Information processing system, information processing method, and program

Abstract

This information processing system is equipped with a control unit. The control unit: inputs, into a generative AI, at least one piece of question information relating to at least one question item on a questionnaire survey; on the basis of the question information and a series of response information from an respondent to the question information, as obtained from a chat between the generative AI and the respondent, sets a prompt in the generative AI so as to generate a set of shaped sentences in which content indicated by the question information and the response information is expressed from the viewpoint of the respondent; and receives the generated shaped sentences from the generative AI.

Description

Information Processing System, Information Processing Method, and Program

[0001] The present invention relates to an information processing system, an information processing method, and a program capable of executing a chat-form web questionnaire with a user terminal using a generative AI.

[0002] Conventionally, there has been a technology for conducting a questionnaire on the web. For example, in Patent Document 1 below, it is disclosed that a server device uses artificial intelligence to transmit the content of each question constituting the target questionnaire as a conversation sentence of a pseudo-conversation to a user terminal, and temporarily stores in a memory the answers to each question returned from the user terminal.

[0003] Japanese Patent Application Laid-Open No. 2004-094463

[0004] By the way, when realizing the above-mentioned questionnaire in the conversation form using a generative AI, it is conceivable to let the generative AI also organize the answer information. However, in that case, so-called "AI summarization" is generally performed, and the summarization operation is often completely entrusted to the ability of the AI. As a result, the frequency of hallucination occurrence depends on the writing style of the original text, so there is a problem that it is difficult to prevent hallucination in summarizing questionnaire answer information.

[0005] In view of the above circumstances, an object of the present invention is to provide an information processing system, an information processing method, and a program capable of shaping the answer information of a chat-type questionnaire using a generative AI into an easy-to-read text while preventing hallucination.

[0006] To achieve the above object, an information processing system according to one embodiment of the present invention includes a control unit. The control unit inputs at least one piece of question information regarding at least one question item of a questionnaire survey to a generative AI, and based on a series of answer information of a respondent to the question information obtained by chatting between the generative AI and the respondent and the question information, sets a prompt to the generative AI to generate a coherent shaped text expressing the content indicated by the question information and the answer information from the perspective of the respondent, and receives the generated shaped text from the generative AI.

[0007] This configuration allows the information processing system to rearrange question and answer information into text from the respondent's perspective, thereby formatting the responses from chat-type surveys using generation AI into easy-to-read text while preventing hallucination.

[0008] The control unit may set the prompt to generate a declarative sentence as the formatted text by combining the question expression indicated by the question information and the affirmative or negative expression indicated by the answer information.

[0009] This configuration allows the information processing system to easily format answer information by generating declarative sentences from question and answer information. For example, if the answer to the question "Do you think XX is YY?" is "Yes," the declarative sentence "I think XX is YY." is generated as formatted text. If the answer to the same question is "No," the declarative sentence "I don't think XX is YY." is generated as formatted text.

[0010] The control unit may remove the question information and answer information from the document if the answer information to any of the question information is not a meaningful answer or does not represent the opinion of the respondent.

[0011] This configuration prevents the information processing system from including meaningless responses in the formatted text. Meaningless responses here refer to sentences that are linguistically incomprehensible, as well as sentences that do not contain any opinion or intention of the respondent and are of no significance to the survey implementer, such as the answer "I don't have any particular opinion" to a question like "Do you have any opinion on XX?".

[0012] The control unit may set the prompt to generate the text by rearranging multiple pairs of question information and answer information in the reverse order of the question and answer sequence.

[0013] With this configuration, the information processing system, due to the nature of AI-generated chat, tends to start with a predetermined conclusion, and later responses in the chat tend to be more specific or informative. By reversing the above order, it becomes possible to generate easy-to-read text that explains the details (causes, etc.) first before reaching the conclusion.

