personal computer, etc.

[0046] FIG. 7 is a block diagram showing an example of the hardware configuration of the server in the information processing system shown in FIG. 6.

[0047] The server 1 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a bus 14, an input / output interface 15, an input unit 16, an output unit 17, a storage unit 18, a communication unit 19, and a drive 20.

[0048] The CPU 11 executes various processes according to a program recorded in the ROM 12 or a program loaded from the storage unit 18 into the RAM 13. In the RAM 13, data and the like necessary for the CPU 11 to execute various processes are also appropriately stored.

[0049] The CPU 11, ROM 12, and RAM 13 are interconnected via the bus 14. The input / output interface 15 is also connected to this bus 14. The input unit 16, output unit 17, storage unit 18, communication unit 19, and drive 20 are connected to the input / output interface 15.

[0050] The input unit 16 is composed of, for example, a keyboard or the like, and inputs various types of information. The output unit 17 is composed of a display such as a liquid crystal display or a speaker, etc., and outputs various types of information as images or sounds. The storage unit 18 is composed of a DRAM (Dynamic Random Access Memory) or the like, and stores various types of data. The communication unit 19 communicates with other devices (for example, the user terminal 2 in FIG. 4) via a network NW including the Internet.

[0051] A removable medium 40 made of a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory or the like is appropriately mounted on the drive 20. The program read from the removable medium 40 by the drive 20 is installed in the storage unit 18 as necessary. Also, the removable medium 40 can store various types of data stored in the storage unit 18 in the same manner as the storage unit 18.

[0052] Although not shown in the figure, the user terminal 2 in FIG. 6 can also have a configuration basically the same as the hardware configuration shown in FIG. 7. Therefore, the description of the hardware configuration of the user terminal 2 is omitted.

[0053] By the cooperation of various types of hardware and various types of software that make up the information processing system in FIG. 6 including the server 1 in FIG. 7, it becomes possible to execute various processes including a question dictionary generation process, a chat-type related article extraction process, and a question dictionary improvement process. As a result, the service provider can provide the above-described service to the user. The "question dictionary generation process" refers to the process of generating the above-described question dictionary. The "chat-type related article extraction process" refers to the process of extracting related articles or the like using the above-described question dictionary while chatting with the user using the above-described chat-type UI, and outputting them included in the user's answer. The "question dictionary improvement process" refers to the process for improving the above-described question dictionary.

[0054] FIG. 8 is a functional block diagram showing an example of a functional configuration for executing a question sentence dictionary generation process, a chat-type related sentence extraction process, and a question sentence dictionary improvement process among the functional configurations of the server in FIG. 7 that constitutes the information processing system in FIG. 6.

[0055] As shown in FIG. 8, when the server 1 executes the question sentence dictionary generation process, in the CPU 11, the dictionary generation unit 101 functions. Further, when the server 1 executes the chat-type related sentence extraction process, in the CPU 11, the keyword reception unit 102, the question sentence candidate extraction unit 103, the question sentence candidate presentation unit 104, the question sentence determination unit 105, the related sentence presentation unit 106, and the chat-type UI control unit 109 function. Further, when the server 1 executes the question sentence dictionary improvement process, in the CPU 11, the history management unit 107 and the dictionary improvement unit 108 function.

[0056] Also, in an area of the storage unit 18 of the server 1, a question sentence dictionary DB 181 is provided. The question sentence dictionary DB 181 stores a question sentence dictionary.

[0057] The dictionary generation unit 101 associates a keyword, a question sentence candidate, and related sentences, etc., so that a question sentence candidate including a keyword that is more general and is a word related to the whole or a component of the related sentences, etc., is extracted when the keyword is input, and stores it in the question sentence dictionary of the question sentence dictionary DB 181. Thereby, a question sentence dictionary is generated to enable a user who uses the chat-type UI of this service to easily reach the desired information.

[0058] The keyword reception unit 102 receives a keyword input in the input field of the chat-type UI. Here, the keyword input in the input field of the chat-type UI refers to not only the case where the keyword alone is input into the input field, but also the concept that includes a sentence containing the keyword or its related words being input into the input field and the keyword being extracted from the sentence by the chat-type UI control unit 109 (AI chatbot). Specifically, for example, in the example of FIG. 1 described above, since only the keyword "delivery date" is input into the input field B1, the keyword reception unit 102 receives "delivery date". In contrast, in the example of FIG. 2 described above, the sentence "Can it meet the delivery date?" is input into the input field B1. As will be described later, the chat-type UI control unit 109 (AI chatbot) extracts the word "meet※" together with "delivery date" from the sentence. Therefore, the keyword reception unit 102 receives "delivery date" and "meet※". Here, "※" means that any character string can be entered. For example, "meet", "make it meet", and "not meet" are included in "meet※".

[0059] The question sentence candidate extraction unit 103 extracts one or more question sentence candidates including at least a part of the keyword received by the keyword reception unit 102 from the question sentence dictionary in the question sentence dictionary DB181. For example, in the example of FIG. 2, in addition to the question sentence candidate "I want to change the order content (delivery date) including "delivery date", the question sentence candidate "There was a defect in the printed matter, but if it cannot be reprinted because it does not meet the delivery date, what should I do?" including "delivery date" and "not meet", and the question sentence candidate "What should I do to make it meet the delivery date? Should I ship it after what operating expenses?" including "delivery date" and "make it meet" are extracted.

[0060] The question sentence candidate presentation unit 104 presents one or more question sentence candidates extracted by the question sentence candidate extraction unit 103 to the user. Specifically, the question sentence candidate presentation unit 104 presents one or more question sentence candidates (question sentence candidate QK2 in the example of FIG. 2) extracted by the question sentence candidate extraction unit 103 to the user by executing control to display them on the user terminal 2.

