Information processing systems, information processing methods, programs

Abstract

The present invention aims to provide a mechanism that allows for the appropriate understanding of the content of the information being analyzed. [Solution] The present invention is characterized by comprising: an output means that outputs information to be analyzed; a pre-set viewpoint for analyzing the information; an instruction to extract topics from the information based on the viewpoint; an instruction to classify the information into topics; and an acquisition means that acquires summary information generated based on the instructions from the output means.

Description

Technical Field

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

Background Art

[0002] Voices of customers (Voice of Customer, hereinafter VoC) collected through questionnaires and the like are very important information in product planning and marketing. However, since VoC in text form is qualitative data, a lot of time and labor are required for sorting and analysis. Furthermore, when there is a large amount of text data, it becomes even more difficult to grasp its content.

[0003] Patent Document 1 discloses a word extraction method performed in a word extraction device that extracts topic words in a specified period from a sequentially updated document set. The method includes a document collection step in which a document collection means collects documents updated since the previous collection from a document storage source at a specified cycle, a target document set creation step in which a target document set creation means creates a document set to be the target of word extraction by collecting documents updated within a period specified independently of the cycle from the collected documents, a word extraction step in which a word extraction means performs word extraction by performing word splitting based on a character string statistic amount, a character string length, and a character type from the target document set, and a topic degree calculation step in which a topic degree calculation means calculates the importance in the target document set and the importance in a document set in a certain period before the target document set for the extracted words, and calculates the degree of increase in importance compared to the previous time as the topic degree of the word, and a topic word list creation step in which a topic word list creation means creates a topic word list in which words are arranged in descending order of topic degree.

[0004] As described above, in Patent Document 1, words and sentences that appear in text data are used as they are as labels representing topics. However, the topic labels obtained by that method may be short and their content may not be understood, or the degree of abstraction may be low, and they may not be suitable for use as they are as topic labels for analysis purposes.

[0005] Furthermore, for example, when analyzing product VoC data, there may be cases where one wants to analyze reputation regarding the product's "features," "price," and "defects." However, Patent Document 1 has the problem that, because it uses words and sentences appearing in the text data as labels, it is not possible to extract topics from the VoC data according to perspectives such as "features," "price," and "defects." [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2005-258678 [Disclosure of the Invention] [Problems that the invention aims to solve]

[0007] Therefore, the present invention aims to provide a mechanism that can appropriately grasp the content of the information to be analyzed. [Means for solving the problem]

[0008] The present invention is characterized by comprising: an output means that outputs information to be analyzed; a pre-set viewpoint for analyzing the information; an instruction to extract topics from the information based on the viewpoint; an instruction to classify the information into topics; and an acquisition means that acquires summary information generated based on the instructions from the output means. [Effects of the Invention]

[0009] According to the present invention, it is possible to provide a mechanism that allows for the appropriate understanding of the content of the information being analyzed. [Brief explanation of the drawing]