[0014] The aforementioned question items may include quantitative question information and qualitative question information. In this case, the control unit may set the prompt to the generation AI to generate the text based on the respondent's series of answers to the qualitative question information and the respondent's answers to the quantitative question items that are prerequisites for the qualitative question information.

[0015] This configuration allows the information processing system to generate more accurate formatted text by taking into account the answers to quantitative questions that form the basis of the chat regarding qualitative questions.

[0016] An information processing method according to another embodiment of the present invention includes inputting at least one question information relating to at least one question item of a questionnaire survey into a generating AI; prompting the generating AI to generate a set of sentences that express the content indicated by the question information and answer information from the perspective of the respondent, based on a series of answer information from the respondent to the question information obtained through a chat between the generating AI and the respondent; and receiving the generated sentences from the generating AI.

[0017] A program according to yet another embodiment of the present invention causes an information processing device to perform the following steps: input at least one question information relating to at least one question item of a questionnaire survey into a generating AI; set a prompt for the generating AI to generate a set of sentences that express the content indicated by the question information and answer information from the perspective of the respondent, based on a series of answer information from the respondent to the question information obtained through a chat between the generating AI and the respondent; and receive the generated sentences from the generating AI.

[0018] As described above, according to the present invention, it is possible to format the response information from a chat-type questionnaire using generation AI into easy-to-read text while preventing hallucination. However, this effect is not limited to the present invention.

[0019] This figure shows the configuration of a chat-type survey system according to one embodiment of the present invention. This figure shows the hardware configuration of a survey management server according to one embodiment of the present invention. This figure shows the configuration of the database owned by the survey management server. This is a flowchart showing the flow of the chat prompt generation process for a survey by the survey management server. This figure shows an example of a chat function management screen output by the survey management server for generating chat prompts. This figure shows an example of the editing screen of the chat function management screen in Figure 5. This figure shows an example of a chat preview screen that the survey management server outputs to the generation AI. This is a flowchart showing the flow of a chat-type survey by the survey management server. This figure shows examples of items for quantitative and qualitative surveys conducted as a prerequisite for the chat-type survey. This figure shows an example of a chat screen that the survey management server outputs to the generation AI. This is a flowchart showing the flow of the chat formatting process by the survey management server.

[0020] Embodiments of the present invention will be described below with reference to the drawings.

[0021] [System Configuration] Figure 1 shows the configuration of the chat-type survey system according to this embodiment.

[0022] As shown in the figure, this system includes a survey management server 100 on the Internet 50, a user terminal 200, and multiple monitor terminals 300.

[0023] The survey management server 100 is a server (information processing device) that conducts web-based surveys requested by survey requesters based on instructions from user terminals 200. The survey management server 100 is connected to the user terminals 200 and multiple monitor terminals 300 via the Internet 50.

[0024] Specifically, the survey management server 100 sends a survey response request to the monitor terminal 300, which includes a request to access the survey screen containing the survey questions, and receives the response information from the monitor terminal 300. The survey management server 100 then aggregates and analyzes this response information and sends it to the user terminal 200 as survey result information.

[0025] The user terminal 200 is a terminal used by the person conducting the survey (the survey company or the user of the survey information), and is, for example, a smartphone, mobile phone, tablet PC (Personal Computer), notebook PC, or desktop PC. The user terminal 200 generates survey information and has the survey management server 100 manage it, and based on the survey information, it has the survey management server 100 conduct the survey with the monitor terminal 300 and receive the survey result information.

[0026] The monitor terminals 300 (200A, 200B, 200C, etc.) are terminals of users who have a contract with the survey implementer to act as respondents (monitors) to surveys conducted by the survey management server 100. These terminals include, for example, smartphones, mobile phones, tablet PCs (personal computers), notebook PCs, and desktop PCs.

[0027] The monitor terminal 300 receives a request to answer a survey from the survey management server 100, inputs answers to the questions using an application or browser, and sends the answer information to the survey management server 100.