[0061] When one question candidate is selected from one or more question candidates presented to the user by the question candidate presentation unit 104, the question text determination unit 105 determines this as the question text from the user. Here, the selection of one question candidate means that, in addition to the case where it is explicitly selected by the user's operation, in this service, it also includes the concept that the chat-type UI control unit 109 (AI chatbot) makes an intention prediction and selects based on the content input in the input field. For example, in the example of Figure 2, when the user clicks on the question candidate "What shipping date after how many business days is needed to meet the delivery date?" among the question candidates QK2, the question candidate can also be determined. Also, as shown in Figure 3, when the user inputs and clicks "Can I meet the delivery date? It's Saturday", the chat-type UI control unit 109 (AI chatbot) makes an intention prediction and selects the question candidate "What shipping date after how many business days is needed to meet the delivery date?" among the question candidates QK2. Therefore, the question text determination unit 105 determines the question candidate.

[0062] The related article etc. presentation unit 106 extracts the related articles etc. associated with the question text determined by the question text determination unit 105 and presents them to the user via the chat-type UI control unit 109 (AI chatbot). Here, presenting to the user via the chat-type UI control unit 109 (AI chatbot) means that, in addition to presenting the related articles etc. as they are, it also includes the concept that the chat-type UI control unit 109 (AI chatbot) processes and presents the related articles etc., or presents an article generated by the chat-type UI control unit 109 (AI chatbot) based on the related articles etc. For example, assume that the answer to the FAQ "What shipping date after how many business days is needed to meet the delivery date?" is "The goods are shipped from the factory on the same day as received at the earliest, and the delivery date is the next day to 3 days after shipping." and this answer is extracted as the related articles etc. This content is included in the answer A1 from the chat-type UI control unit 109 in the example of Figure 3. Furthermore, upon receiving the user's input that the "delivery date" is "Saturday", the chat-type UI control unit 109 (AI chatbot) uses the above-mentioned parrot-like method to create a sentence based on related texts and the like, such as "Since today is Thursday, if you confirm the reception within today, you might make it in time for Saturday. For details, please refer to the FAQ." and includes it in the answer A1. Therefore, the answer A1 in the example of FIG. 3 is an example of related texts and the like presented to the user via the chat-type UI control unit 109 (AI chatbot).

[0063] The history management unit 107 manages, as history information, the keyword received by the keyword etc. reception unit 102, the question sentence candidate selected from among one or more question sentence candidates presented to the user by the question sentence candidate presentation unit 104, and the related texts and the like presented to the user by the related texts and the like presentation unit 106 in response to the selection. In addition, the history management unit 107 extracts keywords that are the keywords received by the keyword etc. reception unit 102 but for which the related texts and the like that were not selected by the user were not presented by the related texts and the like presentation unit 106.

[0064] When the question sentence dictionary improvement unit 108, as history information by the history management unit 107, the question sentence candidate presented by the question sentence candidate presentation unit 104 is not presented to the user and the confirmation by the question sentence confirmation unit 105 is not performed, the question sentence dictionary improvement unit 108 improves the question sentence dictionary including the correspondence between the keyword included in the question sentence candidate and the question sentence candidate.

[0065] The chat-type UI control unit 109 displays the chat-type UI screen C on the user terminal and, while executing control for the AI chatbot to chat (have a conversation) with the user, controls the exchange of information between the keyword etc. reception unit 102 to the related texts and the like presentation unit 106 and the AI chatbot that establish the "keyword search type FAQ", so that for a sentence including the keyword input by the user, an appropriate answer based on the related texts and the like (FAQ and its answer) to be presented by the related texts and the like presentation unit 106 is generated for the AI chatbot.

[0066] Here, as described above, the AI chatbot executes various processes until it generates an answer to the user's input in the chat based on an algorithm that enhances conventional intention prediction search using machine learning.

[0067] For example, the AI chatbot extracts keywords such as those received by the keyword reception unit 102 from the text input by the user. Specifically, for example, in the example of FIG. 2 above, the text "Can I meet the delivery date?" is input in the input field B1. Therefore, the AI chatbot extracts the word "delivery date" and the word "meet※" from the text.

[0068] As a result, the question sentence candidate extraction unit 103 displays the question sentence candidate QK2 to the user, and the input sentence "Can I meet the delivery date? It's Saturday" is thrown at the AI chatbot by the user (see the input field B1 as the history in the example of FIG. 3). Then, the AI chatbot determines, as the question sentence, the question sentence candidate "What business days after shipment should I arrange to meet the delivery date?" among the question sentence candidates QK2 from this input sentence.

[0069] When an answer "The goods will be shipped from the factory on the shortest possible acceptance day, and the delivery date will be the next day to 3 days after shipment." is extracted as related text or the like for this question sentence (FAQ), the AI chatbot refers to the user's input "It's Saturday" (see the input field B1 as the history in the example of FIG. 3) and generates the answer A1 in the example of FIG. 3 based on the related text or the like. That is, the AI chatbot creates the answer A1 in natural language as the summary text of the answer to the user's question. As described above, the chatbot creates responses not only by including relevant sentences as they are, but also by including sentences that are conscious of the "parroting" method of intention prediction search as much as possible in order to reduce the burden of visually checking the validity of search results. In this example, since the user is asked "Can I make it by Saturday?", a parroting expression "I might make it by Saturday" is included at the end of the response summary of Answer A1.

[0070] Furthermore, the AI chatbot can execute various processes for generating the current response based not only on the content input by the user this time, but also on the content of the conversation (chat) up to the previous time and the history of processing from the reception unit 102 to the related article presentation unit 106 based on the keywords and the like based on the conversation up to the previous time. For example, as a continuation of the example of FIG. 3 described above, as shown in FIG. 4, assume that "By what time should I submit?" is input. In this case, since the above-mentioned questions and answers (see FIGS. 1 to 3) are stored in the AI chatbot, based on the premise that "it is necessary to complete the reception by the end of today", the answer A2 in the example of FIG. 4 can be generated by an algorithm that strengthens the conventional intention prediction search using machine learning.