[0010] [Figure 1] This figure shows an example of the system configuration of an analysis report generation device in an embodiment of the present invention. [Figure 2]This is a block diagram showing an example of the hardware configuration of an analysis report generation device in an embodiment of the present invention. [Figure 3] This is a flowchart showing an example of the analysis report generation process in an embodiment of the present invention. [Figure 4] This flowchart shows an example of topic extraction processing in an embodiment of the present invention. [Figure 5] This is a flowchart showing an example of a classification process in an embodiment of the present invention. [Figure 6] This flowchart shows an example of the summarization process in an embodiment of the present invention. [Figure 7] This figure shows an example of document data in an embodiment of the present invention. [Figure 8] This figure shows an example of a viewpoint list in an embodiment of the present invention. [Figure 9] This figure shows an example of data explanation in an embodiment of the present invention. [Figure 10] This figure shows an example of prompt creation and output of the AI ​​for topic extraction in an embodiment of the present invention. [Figure 11] This figure shows an example of prompt creation and generation AI output for batch topic aggregation in an embodiment of the present invention. [Figure 12] This figure shows an example of a topic list in an embodiment of the present invention. [Figure 13] This figure shows an example of prompt generation and output of the AI ​​for perspective classification and topic classification in an embodiment of the present invention. [Figure 14] This figure shows an example of the result of assigning perspectives and topics to document data in an embodiment of the present invention. [Figure 15] This figure shows an example of the output of the AI ​​for prompting and generating viewpoint summaries and topic summaries in an embodiment of the present invention. [Figure 16] This figure shows an example of the output of the AI ​​for prompting and generating unclassified summaries in an embodiment of the present invention. [Figure 17]This is a diagram showing an example of prompt creation for batch summary aggregation and the output of a generative AI in an embodiment of the present invention. [Figure 18] This is a diagram showing an example of the output result of a summary in an embodiment of the present invention. [Figure 19] This is a diagram showing an example of a prompt template in a second embodiment of the present invention. [Figure 20] This is a diagram showing an example of the system configuration of a document additional information automatic registration device in a second embodiment of the present invention. [Figure 21] This is a flowchart showing an example of document additional information automatic registration processing in a second embodiment of the present invention. [Figure 22] This is a diagram showing an example of document data having a plurality of text items in a second embodiment of the present invention. [Figure 23] This is a diagram showing an example of prompt creation for document additional information generation and the output of a generative AI in a second embodiment of the present invention. [Figure 24] This is a diagram showing an example of an analysis report creation instruction including prompt template setting in a second embodiment of the present invention. [Figure 25] This is an example of a conceptual diagram of the processing flow in an embodiment of the present invention.

Modes for Carrying Out the Invention

[0011] FIG. 1 is a diagram showing an example of the system configuration of an analysis report generation device 100 in an embodiment of the present invention. <00​​​​​​​​​​The document data acquisition unit 102 can acquire all document data and perform filtered searches from the document DB 120 which stores the document data.

[0015] The analysis report generation unit 103 receives instructions for creating a report and document data, and performs report generation processing. In addition, the analysis report generation unit 103 can send a text-formatted prompt to the generation AI server 130 during the report generation process and receive a reply in text format.

[0016] The analysis report generation unit 103 can retrieve prompt templates stored in the prompt template DB 140. By substituting values ​​into the variables included in the prompt template, it can create prompts to send to the generation AI server 130.

[0017] Document DB120 stores document data. An example of document data is shown in Figure 7. The document data consists of multiple documents, and each document contains text fields and other fields. In the example in Figure 7, the document has a text field called "Body" and a field called "Posting Date and Time".

[0018] The generation AI server 130 operates general-purpose generation AI such as Large Language Models (LLMs). The generation AI server 130 may be located inside or outside the information processing device. If it is located outside, the analysis report generation unit 103 outputs a prompt to the generation AI, and processing is achieved by obtaining the processing results from the generation AI.

[0019] Furthermore, in this embodiment, an example using a general-purpose generative AI such as a large-scale language model as the generative AI server 130 is described, but it is not limited to this, and any AI, machine learning model, or other method capable of generating text may be used.

[0020] The prompt template DB140 stores prompt templates as shown in Figure 19. A prompt template consists of a variable list 1901 and a template string 1902, where the template string contains a fixed string portion and placeholders that are replaced by variables.

[0021] A single string is created by assigning a value to a variable and replacing placeholders with the variable's value. For example, in Figure 19, the parts enclosed in curly braces, such as {VAR01}, are placeholders and are replaced by the variable's value. Furthermore, variables have types. Variables of type string array can be assigned multiple strings, and within the template string, replacement is possible by repeating each element of the array. Parts such as {foreach text in VAR03} and {endfor} are control structures that represent repetition, and these are not included in the created string.

[0022] The prompt template acquired by the analysis report generation unit 103 differs for each processing step. The type of prompt template used for each step is included in the instructions for creating the analysis report. Furthermore, the type of value to be assigned to the variables in the prompt template is also included in the instructions. However, it is also possible to configure the system so that the selection of the prompt template is not included in the instructions. In that case, the system acquires the default prompt template, which is the setting value of the analysis report creation device 100, and uses the default value to assign to the variables.

[0023] Figure 2 is a block diagram showing an example of the hardware configuration of the client terminal 150 and the analysis report generation device 100 in an embodiment of the present invention.