[0028] In this embodiment, the survey management server 100 is connected to a generation AI such as ChatGPT, which is a learning model based on a large-scale language model (LLM), via an API (Application Programming Interface).

[0029] Traditionally, online surveys have shown that respondents are not very cooperative. As a result, response rates to quantitative questions (answered with multiple-choice options) are high, while cooperation rates to qualitative questions (collecting opinions and attitudes in text format) are low.

[0030] For example, asking questions like, "Do you have any complaints about the ~~~ service?" only yielded superficial responses such as "difficult to use" or "not good," making it difficult to use the large amount of text collected for "research purposes" such as service improvement or product development.

[0031] Therefore, the inventors have devised a web survey system that uses generation AI to provide appropriate stimuli to respondents, thereby simultaneously acquiring the intent and context of their statements. This system is useful for service improvement and enables a more realistic analysis of the respondents' behavior.

[0032] In other words, the survey management server 100 generates chat prompts for qualitative questions among the questions in the web survey, in order to obtain answers that are in line with the purpose of the survey through a conversation (chat-style interview) between the generating AI and the monitor terminal 300, and sets these prompts to the generating AI via API. The survey management server 100 also formats (cleanses) the series of answer information obtained through the chat between the generating AI and the monitor terminal 300 into a single sentence and outputs it to the user terminal 200. Details of these chat prompt generation and chat formatting processes will be described later.

[0033] [Hardware Configuration of Survey Management Server] Figure 2 shows the hardware configuration of the survey management server 100. As shown in the figure, the survey management server 100 includes a CPU (Central Processing Unit) 11, ROM (Read Only Memory) 12, RAM (Random Access Memory) 13, an input / output interface 15, and a bus 14 that connects these components to each other.

[0034] The CPU 11 accesses RAM 13 and other memory as needed, performing various calculations and comprehensively controlling each block of the survey management server 100. ROM 12 is a non-volatile memory in which the OS, programs, and firmware such as various parameters to be executed by the CPU 11 are permanently stored. RAM 13 is used as a working area for the CPU 11 and temporarily holds the OS, various running applications, and various data being processed.

[0035] The input / output interface 15 is connected to a display unit 16, an operation reception unit 17, a storage unit 18, a communication unit 19, and the like.

[0036] The display unit 16 is a display device that uses, for example, an LCD (Liquid Crystal Display), an OLED (Organic ElectroLuminescence Display), or a CRT (Cathode Ray Tube).

[0037] The operation reception unit 17 is, for example, a pointing device such as a mouse, a keyboard, a touch panel, or other input device. If the operation reception unit 17 is a touch panel, the touch panel may be integrated with the display unit 16.

[0038] The storage unit 18 is a non-volatile memory such as an HDD (Hard Disk Drive), flash memory (SSD; Solid State Drive), or other solid-state memory. The OS, various applications, and various data are stored in this storage unit 18.

[0039] In this embodiment in particular, the storage unit 18 stores applications and other programs for the survey management server 100 to execute web surveys. As will be described later, the storage unit 18 has a monitor information database, a chat prompt information database, and a survey information database as databases containing such data.

[0040] The communication unit 19 is various modules for wireless communication such as, for example, an NIC (Network Interface Card) for Ethernet or a wireless LAN, and is responsible for communication processing between the user terminal 200 and the monitor terminal 300.

[0041] Although not shown, the hardware configurations of the user terminal 200 and the monitor terminal 300 are also the same as the hardware configuration of the questionnaire management server 100. Also, an application for creating and managing Web questionnaire survey information (hereinafter also referred to as a questionnaire app) is installed in the user terminal 200, and communication with the questionnaire management server 100 is enabled via the questionnaire app.

[0042] [Database Configuration of Questionnaire Management Server] Fig. 3 is a diagram showing the configuration of the database possessed by the questionnaire management server 100.

[0043] As shown in the figure, the questionnaire management server 100 has a monitor information database 31, a chat prompt information database 32, and a questionnaire information database 33 in the storage unit 18.