[0071] Furthermore, when the chat-type UI screen C of this service is embedded in a predetermined website (for example, in the case of the example of FIG. 5), the AI chatbot can execute various processes for generating the current response based not only on the content input by the user, but also on the content of one or more web pages included in the website. For example, in the example of FIG. 5, with the page "A4 size flyer / leaflet printing price list" open, the user inputs "Where should I enter the data?" in the input field B4 of the chat-type UI screen C and presses the enter key. Therefore, the AI chatbot generates Answer A4 in line with the context of "wanting to print a flyer" based on the page "A4 size flyer / leaflet printing price list". Note that what the AI chatbot can use as context is not limited to only the currently open page, and the pages that the user has viewed in the past can also be used in the same way.

[0072] As described above, since the information processing apparatus in FIG. 7 has the above-described functional configuration shown in FIG. 8, the user who uses this service can easily reach the desired information and can receive the desired information in a more appropriate form of natural language.

[0073] Hereinafter, it will be further explained why a user who uses the chat-type UI of this service to which the "keyword-type FAQ" is applied can easily reach the desired information (desired FAQ and its answer), unlike a conventional FAQ search site.

[0074] FIG. 9 is a diagram showing an example of a search on a conventional approach FAQ search site. That is, in the conventional approach, the character string input by the user himself / herself is received, and a list of pages (articles) of the FAQ and its answer that contain the character string is presented to the user as the ones that have been hit. Then, the user has to select an article that he / she would have searched for from the list. However, usually, many articles that contain the character string (search keyword) input by the user are hit.

[0075] As a result, there were cases where many incorrect answers could not be selected. That is, although there may be a correct answer (the article to be presented to the user) somewhere in the list, there were cases where the user himself / herself did not know which article in the list should be viewed. Therefore, the user had to check the list in order from the top (from the beginning).

[0076] Also, as a result, there were cases where nothing was listed. That is, articles containing the character string input by the user are listed. Therefore, if the character string input by the user is not a term or the like used in the article, nothing is listed, and the user has to reconsider on their own and input another character string.

[0077] Assuming the conventional approach of listing articles containing the character string input by the user, the following countermeasures can be considered. That is, it is conceivable to prepare a large number of articles. Specifically, for example, by preparing multiple patterns of articles in which words are replaced with other words or the way of approaching the problem is changed, it is possible to ensure that any character string input by the user will hit in the search. However, this countermeasure also has the drawback that although there will be hits, misfitting articles will also be hit. Also, for example, there may be multiple hits of similar articles (answers), causing confusion. Also, the maintainability deteriorates.

[0078] Also, for example, instead of creating a list on the condition that the article contains the character string input by the user, there has been a conventional method of adopting an AI or the like that evaluates the relevance between the character string input by the user and the article to be presented to the user. When a conventional machine learning type AI is adopted, the situation where there are no hit articles becomes less likely. However, the problem that the list of hit articles contains many articles that are not the articles originally intended by the user is not solved. Also, when a conventional AI is adopted, articles with a high degree of relevance as determined by the AI are presented as a result, but there are cases where answers that pose compliance or business problems are given, and it is difficult to reliably remove them. Also, when a conventional AI is adopted, it is vulnerable to changes in the service or business environment. That is, since past search results and the like are used for the learning of the AI, it becomes difficult to respond in a short period of time when the number of specific inquiries increases, or when the service or business environment or the service itself changes. In addition, when using conventional AI, a large amount of data is required for individual tuning for each search target (e.g., service). Also, a large amount of engineering resources are required for cleaning the large amount of data, resulting in high costs.

[0079] Thus, the conventional approach and the method using conventional AI, etc. are nothing but evaluating the relationship between the character string input by the user and the article (including evaluating the relationship of simply being included or not), and have the above-mentioned problems. On the other hand, users who use the chat-type UI of this service to which "keyword-type FAQ" is applied can easily reach the desired information (FAQ and its answer articles), unlike the above-mentioned conventional FAQ search sites. The reason will be explained below.

[0080] FIG. 10 is a diagram for explaining the concept of intention expansion in this service. In the example of FIG. 10, the target text T1 to be presented to the user is the FAQ of "Guide on refund method" and its answer article (related articles, etc.). In fact, the information presented to the user in this service is not the FAQ of "Guide on refund method" and its answer article itself, but the information based on the FAQ and its answer. Here, "the information based on the FAQ and its answer" means the following. That is, since a chat-type UI is used in this service, "the FAQ and its answer" are not presented to the user as they are, but are presented as natural language answers generated by the AI chatbot. The natural language answers generated by this AI chatbot are "the information based on the FAQ and its answer". However, for the sake of convenience of explanation, the following explanation will be given assuming that the target text, etc. is presented to the user. Here, the article "Guide on Refund Methods" (target text, etc.) should be presented to users of e-commerce sites, etc., who are users who "are troubled by receiving defective products", users who "received a product different from the ordered product", or users who, for some reason, "want their money returned".

[0081] Therefore, in this service, the technical writer expands the target text, etc. T1 into the intended expression (intermediate) T2 (expands three times in the example of Fig. 8). That is, the technical writer expands into an expression considering what intention the user has. That is, in Fig. 8, for example, the technical writer assumes a user who intends to obtain an answer by conveying the fact that "there is a defect in the product" and expands. Also, for example, the technical writer assumes a user who intends to obtain an answer by conveying the fact that "a different product was received" and expands. Also, for example, the technical writer assumes a user who intends to obtain an answer by conveying the hope (solution method) of "wanting a refund" and expands. In this way, the technical writer expands from the target text, etc. T1 into the intended expression (intermediate) T2 from the perspective of for what intention users the article should be presented.