[0024] As shown in Figure 2, the information processing device is connected via a system bus 204 to a CPU (Central Processing Unit) 201, ROM (Read Only Memory) 202, RAM (Random Access Memory) 203, input controller 205, video controller 206, memory controller 207, and communication I / F controller 208.

[0025] CPU201 provides comprehensive control over all devices and controllers connected to the system bus 204.

[0026] ROM202 or external memory211 holds the BIOS (Basic Input / Output System) and OS (Operating System), which are control programs executed by the CPU201, as well as computer-readable and executable programs and various necessary data (including data tables) for realizing this information processing method.

[0027] RAM203 functions as the main memory, work area, etc., of the CPU201. The CPU201 loads the necessary programs, etc., from ROM202 or external memory211 into RAM203, and then executes the loaded programs to perform various operations.

[0028] The input controller 205 controls input from input devices such as a keyboard 209 or a pointing device such as a mouse (not shown). If the input device is a touch panel, the user can give various instructions by pressing (touching with a finger, etc.) icons, cursors, or buttons displayed on the touch panel.

[0029] Furthermore, the touch panel may be a multi-touch screen or other touch panel capable of detecting the positions of multiple fingers touching it.

[0030] The video controller 206 controls the display to an external output device such as the display 210. The display may include the display of a notebook computer integrated with the main unit. The external output device is not limited to a display; for example, it may be a projector. Furthermore, for the aforementioned touch-enabled device, an input device is also provided.

[0031] The video controller 206 can control the video memory (VRAM) used for display control. It can utilize a portion of the RAM 203 as the video memory area, or it can provide a separate, dedicated video memory.

[0032] The memory controller 207 controls access to the external memory 211. The external memory can include an external storage device (hard disk), a flexible disk (FD), or a CompactFlash® memory connected to a PCMCIA card slot via an adapter, which stores boot programs, various applications, font data, user files, editing files, and other data.

[0033] The communication interface controller 209 connects to and communicates with external devices via a network and performs communication control processing over the network. For example, it can communicate using TCP / IP, telephone lines such as ISDN, and 3G mobile phone lines.

[0034] Furthermore, the CPU 201 enables display on the display 210 by, for example, performing the process of expanding (rasterizing) outline fonts into the display information area in RAM 203. The CPU 201 also enables user input via a mouse cursor (not shown) on the display 210.

[0035] First, we will explain the flow of the series of processes using Figure 25. Figure 25 shows an example of a conceptual diagram of the process flow in an embodiment of the present invention.

[0036] First, in Step 1, we collect VoC (Voice of the Customer) information such as SNS posts about the product and call logs of inquiries received by the call center. In this example, the first embodiment assumes the collection of SNS posts and reviews on evaluation sites, while the second embodiment assumes the collection of relatively long texts such as call logs. An example of document data of collected posting information is shown in Figure 7, and an example of document data of a call log with multiple text items is shown in Figure 22. Further details will be described later.

[0037] In this embodiment, SNS posts and call logs were assumed to be the subjects of analysis, but other information such as review sites may also be collected, or any other information such as customer opinions, requests, impressions, complaints, and survey results may be used. Furthermore, the subjects of analysis are not limited to customer opinions, but may also include the results of surveys in which employees evaluate products, or feedback after events have been held.

[0038] In Step 2, topics are extracted from the VoC information collected in Step 1, referencing pre-defined perspectives and their definitions, and a topic list is generated. By providing the generating AI with a document containing the VoC information, a perspective list with items of perspectives and their definitions as shown in Figure 8, a data description as shown in Figure 9, and a prompt including instructions to extract topics, a topic list is extracted for each perspective. Finally, a topic list with topics extracted for each perspective is generated as shown in Figure 12. The specific processing details will be described later in Figure 4.

[0039] Furthermore, the topic extraction process described above is not limited to simply extracting topical strings from the document (information) being analyzed, but also includes the process of generating strings (sentences) that indicate topics interpreted from the content of the document being analyzed.