[0044] The monitor information database 31 stores information about monitors (users of the monitor terminal 300), who are consumers / service users who have concluded a monitoring contract with the operator of the questionnaire management server 100. Specifically, the monitor information database 31 stores, for example, user ID, password, name, email address, phone number, address (residence), age (group), gender, family composition (married / unmarried, or number of family members, presence or absence of children, etc.), occupation (job type, school), annual income (household annual income, personal annual income), SNS in use, housing form (detached house, condominium, owned house, rental, etc.), mobile phone (carrier information), automobile ownership status, presence or absence of a license, etc., and information on hobby-related concerns for each monitor.

[0045] The chat prompt information database 32 stores chat prompt information generated for a chat (interview) with the monitor conducted by the generation AI in the Web questionnaire, which is input from the user terminal 200 on the chat management screen described later.

[0046] The chat is executed to elicit answers from the monitor that conform to the purpose of the Web questionnaire survey for qualitative questions. As chat prompt information, for each qualitative question item, information such as the purpose of the Web questionnaire survey, the purpose of the qualitative question, the selection information of the chat type (for digging deeper into answers or for broadening the scope of answers), and the identification information (question number) of the previous question (quantitative or qualitative question) to be referred to as a premise is stored. The chat prompt information database 32 also stores base chat prompts for the chats by the generation AI for each chat type.

[0047] The questionnaire information database 33 stores information related to the Web questionnaire created for the monitor, that is, the question information (survey form) of the Web questionnaire, data necessary for displaying the questionnaire screen, the answer information received from each monitor terminal 300 in the conducted questionnaire, the formatted data obtained by formatting the answer information for qualitative questions, the aggregated data of the answers for quantitative questions, and the like.

[0048] In addition to the chat prompts, the questionnaire information database 33 also stores formatting prompts for causing the generation AI to execute chat formatting processing described later. Although details will be described later, the formatting prompts mainly instruct the deletion of meaningless answers, the synthesis of questions and answers, and the rearrangement of the order of answers.

[0049] These databases are mutually referred to and used as necessary in the chat prompt generation processing and chat formatting processing by the questionnaire management server 100 described later.

[0050] [Operation of the Survey Management Server] Next, the operation of the survey management server 100 configured as described above will be explained. This operation is performed through the cooperation of the hardware of the survey management server 100, such as the CPU 11 and communication unit 19, and the software stored in the storage unit 18. For convenience, in the following explanation, the CPU 11 will be considered the main operator.

[0051] (Chat prompt generation process) Figure 4 is a flowchart showing the flow of the chat prompt generation process by the survey management server 100. As a prerequisite for this process, it is assumed that the user terminal 200 is running the survey application mentioned above.

[0052] First, the CPU 11 receives input of a set of questions for the web survey from the user terminal 200 via the survey application mentioned above, and generates a set of questions (step 41). The set of questions is generated for each question with a question number (Q1, Q2, Q3, etc.) that identifies the question. The questions include quantitative questions and qualitative questions.

[0053] Quantitative questions are questions in which respondents select numerical values ​​or attributes from a set of options, such as "What is your gender?" (options: male / female), "What is your age?" (options: under 10 / 20s...60s / 70s and over), or "How satisfied are you with XX? Please answer on a scale of 0 to 10." (options: 0 / 1...9 / 10). The options are set using radio buttons, for example.

[0054] Qualitative questions are questions that require respondents to provide free-response answers about their opinions, actions, impressions, etc., such as, "What was the reason for giving Q3 a score of ○ for satisfaction?" or "Please let us know if there are any points that concern you regarding XX."

[0055] Next, the CPU 11 places chat-type questions for the questions selected by the user from the created set of questions, based on the input from the user terminal 200 (step 42). The survey application is assumed to have a user interface, such as a button, for placing chat-type questions for each of the created set of questions.