[0082] And further, the intended expression (intermediate) T2 is further expanded by the server 1. That is, the server 1 expands the intended expression (intermediate) T2 into the intended expression T3 by a predetermined algorithm (expands 16 times in the example of Fig. 8). Specifically, for example, as shown in Fig. 8, the intended expression (intermediate) T2 of "There is a defect" is expanded into "The product is broken", "The product was a defective product", etc. That is, for example, the server 1 expands by supplementing the subject, replacing words with synonyms, etc., and changing the combination of endings, auxiliary verbs, etc. to form a synonymous sentence. The intended expression T3 generated (expanded) in this way is stored in the above-mentioned question sentence dictionary DB181 as a question sentence candidate. Similarly, each of the plurality of target sentences, etc. T1 is developed through the stages of the intended expression (intermediate) T2 and the intended expression T3, and the developed ones are stored in the above-described question sentence dictionary DB181 as question sentence candidates.

[0083] As a result, the user can perform the following searches. FIG. 11 is a diagram showing an example of user guidance realized by the intention development of this service. That is, for example, as shown in FIG. 11, the user will be able to perform a search by entering, as words (keywords) that come to mind, the intentions such as "fugu taste", "different", and "refund". That is, since a plurality of questions (intended expressions) are generated for "a certain article" (target sentence, etc.), the user will be able to reach that article from various expressions. In this service, as shown in FIGS. 10 and 11, even when the keywords input by the user are different (for example, when they are different like "defective product" and "fugu taste", "different" and "not the same", "return the money" and "refund"), appropriate question sentence candidates will be presented to the user. This will be described later with reference to FIGS. 13 and 14.

[0084] The above is an explanation of the intention development, which is one of the differences between this service and the conventional FAQ search site. Next, other functions, etc. of this service that improve the convenience of the user will be described.

[0085] FIG. 12 is a diagram showing an example of user guidance realized by intention prediction in this service. As shown in FIG. 12, the server 1 of this service has a function of predicting the user's intention just by one character being input by the user. As a result, the user can confirm the question sentence candidates without inputting all the question sentences (keywords). That is, in this service, while the user is entering a keyword, there is a prediction function that predicts, from the characters entered at that time, what keyword the user ultimately intends to enter. The prediction function is adjusted for each service that is the search target of this service. That is, for example, in the case of an FAQ search site of a service where the term "point" is used, when the user enters "po", it is predicted that the keyword "point" will be entered. Also, for example, in the case of an FAQ search site of a service where the term "portfolio" is used, when the user enters "po", it is predicted that the keyword "portfolio" will be entered.

[0086] Note that even when the prediction function is functioning, as described above, question text candidates containing the keyword are presented to the user. That is, usually, when the user enters the keyword "point", even if the user is a natural person, the intention regarding what the user wants to do with the point cannot be grasped. However, in this service, since question text candidates containing the keyword are presented to the user, the user can surely reach an answer by selecting a question text candidate according to their intention.

[0087] Hereinafter, the feature that question text candidates containing the keyword entered by the user are presented to the user will be supplemented and explained. Figure 13 is a diagram showing an example of the presentation of question text candidates in this service. As shown in Figure 13, when a keyword is entered by the user, question text candidates containing the keyword are presented to the user. Then, the user can confirm the target text, etc. by selecting a question text candidate according to their intention from the presented question text candidates. Server 1 associates the definitions of terms, explanatory texts, synonyms, paraphrases, etc., and stores and manages them as a definition database. Specifically, for example, in the example of FIG. 13, "automatic braking device" and "ASV" are associated and stored as a definition database. Then, when Server 1 searches for candidate question sentences based on the keyword input by the user, it uses the definition database to search for candidate question sentences and presents them to the user. As a result, the intended expansion T3 of "What is an automatic braking device?" is hit. Here, when the user searches using the keyword "ASV", what is presented to the user as a candidate question sentence is the one obtained by replacing "automatic braking device" in the intended expansion T3 of "What is an automatic braking device?" with the word "ASV". As a result, since the user's words are "necessarily" included in the question, candidate question sentences that do not depend on the user's search level are presented to the user. In other words, at the level of the user's words, the system (Server 1) can present corresponding candidate question sentences. From the user's perspective, it becomes possible for the user to search using their own words, and it becomes easier to select from multiple candidate question sentences. In other words, the user only needs to input a keyword, and candidate question sentences expressed using the keyword input by the user are presented, so it becomes easier to select from multiple candidate question sentences.

[0088] FIG. 14 is a diagram showing an example in which a user selects a candidate question sentence in this service. As shown in FIG. 14, candidate question sentences containing the keyword input by the user are presented to the user. As a result, as described with reference to FIG. 13, it becomes possible for the user to search using their own words. And as a result of the user selecting a candidate question sentence containing the keyword input by themselves, they can reach the target text corresponding to that question and so on.

[0089] As described above, the target text and the like have been explained as related texts such as FAQs and the texts (articles) of answers thereto. However, as described above, this service can be applied to a predetermined word, phrase, or text presented to the user in order to achieve the user's predetermined purpose as the target text.

[0090] That is, for example, the target text and the like may be an explanatory text of a product (service). Specifically, for example, assume that the user has the intention of "being worried about the education funds for children" on the premise of a financial service. In this case, the user inputs a keyword such as "child" in the input field displayed together with the text "What do you want to ask?" displayed on the financial service site. As a result, candidates for the question text such as "I want to prepare education funds for my children" and "What age children is ○○ targeted at?" are presented to the user. Here, "○○" is the service name of the service for preparing education funds for children, that is, the so-called school expense insurance service. Next, by selecting the candidate question text "I want to prepare education funds for my children" that has not yet been verbalized and input as a keyword, the user can view the product description page of "○○", which is school expense insurance.