[0040] In other words, Step 2 is a step that demonstrates an example of the process of obtaining information to be analyzed, pre-defined perspectives for analyzing the information, and topics extracted based on instructions to extract topics from the information based on those perspectives.

[0041] Step 3 involves classifying the collected VoC information into perspectives and topics. The specific processing details will be described later in Figure 5. By providing the generating AI with documents containing VoC information, the data description shown in Figure 9, the topic list generated in Step 2, and instructions for extracting topics, topics are assigned to documents containing VoC information. By assigning perspectives corresponding to the topics, the documents containing VoC information are ultimately classified into perspectives and topics, as shown in Figure 14. The specific processing details will be described later in Figure 5.

[0042] In other words, Step 3 is a step that demonstrates an example of a process that outputs an instruction to classify the information to be analyzed into the topic from which it was obtained, and then obtains the classification result based on that instruction to classify the information to be analyzed into the topic from which it was obtained.

[0043] Step 4 involves generating an analysis report that summarizes the collected VoC information according to the user's desired perspective or topic. The AI ​​generates the summary by providing it with the specific perspective or topic to be summarized, the documents to be summarized, the data description shown in Figure 9, and instructions for generating the summary. The specific details of the process will be described later in Figure 6.

[0044] In other words, Step 4 is a step that shows an example of a process that outputs instructions to generate a summary of the information to be analyzed based on the acquired classification results, and then retrieves the generated summary information.

[0045] In other words, Steps 2 to 4 are steps that demonstrate an example of a process that outputs the information to be analyzed, a pre-set perspective for analyzing the information, instructions to extract topics from the information based on the perspective, instructions to classify the information into topics, and instructions to generate a summary of the information to be analyzed based on the topics, and then retrieves the summary information generated based on the instructions.

[0046] By performing the processes described in Steps 1 to 4 above, it becomes possible to generate an analysis report that summarizes the collected VoC information by perspective or topic. Furthermore, as explained in Step 2, in this embodiment, topics are extracted by referring to a predefined list of perspectives, so that VoC information can be classified based on the items that the user wants to analyze, and as a result, the user's desired analysis report can be generated.

[0047] Furthermore, in this process, topic extraction, classification, and summary generation were performed step by step from Step 2 to Step 4. By dividing the processing into each step, the instructions to the generating AI become clearer, and an improvement in output accuracy is expected.

[0048] Furthermore, depending on the generation AI used, there may be limitations on the number of tokens that can be input, making it impossible to input all of the vast amount of VoC information at once. In such cases, for example, in topic extraction processing, it is necessary to divide the entire document, generate topics for each document batch, and then aggregate the generated topics. By processing in separate steps, it becomes possible to process each document batch even when there are limitations on the number of tokens, making it possible to handle vast amounts of VoC information.

[0049] Alternatively, instead of performing Steps 2 through 4 sequentially, the topic extraction, classification, and summary generation processes can be performed simultaneously with a single prompt. In this case, for example, by providing the generating AI with a prompt containing the collected VoC information, a predefined list of perspectives, and a description of the data, along with instructions such as "Output the main topics contained in this text data. Assign the extracted topics to one of the following perspectives," "Classify which topic each text in the text data belongs to," and "Summarize and output each topic," a summary result for each topic can be obtained. Alternatively, instructions can be given to output both the topic list or the classification results with topics assigned to the VoC information.

[0050] (First Embodiment) Next, the processes performed by the analysis report generation device 100 in the first embodiment of the present invention will be explained using the flowcharts in Figures 3 to 6.

[0051] The flowchart in Figure 3 shows the flowchart of the analysis report generation process performed by the analysis report generation device 100.

[0052] In step S301, the user input receiving unit 101 receives an instruction to create an analysis report as input from the client terminal 150.

[0053] The instructions for creating the analysis report include a list of perspectives, a description of the data, and filtering criteria for the document data. It is also possible to include prompt template settings in the analysis report creation instructions. This configuration will be described later.