[0056] Next, the CPU 11 receives a selection of the chat-type questions to be managed from the user terminal 200 among the chat-type questions arranged above, and displays the chat function management screen on the survey application (step 43).

[0057] Figure 5 shows an example of the chat function management screen. As shown in the figure, the chat function management screen displays a list of placed chat-type questions. For each question, the following items are displayed: question name, chat type, chat purpose, reference question number, and last updated date. When the edit button 70 provided for each question is pressed on the user terminal 200, the CPU 11 overlays the editing screen for each item, as shown in Figure 6, onto the list.

[0058] Next, the CPU 11 accepts input for the purpose of the chat-type question survey and the purpose of the chat on the editing screen mentioned above, and also accepts selection of the chat type and reference question number (step 44).

[0059] As shown in Figure 6, the editing screen includes fields for selecting a reference question number 71, a chat type 72, and a chat purpose input field 73. The reference question number selection field 71 allows users to select quantitative or qualitative questions preceding the chat-type questions in the target survey, for example, using checkboxes. The chat type selection field 72 allows users to select the type of chat-type question, for example, using a dropdown menu. The chat purpose input field 73 allows users to input the detailed purpose of conducting the chat-type questions as text. Although not shown in the figure, the editing screen also includes an input field for the purpose of the survey. After the user has made selections and entered information for each of the above items, the editing is completed when the user presses a confirmation button 74, for example, located at the bottom of the editing screen.

[0060] The purpose of the survey could be, for example, "to discover areas for improvement in a specific business." The purpose of the chat refers to what kind of information you want to elicit from the respondents. For example, if a certain value (e.g., low satisfaction) is obtained in a quantitative question, you can further ask why they gave that value, and if they answer that reason (e.g., high price), you can describe in detail what improvements would eliminate that reason. Ultimately, you can describe the information you want to elicit from the respondents in this way.

[0061] Chat types include in-depth questioning (aiming to elicit deeper reasons and pursue further details) and broadening the scope (aiming to hear a wide range of opinions). The former involves repeatedly asking questions such as "Why is that?" in response to the respondent's answer to elicit deeper reasons, while the latter involves repeatedly asking questions such as "What about XX?", "Anything else?", and "What are your next concerns?" in response to the respondent's answer to elicit a variety of opinions and reasons other than the initial answer, without ignoring the respondent's opinion.

[0062] The reference question number identifies the previous question (quantitative or qualitative) that should be referenced (or used as a prerequisite) when conducting the above-mentioned chat-type question. For example, if you want to know the supporting evidence for the service satisfaction question in Q2, then "Q2" would be selected as the reference question number.

[0063] Next, the CPU 11 obtains a base chat prompt corresponding to the chat type (step 45). This base chat prompt consists of instructions for establishing a chat and instructions for the purpose of the conversation in line with the theme. When instructions corresponding to the purpose of the survey, such as "This time, in a mobile carrier satisfaction survey...", are added to this base chat prompt to generate a new prompt, the content of the prompt is rewritten to match the purpose.

[0064] Examples of suitable base chat prompts include, but are not limited to, the following. The following are base chat prompts that broaden the scope of the chat; different types of base chat prompts are provided for deeper conversations. In practice, the operation is performed using a set of these base chat prompts and "conversation rule (speaking style and conversation length) prompts."

[0065] "You are a marketer for a research company. You will conduct interviews with research participants on various research topics. Through these interviews, we expect you to: (1) gain a deep understanding of the research participants' information regarding the research topic; (2) gain a certain level of insight into the participants' personalities, attitudes, and environments; and (3) combine the superficial information from (1) (consumer behavior and habits) with the internal information from (2) (likes, background, family structure, hobbies) to derive certain insights. The topic for this assignment is "○○". (Within the scope of the discussion on ○○) What kind of person is this person? What kind of ○○ habits does this person have? What kind of ○○ environment does this person have? Does this person have any particular preferences? Are they satisfied with their current ○○ environment? We would like to gradually find out these things. When the participant seems likely to talk a lot, or has a lot to say, try to lead the conversation less and instead focus on nodding and showing empathy." Conversely, for those who "don't talk much spontaneously and don't seem to have much to say," try stimulating them from various perspectives with questions like, "What do you think in this situation?", "How do you feel about that?", or "Are you familiar with △△?"