[0091] Note that the above-mentioned keyword and candidate question text also function appropriately for a user who intends to use the ○○ service and wants to know "up to what age children are targeted". Also, as described above, when a product introduction is adopted for the target text and the like of this service, the chat-type UI can be made to function not as a part of the FAQ search site but as a dedicated page for product introduction. In this way, this service can be applied not only to FAQs but also to a predetermined word, phrase, or text presented to the user in order to achieve the user's predetermined purpose. In the above-described embodiment, it is assumed that the user inputs keywords on the premise of trying to input their own intention in the form of a question sentence. However, what is presented to the user is not limited to a question sentence as long as it expresses the user's own intention, and a guiding sentence desired (intended) by the user is sufficient.

[0092] The functional configuration of the server 1 that presents candidates for such guiding sentences, target sentences, etc. will be described with reference to FIG. 15. FIG. 15 is a functional block diagram showing an example of the functional configuration for executing a process of causing the user to search for and present a target sentence, etc. among the functional configurations of the server in FIG. 8.

[0093] As shown in FIG. 15, when the server 1 executes the guiding sentence dictionary generation process, in the CPU 11, the dictionary generation unit 111 functions. Further, when the server 1 executes the chat-type target sentence extraction process, in the CPU 11, the keyword reception unit 112, the guiding sentence candidate extraction unit 113, the guiding sentence candidate presentation unit 114, the guiding sentence selection reception unit 115, the target sentence presentation unit 116, and the chat-type UI control unit 119 function. Further, when the server 1 executes the guiding sentence dictionary improvement process, in the CPU 11, the history management unit 117 and the dictionary improvement unit 118 function.

[0094] Also, in an area of the storage unit 18 of the server 1, a guiding sentence dictionary DB 182 is provided. The guiding sentence dictionary DB 182 stores a guiding sentence dictionary.

[0095] The dictionary generation unit 111 associates a keyword, a guiding sentence candidate, and a target sentence, etc. so that a guiding sentence candidate including a keyword with a higher degree of generality, which is a word related to the whole or a constituent element of the target sentence, etc., is extracted when the keyword is input, and stores it in the guiding sentence dictionary of the guiding sentence dictionary DB 182. Thereby, a guiding sentence dictionary is generated to enable a user who searches for a target sentence, etc. to easily reach the desired information.

[0096] The keyword reception unit 112 receives the keyword K input in the input field of the chat-type UI. Specifically, for example, in the above example of school expense insurance, when words such as "child" and "education funds" or their synonyms are input, the keyword reception unit 112 receives keywords such as "child" and "education funds" via the chat-type UI control unit 119 (AI chatbot).

[0097] The leading sentence candidate extraction unit 113 extracts one or more leading sentence candidates including at least a part of the keyword received by the keyword reception unit 112 from the leading sentence dictionary in the leading sentence dictionary DB 181.

[0098] The leading sentence candidate presentation unit 114 presents one or more leading sentence candidates extracted by the leading sentence candidate extraction unit 113 to the user. Specifically, the leading sentence candidate presentation unit 114 presents one or more leading sentence candidates extracted by the leading sentence candidate extraction unit 113 to the user by executing control to display them on the user terminal 2.

[0099] When one of the one or more leading sentence candidates presented to the user by the leading sentence candidate presentation unit 114 is selected, the leading sentence selection reception unit 115 determines this as the leading sentence from the user. Here, when one leading sentence candidate is selected, in addition to the case where it is explicitly selected by the user's operation, in this service, it also includes the concept that the chat-type UI control unit 119 (AI chatbot) makes an intention prediction and selects based on the content input in the input field.

[0100] The target sentence presentation unit 116 extracts the target sentence and the like associated with the leading sentence determined by the leading sentence selection reception unit 115, and presents it to the user via the chat-type UI control unit 119 (AI chatbot). Here, presenting to the user via the chat-type UI control unit 119 (AI chatbot) includes, in addition to presenting the target sentence or the like as it is, the chat-type UI control unit 119 (AI chatbot) processing and presenting the target sentence or the like, or presenting a sentence generated by the chat-type UI control unit 119 (AI chatbot) based on the target sentence or the like.

[0101] The history management unit 117 manages, as history information, the keyword received by the keyword or the like reception unit 112, the candidate guidance sentence selected by the user from among one or more candidate guidance sentences presented to the user by the candidate guidance sentence presentation unit 114, and the target sentence or the like presented to the user by the target sentence or the like presentation unit 116 in response to the selection. In addition, the history management unit 117 extracts keywords that are keywords received by the keyword or the like reception unit 112 but for which the target sentence or the like not selected by the user was not presented by the target sentence or the like presentation unit 116.

[0102] When the candidate guidance sentence presented by the candidate guidance sentence presentation unit 114 is not presented to the user and the confirmation by the guidance sentence selection reception unit 115 is not performed after that, the dictionary improvement unit 118 improves the guidance sentence dictionary including the correspondence between the keyword included in the candidate guidance sentence and the candidate guidance sentence, using the history information managed by the history management unit 117.

[0103] The chat-type UI control unit 119 displays the chat-type UI screen C on the user terminal, and while executing control for the AI chatbot to chat (have a conversation) with the user, controls the exchange of information between the keyword or the like reception unit 112 to the target sentence or the like presentation unit 116 and the AI chatbot, so as to generate, for the AI chatbot, an appropriate response based on the target sentence or the like extracted by the target sentence or the like presentation unit 116 for the sentence including the keyword input by the user.

[0104] As described above, an example of the functional configuration of the server 1 applied to the target sentence or the like including not only FAQs but also product descriptions has been explained. The features of this service realized by the above-mentioned server 1 are summarized below.