[0054] A list of viewpoints consists of multiple viewpoints, each viewpoint comprising a viewpoint name string and a definition string. Figure 8 shows an example of a viewpoint list. There is a viewpoint whose name is "Function" and whose definition is "Opinions on functions such as printing, scanning, and copying." Also, the viewpoint named "Price" has an empty definition string. Setting a definition makes the content of a viewpoint clear and makes it possible to generate topics that are relevant to the content of that viewpoint, but it is also possible to generate topics using only the viewpoint's keywords.

[0055] The perspective names are items that can serve as keywords for analyzing VoC information, such as "Function," "Price," and "Defects." The definition statements are keywords that specifically describe the content of the perspective. Users can pre-set the perspectives they want to analyze and their definitions, as shown in Figure 8. This makes it possible to extract topics that take into account the perspectives and definitions that the user wants to analyze.

[0056] Figure 9 shows an example of a data description in an embodiment of the invention. The data description is a single string of characters and is an explanatory text that describes the information contained in the document data. The data description is set to "SNS post data related to printer product A".

[0057] In addition, in cases like the second embodiment, a string such as "call log from a call center regarding printer product A" may be added as a description of the data. By providing the data description to the generation AI, it becomes possible to generate, classify, and summarize topics that take into account the background information that the data is "SNS post data regarding printer product A."

[0058] In step S302, the document data acquisition unit 102 retrieves a list of documents from the document database 120 that match the filtering conditions. For example, if a user-inputted filtering condition such as "post date and time is 2025 / 1 / 1 or later" is specified, the document database 120 will be filtered to retrieve a list of documents whose post date and time is 2025 / 1 / 1 or later. Note that the filtering conditions are not limited to this; filtering may also be performed using conditions related to user attributes, specific keywords, product names, or other information.

[0059] From step S303 onward, the analysis report generation unit 103 performs the processing.

[0060] Step S303 performs topic extraction. Details of the topic extraction process will be described later in Figure 4. In this step, topics contained in the text content of the document list are extracted, resulting in a topic list like the one shown in Figure 12. The topic list consists of multiple topics, each topic relating to one viewpoint and including a topic name.

[0061] Figure 12 shows an example of the results of extracting topics based on the document list in Figure 7 and the perspective list in Figure 8. For the "Functionality" perspective, topics titled "Scanning function is well-received" and "Opinions on smartphone operation" have been extracted. Similarly, topics are extracted for each perspective, such as "Price" and "Bugs." Each perspective may have one or more topics, and if no suitable topics are found for a particular perspective, it may be left as "no topics."

[0062] Step S304 performs a classification process related to topics. Details of the classification process will be described later in Figure 5. In this step, zero or more topics from the topic list are assigned to each document included in the document list. The number of topics can be any number. As a result, the documents will be assigned topics as shown in Figure 14. Furthermore, documents that have been assigned topics will also be assigned perspectives related to those topics, based on the topic list in Figure 12.

[0063] Document 1401 in Figure 14 has the topic "Scanning function is well-received" assigned to it as a topic name. When a topic is assigned, the perspective of "function," which is related to that topic, is also assigned to the document.

[0064] Document 1402 has the topics "Scanning function is well-received" and "Opinion that the price is high" assigned to it. Thus, multiple topics can be assigned to a single document. The former topic relates to the perspective "function," and the latter relates to the perspective "price," so two perspectives, "function" and "price," are assigned to the document.

[0065] Document 1403 is assigned two topics related to the perspective "price": "opinion that the price is high" and "sale information." Since both of the aforementioned topics relate to "price," it is assigned one perspective: "price."

[0066] In the processing of step S304, there may be cases where no topic is assigned to the document, as shown in document 1404. In the following step S305, a perspective is assigned to such a document.

[0067] In step S305, the search is narrowed down to the list of documents that have not been assigned a topic. A classification process related to perspectives is then performed on these documents. Details of the classification process will be described later in Figure 5. In this step, zero or more perspectives from the perspective list are assigned to each of the narrowed-down documents. The number of perspectives can be any number. As a result, even documents that were not assigned a topic will now have perspectives assigned to them, as shown in Figure 14.

[0068] The processing in step S305 makes it possible to assign perspectives to documents on topics that were not extracted in S303. One case in which a topic is not extracted in S303 is when there are few documents on a similar topic.