[0066] Next, the CPU 11 combines the information on the purpose, type, and reference question number of the chat selected and entered on the editing screen with the base chat prompt to generate a chat prompt for the target chat-type question (step 46).

[0067] Next, the CPU 11 inputs the generated chat prompt to the generation AI (step 47), and uses this chat prompt to virtually chat with the respondent, causing the generation AI to generate a predicted example of a chat (question and answer) and display it as a preview (step 48).

[0068] Figure 7 shows an example of the preview screen for the virtual chat. As shown in the figure, multiple patterns of the virtual chat are generated based on several different virtual answers to previous quantitative and qualitative questions (for example, 8 points / 2 points as the answer to Q3, and "because the price is cheap" and "because the connection is bad" as the answers to Q4).

[0069] Then, if the user terminal 200 inputs an operation to register approval in response to the above preview display (Yes in step 49), the CPU 11 registers the generated chat prompt in the chat prompt information database 32 (step 50).

[0070] On the other hand, if the user terminal 200 inputs an operation to reject registration in response to the above preview display (No. in step 49), the CPU 11 returns to step 44, prompts the user terminal 200 to re-edit the chat purpose, reference question number, etc., and repeats the subsequent processing.

[0071] (Chat processing using chat prompts) Next, we will explain the operation of the survey management server 100 when executing a web survey using the chat prompts generated as described above. Figure 8 is a flowchart showing the flow of a chat-type survey by the survey management server 100.

[0072] As shown in the figure, when the CPU 11 receives a request to start the survey from the monitor terminal 300, it retrieves the survey information stored in the survey information database 33 and starts the web survey (step 51). That is, it generates a survey screen based on the survey information and displays it on the application or browser of the monitor terminal 300. Prior to this, the CPU 11 sends a message or push notification to the monitor terminal 300 in response to a request from the user terminal 200, which includes a hyperlink or the like for accessing the survey start screen.

[0073] When the survey is started, the CPU 11 first receives information from the monitor terminal 300 regarding quantitative and qualitative questions, such as monitor attributes and satisfaction levels (step 52).

[0074] Figure 9 shows examples of such quantitative and qualitative questions and their responses. In the example shown, the quantitative questions include gender (Q1), age (Q2), and career recommendation level (Q3), while the qualitative question is the reason for the recommendation level in Q3 (Q4).

[0075] Next, the CPU 11 determines whether the survey has proceeded to a qualitative question for which a chat prompt has been registered (i.e., whether or not it has received the answer information to the question immediately preceding the qualitative question) (step 53).

[0076] If the CPU determines that the survey has not progressed to the above qualitative questions (No. in step 53), the CPU 11 displays the information from the previous questions on the monitor terminal 300 and receives the corresponding answer information (step 52).

[0077] If the CPU determines that the survey has progressed to the qualitative questions (Yes in step 53), the CPU 11 retrieves the chat prompt from the chat prompt information database 32 along with the reference question number information, inputs it into the generating AI along with the answer information to the previous quantitative questions, and instructs the AI ​​to start a chat for the qualitative questions (step 54).

[0078] The generating AI performs a chat with the monitor terminal 300 according to the chat prompt and receives information answering qualitative questions (step 55).

[0079] Figure 10 shows an example of a chat screen output by the generating AI. As shown in the figure, the chat screen displays the question information 81 from the generating AI and the answer information 82 received from the monitor terminal 300 alternately, for example, by dividing the area into left and right sections.