[0105] The search using the intended expansion of this service can be said to be a search method that fits well with the user's behavior characteristics. Specifically, the user is a layperson regarding products (services) to be searched for. Therefore, the user often does not understand the technical terms usually used in FAQs or product descriptions. That is, the user cannot come up with keywords for searches using such technical terms. Furthermore, even if the user enters a keyword, it is difficult to select which one is the correct answer (which one matches the user's own intention) from the list of articles that simply contain the keyword. In addition, in reality, many users will give up if they cannot reach the target text they intend after several (on average, about three times) attempts. As described above, this service can solve such problems.

[0106] Also, the search using the intended expansion of this service can be said to be a mechanism that searches for predicted guiding texts (question texts) instead of searching for target texts as answers from the keywords input by the user. Specifically, it is difficult for the user to select an appropriate answer even if the user is presented with an answer (target text) that they want to know in the future. In this service, since the questions (guiding texts) that the user himself / herself is thinking about are presented, it becomes easier for the user to make a selection. As a result, the correct answer reach rate of the user is improved.

[0107] The above describes one embodiment of the present invention. However, the present invention is not limited to the above-described embodiment, and modifications, improvements, etc. within the scope that can achieve the object of the present invention are considered to be included in the present invention.

[0108] For example, in the above-described embodiment, the related articles, which are an example of the target text, etc. are regarded as the answer to one FAQ, but they may also be regarded as the answers to two or more FAQs.

[0109] Also, for example, it is not limited to the above specific examples. For example, assume that incorrect payment processing has been performed for three users using online payment. In this case, each of the three users inputs a natural sentence or the like including keywords such as "problem", "refund the money", and "defective" into the chat-type UI screen C of this service, and selects the displayed question sentence candidates respectively (including the case where the AI chatbot makes an intention prediction and selects). Then, the answers to the same FAQ desired by the three users are extracted as related articles or the like, and articles based on the same related articles or the like are presented to the three users respectively. Although the articles presented to each of the three users are based on the same related articles or the like, since they are generated by the AI chatbot according to the history of each of the three chats (conversations), they are generally different for each of the three users.

[0110] Also, for example, in the above-described embodiment, question sentence candidates including a word that is more general among words related to a predetermined word, phrase, or sentence are extracted when the word is input, but it is not limited to this. It is also possible to extract question sentence candidates including a word that is less general or has the same degree of generality among words related to a predetermined word, phrase, or sentence when the word is input. Specifically, for example, when the word "erase" is input into the search window, an article including the word "delete" may be extracted.

[0111] Also, the system configuration shown in FIG. 6 and the hardware configuration of the server 1 shown in FIG. 7 are merely examples for achieving the object of the present invention and are not particularly limited.

[0112] Also, the functional block diagrams shown in FIGS. 8 and 15 are merely examples and are not particularly limited. That is, it is sufficient that the information processing system of FIG. 6 is provided with a function capable of executing the above-described various processes as a whole, and the functional blocks and databases used to realize this function are not particularly limited to the examples of FIGS. 8 and 15.

[0113] Also, the locations of the functional blocks and the database are not limited to FIGS. 8 and 15, and may be arbitrary. For example, at least a part of the functional blocks and the database arranged on the server 1 side may be configured to be provided on the user terminal 2 side or other information processing apparatuses (not shown).

[0114] Also, the above-described series of processes can be executed by hardware or by software. Also, one functional block may be configured by hardware alone, by software alone, or by a combination thereof.

[0115] When the series of processes are executed by software, the program constituting the software is installed in a computer or the like from a network or a recording medium. The computer may be a computer incorporated in dedicated hardware. Also, the computer may be a computer capable of executing various functions by installing various programs, for example, a general-purpose smartphone or personal computer other than a server.

[0116] A recording medium containing such a program is not only constituted by a removable medium (not shown) distributed separately from the apparatus main body to provide the program to the user, but also constituted by a recording medium or the like provided to the user in a state pre-installed in the apparatus main body.

[0117] Note that in this specification, the steps of describing the program recorded on the recording medium include not only processes performed in chronological order according to the order, but also processes that are not necessarily processed in chronological order and are executed in parallel or individually.

[0118] To summarize the above, it suffices for the information processing apparatus to which the present invention is applied to have the following configuration, and various embodiments can be adopted. That is, the information processing apparatus to which the present invention is applied has reception means (for example, the keyword reception unit 102 in FIG. 8, the keyword reception unit 112 in FIG. 15) that receives words (for example, keywords) related to a target sentence or the like (for example, a related sentence such as an answer to an FAQ) by using a predetermined word, phrase, or sentence presented to the user to achieve a predetermined purpose of the user as the target sentence or the like; extraction means (for example, the question candidate extraction unit 103 in FIG. 8, the guiding sentence candidate extraction unit 113 in FIG. 15) that extracts one or more guiding sentence candidates (for example, question candidates) including at least a part of the word or a similar word from a guiding sentence dictionary (for example, the question sentence dictionary DB181 in FIG. 8, for example, the guiding sentence dictionary DB182 in FIG. 15) in which a plurality of guiding sentence candidates (for example, question sentence candidates) intended by the user to reach the target sentence or the like are registered in advance in association with the target sentence or the like; first presentation means (for example, the question candidate presentation unit 104 in FIG. 8, the guiding sentence candidate presentation unit 114 in FIG. 15) that presents the one or more question candidates extracted by the extraction means to the user (for example, presents them like the question candidate QK1 in FIG. 1); second presentation means (for example, the related sentence presentation unit 106 in FIG. 8, the target sentence presentation unit 116 in FIG. 15) that extracts the target sentence or the like associated with the selected guiding sentence candidate from the guiding sentence dictionary among the one or more guiding sentence candidates presented to the user by the first presentation means and presents it to the user; While the chatbot executes control for chatting with the user (for example, while displaying the chat-type UI screen C on the user terminal 2 as shown in FIGS. 1 to 5), by controlling the exchange of information between the reception means to the second presentation means and the chatbot, for the text input by the user (for example, the text input in the input field B1 in FIG. 3), a chat control means (for example, the chat-type UI control unit 109 in FIG. 8, the chat-type UI control unit 119 in FIG. 15) that causes the chatbot to generate a response (for example, response A1 in FIG. 3) based on the target text such as the target text to be presented by the second presentation means. An information processing apparatus having this is sufficient. For example, the leading text candidate can be a question text candidate, and the leading text dictionary can be a question text dictionary.

[0119] As a result, the user can easily reach the desired target text or the like through the chat and obtain an appropriate response based on the target text or the like.

[0120] The chatbot can be configured to execute one or more processes before generating the response to the user's input in the chat based on an algorithm for intention prediction search generated using predetermined machine learning. As a result, for example, it becomes possible to present response A1 in the example of FIG. 3 to the user for the content input in input field B1 in the example of FIG. 3.

[0121] The chatbot can execute the one or more processes including a process of extracting the words received by the reception means based on the input content of the user. For example, in the example of FIG. 2, the text "Can it meet the delivery date?" is input in input field B1. Therefore, the AI chatbot can extract the word "delivery date" and the word "meet※" from the text.

[0122] In addition to the user's current input content, the chatbot can execute the above-described one or more processes including a process of extracting the word received by the reception means based on the chat content up to the previous time and the history of the processing of the reception means to the second presentation means based on the chat up to the previous time. For example, as a continuation of the example in FIG. 3, as shown in FIG. 4, it is assumed that the input is "By what time should I submit?" In this case, since the above-described previous questions and answers (see FIGS. 1 to 3) are stored in the AI chatbot, it is possible to generate Answer A2 in the example of FIG. 4 on the premise that "it is necessary to complete the reception by the end of today".

[0123] The chatbot is embedded in a predetermined website, and in addition to the user's current input content, it can execute the above-described one or more processes including a process of extracting the word received by the reception means based on the content of one or more web pages included in the website. For example, in the example of FIG. 5, with the page "A4 size leaflet / flyer printing rate table" open, the user enters "Where should I enter the data?" in the input field B4 of the chat-type UI screen C and presses the enter key. Therefore, the AI chatbot can generate Answer A4 in line with the context of "wanting to print a leaflet" based on the page "A4 size leaflet / flyer printing rate table".

Explanation of Reference Numerals

[0124] 1... Server, 2... User terminal, 11... CPU, 12... ROM, 13... RAM, 14... Bus, 15... Input / output interface, 16... Input section, 17... Output section, 18... Storage section, 19... Communication section, 20... Drive, 40... Removable media, 101... Dictionary generation section, 102... Keyword reception section, etc., 103... Question sentence candidate extraction section, 104... Question sentence candidate presentation section, 105... Question sentence determination section, 106... Related article presentation section, etc., 107... History management section, 108... Dictionary improvement section, 109... Chat-type UI control section, 111... Dictionary generation section, 112... Keyword reception section, etc., 113... Guidance sentence candidate extraction section, 114... Guidance sentence candidate presentation section, 115... Guidance sentence selection reception section, 116... Target article presentation section, etc., 117... History management section, 118... Dictionary improvement section, 119... Chat-type UI control section, 181... Question sentence dictionary DB, 182... Question sentence dictionary DB

Claims

1. Receiving means for receiving words related to a target sentence or the like, with a predetermined word, phrase, or sentence presented to the user to achieve a predetermined purpose of the user as the target sentence or the like; Extracting means for extracting one or more candidate guiding sentences including at least a part of the word or a word similar thereto from a guiding sentence dictionary in which a plurality of candidate guiding sentences intended by the user to reach the target sentence or the like are registered in advance in association with the target sentence or the like; First presenting means for presenting the one or more candidate guiding sentences extracted by the extracting means to the user; Second presenting means for extracting the target sentence or the like associated with the selected candidate guiding sentence from among the one or more candidate guiding sentences presented to the user by the first presenting means and presenting it to the user; Chat control means for controlling the exchange of information between the receiving means to the second presenting means and the chatbot while the chatbot executes control for chatting with the user, so that the chatbot generates an answer based on the target sentence or the like to be presented by the second presenting means for the sentence input by the user; comprising The chatbot, under the control of the chat control means, executes one or more processes until generating the answer to the input of the user in the chat based on an algorithm for intention prediction search generated using predetermined machine learning, and executes the one or more processes including a process of extracting the word received by the receiving means based on the current input content of the user, the chat content up to the previous time, and the history of the processes of the receiving means to the second presenting means based on the chat up to the previous time among the input content of the user; An information processing apparatus.

2. Receiving means for receiving words related to a target sentence or the like, with a predetermined word, phrase, or sentence presented to the user to achieve a predetermined purpose of the user as the target sentence or the like; Extracting means for extracting one or more candidate guiding sentences including at least a part of the word or a word similar thereto from a guiding sentence dictionary in which a plurality of candidate guiding sentences intended by the user to reach the target sentence or the like are registered in advance in association with the target sentence or the like; First presenting means for presenting the one or more candidate guiding sentences extracted by the extracting means to the user; A second presenting means for extracting, from the induction text dictionary, the target sentence or the like associated with the selected induction text candidate among the one or more induction text candidates presented to the user by the first presenting means, and presenting the same to the user; A chat control means for controlling the exchange of information between the reception means to the second presenting means and the chatbot while the chatbot executes control for chatting with the user, so that the chatbot generates, for the sentence input by the user, a response based on the target sentence or the like to be presented by the second presenting means; comprising; The chatbot, under the control of the chat control means, executes one or more processes until generating the response to the input of the user in the chat based on an algorithm for intention prediction search generated using predetermined machine learning, and is embedded in a predetermined website, and executes the one or more processes including a process of extracting the word received by the reception means based on the current input content among the input contents of the user and the contents of one or more web pages included in the website; An information processing apparatus.