[0080] Below these, an answer input field 83 and an answer button 84 are displayed. When the monitor enters an answer in the answer input field 83 and presses the answer button 84, the answer is sent from the monitor terminal 300 to the generating AI and displayed as new answer information 82.

[0081] In this case, the survey management server 100 can increase the respondent's motivation to answer by providing various patterns of expressions of agreement and empathy in the chat prompt.

[0082] In the example shown in the diagram, when a monitor was asked why the support was poor, they replied, "I called 3-4 times on weekdays, but I couldn't get through." The AI ​​then responded with empathy and agreement, saying, "That's unfortunate. I hope they make it easier to get through."

[0083] Other examples of chats that include empathy and agreement include the following: (Example 1) Respondent: "Because the customer experience wasn't good." Generating AI: "That's a shame. What aspects were you dissatisfied with?" (Example 2) Respondent: "The plan descriptions all sound the same." Generating AI: "Thank you. When they all sound the same, it can be hard to know what's truly valuable to you. Are there any services or benefits in these plans that particularly interest you?"

[0084] In addition, it can generate chat prompts such as "I see," "That's true," and "Me too," without negating the respondent's statements, to facilitate smooth conversation.

[0085] Next, the generating AI determines whether sufficient information has been obtained from the respondent through a chat following the above chat prompt (step 56).

[0086] If the AI ​​determines that it has not received a sufficient response (No. in step 56), it continues the chat to receive response information from the monitor terminal 300 (step 55).

[0087] On the other hand, if the AI ​​determines that it has received a sufficient response (Yes in step 56), it terminates the chat using the above chat prompt (step 57).

[0088] Next, the CPU 11, based on the series of response information received via the chat and the corresponding question information, instructs the generation AI to generate a set of formatted text that expresses the content of the question information and response information from the respondent's perspective, in accordance with the formatting prompt (step 58).

[0089] The CPU 11 then receives the formatted text generated by the generation AI, stores it in the survey information database 33, and outputs it to the user terminal 200 (step 59).

[0090] Figure 11 is a flowchart showing the detailed flow of the chat formatting process performed by the survey management server 100 in step 58 of Figure 8.

[0091] As shown in the figure, the generating AI that receives the formatting prompt first integrates the chat data executed by the chat prompt, the chat model, the information of the answers to the underlying quantitative questions, and the reference question information (step 61).

[0092] Next, the generating AI removes non-meaningful responses or responses that do not represent the respondent's opinion from the integrated response information (step 62). Non-meaningful responses include sentences that are linguistically nonsensical, as well as sentences that do not contain any opinion or intention of the respondent and are of no significance to the survey implementer, such as the answer "I don't have any particular opinion" to a question like "Do you have any opinion on XX?".

[0093] Next, the generating AI synthesizes the question expression from the question information and the affirmative or negative expression from the answer information into a declarative sentence from the respondent's perspective (step 63). For example, in the chat example in Figure 10, the question information, "I wish it were easier to connect," and the answer information, "Yes," synthesize the declarative sentence, "I wish it were easier to connect." Also, the question information, "So, are there any concerns regarding connection quality or area coverage?" and the answer information, "No," synthesize the declarative sentence, "There are no concerns regarding connection quality or area coverage." In addition, for example, the question information, "Are you satisfied with the price?" and the answer information, "Yes," synthesize the declarative sentence, "I am satisfied with the price."

[0094] Next, the generating AI rearranges multiple pairs of question and answer information in the reverse order of the original question and answer sequence to generate text (step 64). For example, in the top two pairs of question and answer information in Figure 10, the conclusion is "the support is poor," followed by the reason for that feeling, and then a request. However, by rearranging the order of this conclusion, reason, and request, the text is formatted to say, "I would like the phone support to be easier to get through to. I called 3-4 times on a weekday, but I couldn't get through. Because of that experience, I feel that the quality of support is poor."

[0095] Thus, due to the nature of AI-generated chat, the conclusion is often predetermined, and later responses tend to be more specific or informative. By rearranging the order as described above, it becomes possible to generate easy-to-read text that explains the details (causes, etc.) first before reaching the conclusion.