3. The induction text candidate is a question text candidate, and the induction text dictionary is a question text dictionary. The information processing apparatus according to claim 1 or 2.

4. In an information processing method executed by an information processing apparatus, a reception step of receiving, as a target sentence or the like, a predetermined word, phrase, or sentence presented to the user to achieve a predetermined purpose of the user, and receiving a word related to the target sentence or the like; an extraction step of extracting one or more induction text candidates including at least a part of the word or a word similar thereto from an induction text dictionary in which a plurality of induction text candidates intended by the user to reach the target sentence or the like are registered in advance in association with the target sentence or the like; a first presentation step of presenting the one or more induction text candidates extracted in the extraction step to the user; a second presentation step of extracting, from the induction text dictionary, the target sentence or the like associated with the selected induction text candidate among the one or more induction text candidates presented to the user in the first presentation step, and presenting the same to the user; While executing control for the chatbot to chat with the user, by controlling the exchange of information between the chatbot in the reception step to the second presentation step, for the text input by the user, the chatbot is caused to generate a response based on the target text such as the target text in the second presentation step. A chat control step, including under the control in the chat control step, the chatbot executes one or more processes until generating a response to the user's input in the chat based on an algorithm for intention prediction search generated using predetermined machine learning, and among the input contents of the user, based on the current input content, the chat contents up to the previous time, and the history of the processes in the reception step to the second presentation step based on the chat up to the previous time, the chatbot executes the one or more processes including a process of extracting the word received in the reception step. An information processing method. In the information processing method executed by an information processing apparatus according to claim 5, a reception step of receiving, as a target text or the like, a predetermined word, phrase, or sentence presented to the user to achieve a predetermined purpose of the user, and receiving a word related to the target text or the like; an extraction step of extracting one or more candidate guiding texts including at least a part of the word or a word similar thereto from a guiding text dictionary in which a plurality of candidate guiding texts intended by the user to reach the target text or the like are registered in advance in association with the target text or the like; a first presentation step of presenting the one or more candidate guiding texts extracted in the extraction step to the user; a second presentation step of extracting the target text or the like associated with the selected candidate guiding text from the guiding text dictionary among the one or more candidate guiding texts presented to the user in the first presentation step and presenting it to the user; While executing control for the chatbot to chat with the user, by controlling the exchange of information between the chatbot in the reception step to the second presentation step, for the text input by the user, the chatbot is caused to generate a response based on the target text such as the target text in the second presentation step. A chat control step, including under the control in the chat control step, the chatbot Based on the algorithm of intention prediction search generated using predetermined machine learning, one or more processes are executed until the response to the user input in the chat is generated, and is embedded in a predetermined website, and one or more processes including a process of extracting the word received in the reception step are executed based on the current input content among the input contents of the user and the contents of one or more web pages included in the website. Information processing method.

6. On a computer, a reception step of receiving, as a target sentence or the like, a predetermined word, phrase, or sentence presented to the user to achieve a predetermined purpose of the user, and receiving a word related to the target sentence or the like; an extraction step of extracting one or more induction sentence candidates including at least a part of the word or a word similar thereto from an induction sentence dictionary in which a plurality of induction sentence candidates intended by the user to reach the target sentence or the like are registered in advance in association with the target sentence or the like; a first presentation step of presenting the one or more induction sentence candidates extracted in the extraction step to the user; a second presentation step of extracting the target sentence or the like associated with the selected induction sentence candidate from among the one or more induction sentence candidates presented to the user in the first presentation step, and presenting it to the user; while the chatbot executes control for chatting with the user, by controlling the exchange of information between the chatbot in the reception step to the second presentation step, for the sentence input by the user, a chat control step of causing the chatbot to generate a response based on the target sentence or the like to be presented in the second presentation step; causing a control process including the above to be executed, as the chat control step, further causing the chatbot to execute one or more processes until the response to the user input in the chat is generated based on the algorithm of intention prediction search generated using predetermined machine learning, and Among the input content of the user, based on the current input content, the chat content up to the previous time, and the history of the processing in the reception step to the second presentation step based on the chat up to the previous time, execute one or more processes including a process of extracting the word received in the reception step. Execute control processing including steps. Program.

7. A computer A reception step of receiving, as a target sentence or the like, a predetermined word, phrase, or sentence presented to the user to achieve a predetermined purpose of the user, and receiving a word related to the target sentence or the like. An extraction step of extracting one or more leading sentence candidates including at least a part of the word or a word similar thereto from a leading sentence dictionary in which a plurality of leading sentence candidates intended by the user to reach the target sentence or the like are registered in advance in association with the target sentence or the like. A first presentation step of presenting the one or more leading sentence candidates extracted in the extraction step to the user. A second presentation step of extracting the target sentence or the like associated with the selected leading sentence candidate from the leading sentence dictionary among the one or more leading sentence candidates presented to the user in the first presentation step and presenting it to the user. While the chatbot executes control for chatting with the user, by controlling the exchange of information between the chatbot in the reception step to the second presentation step, for the sentence input by the user, generate an answer based on the target sentence or the like to be presented in the second presentation step for the chatbot. A chat control step. Execute control processing including As the chat control step, further Cause the chatbot to execute one or more processes until generating an answer to the input of the user in the chat based on an algorithm of intention prediction search generated using predetermined machine learning, and Embedded in a predetermined website, among the input content of the user, based on the current input content and the content of one or more web pages included in the website, execute one or more processes including a process of extracting the word received in the reception step. Program.

Images