[0096] The generating AI then sends the formatted text, processed as described above, to the survey management server 100 (step 65).

[0097] [Summary] As described above, according to this embodiment, when conducting a web survey using generation AI, the survey management server 100 can easily generate prompts that allow the survey implementer to obtain the necessary answers, even without possessing advanced technical skills.

[0098] In other words, according to this embodiment, by creating a new chat prompt for each web-based survey, it is possible to conduct interviews that are as aligned as possible with the survey objectives, even when the purpose of the survey, the direction of the specific information to be extracted by the AI ​​generation, or the amount of information about the respondents obtained before the interview differs.

[0099] Furthermore, while conventional so-called "AI summarization" tended to have a high frequency of hallucination, this embodiment allows for the restructuring of question and answer information into text from the respondent's perspective, thereby formatting the response information from a chat-type questionnaire using generation AI into easy-to-read text while preventing hallucination.

[0100] [Modifications] The present invention is not limited to the embodiments described above, and can be modified in various ways without departing from the spirit of the invention.

[0101] In the above embodiment, an example was shown in which quantitative questions were placed in the first half of the questionnaire survey, followed by qualitative questions. However, the questionnaire survey may consist only of qualitative questions, without any quantitative questions. In this case, the chat type set in the above chat prompt should be a broad type.

[0102] The survey management server 100 described in the above embodiment may be configured as a system in which its functions are distributed across multiple servers or information processing devices. For example, at least one or each of the following processes—the generation of the chat prompt, the implementation of the survey using the prompt, and the generation of formatted chat text—may be performed on separate servers.

[0103] Of the inventions described in the claims of this application, the invention described as "information processing method" is one in which each step is performed automatically by at least one device such as a computer through information processing by software, and not by a human using a computer or other device. In other words, the "information processing method" is an information processing method using computer software, and not a method in which a human operates a computer as a calculating tool.

Claims

1. An information processing system comprising: inputting at least one question information relating to at least one question item of a survey into a generating AI; prompting the generating AI to generate a set of formatted text that expresses the content indicated by the question information and answer information from the perspective of the respondent, based on a series of response information from the respondent to the question information obtained through a chat between the generating AI and the respondent; and a control unit that receives the formatted text generated from the generating AI.

2. An information processing system according to claim 1, wherein the control unit sets the prompt to generate a declarative sentence as the formatted text by combining the question expression indicated by the question information and the affirmative or negative expression indicated by the answer information.

3. An information processing system according to claim 1 or 2, wherein the control unit removes the question information and answer information from the text if the answer information to any of the question information is not a meaningful answer or does not represent the opinion of the respondent.

4. An information processing system according to claim 1 or 2, wherein the control unit sets the prompt to generate the text by rearranging a plurality of pairs of question information and answer information in the reverse order of the question and answer order.

5. An information processing system according to claim 3, wherein the question item comprises quantitative question information and qualitative question information, and the control unit sets the prompt to the generation AI to generate the text based on a series of response information from the respondent to the qualitative question information and the respondent's response information to the quantitative question item that is the premise of the qualitative question information.

6. An information processing method comprising: inputting at least one question information relating to at least one question item of a survey into a generating AI; prompting the generating AI to generate a coherent text that expresses the content indicated by the question information and answer information from the perspective of the respondent, based on the series of answer information from the respondent obtained through a chat between the generating AI and the respondent; and receiving the generated text from the generating AI.

7. A program that causes an information processing device to perform the following steps: inputting at least one question information relating to at least one question item of a questionnaire survey into a generating AI; setting a prompt for the generating AI to generate a set of sentences that express the content indicated by the question information and answer information from the perspective of the respondent, based on a series of answer information from the respondent to the question information obtained through a chat between the generating AI and the respondent; and receiving the generated sentences from the generating AI.

Images