[0069] For example, in document 1404 in Figure 14, the perspective "defect" is assigned. Since the topic extraction process extracts only the main text content in the document list, text content that appears infrequently is not treated as a topic. In the topic list in Figure 12, the only topic related to "defect" is "print quality issues," so the text "My new PC isn't recognizing printer product A, and I'm having trouble" is not assigned the topic of "defect." On the other hand, considering the content, document 1404 is a document that should be classified under the perspective "defect," so this perspective is assigned in this step. This makes it possible to classify even minor opinions that were not extracted as topics according to their respective perspectives.

[0070] Step S306 initiates the repetition of steps S307 and S308 zero or more times. The number of repetitions is pre-set in the analysis report generation unit 103 as the number of times the classification is reconfirmed. If the number of repetitions is zero, steps S307 and S308 are not performed.

[0071] Step S307 narrows down the list to documents that have not yet been assigned a viewpoint or topic. These documents are then subjected to a topic-related classification process (similar to the process in step S304).

[0072] Step S308 narrows down the list to documents that have not yet been assigned a viewpoint or topic. For these documents, a classification process related to viewpoints (similar to the process in step S305) is performed.

[0073] The above iterative process is performed to re-verify the classification of documents that failed to be classified by the generating AI. Even if a document was not assigned any viewpoint or topic by the generating AI, it may be desirable for a human to assign some viewpoint or topic upon review. For such documents, the goal is to perform the generating AI classification process several more times in the hope that viewpoints and topics will be assigned.

[0074] In other words, steps S307 to S308 are steps that show an example of a process that outputs information that was not classified as a topic, the extracted topics, and instructions to classify the information that was not classified as a topic into a topic, and obtains the classification result based on the instructions.

[0075] Because the classification process by the generative AI can change even with subtle input differences, such as rearranging the order of text data in the prompts, multiple classification attempts are effective. Through the above process, even for documents where the generative AI classification process was unsuccessful in steps S304 and S305, it becomes possible to assign perspectives and topics.

[0076] Furthermore, by re-examining only the documents for which no perspective or topic was assigned, the amount of text input to the generation AI can be reduced compared to the entire document list. This allows for efficient handling of failures where perspectives or topics are not assigned, while saving on AI usage fees and token counts.

[0077] During the repeated processing of steps S307 and S308, if any viewpoints or topics are assigned even once, they will be reflected in the final document assignment. If no viewpoints or topics are assigned even once, the final viewpoints and topics will be blank, as shown in document 1405 in Figure 14.

[0078] In step S309, the process of repeating step S310, which will be described later, for each viewpoint is initiated.

[0079] In step S310, summarization processing is performed on the relevant viewpoint. Details of the summarization processing will be described later in Figure 6. In this step, the list of documents to which the viewpoint to be summarized is assigned is narrowed down, and the generating AI performs summarization of multiple documents. As a result, as shown in summary 1801 in Figure 18, a text summarized on the entire viewpoint of the viewpoint to be summarized is output. Note that in Figure 18, the summarization results are saved in a tabular list format, but when displaying to the user, it may be possible to display summaries of viewpoints or topics specified by the user.

[0080] Summary 1801 in Figure 18 shows an example of outputting a summary of the entire "Function" perspective, targeting the body text of documents with the "Function" perspective from the document list in Figure 14. The summary text reads: "This summarizes posts about the functions of printer product A. - Many positive posts are found regarding the scanning function..." Furthermore, it also outputs information indicating that there are 1029 documents in the document list that have the "Function" perspective.

[0081] In step S311, the process of repeating step S312 for each topic begins.

[0082] In step S312, summarization processing is performed for each topic. Details of the summarization processing will be described later in Figure 6. In this step, the list of documents to be summarized is narrowed down to include the topic to be summarized, and the generating AI performs summarization of multiple documents. As a result, the summary text for the topic is output, as shown in summary 1802 in Figure 18.

[0083] In other words, step S312 is a step that demonstrates an example of a process for generating a summary of information classified under a topic for each topic.

[0084] Summary 1802 in Figure 18 shows an example of outputting a summary related to the topic "Scanning function is popular" from the document list in Figure 14, focusing on the text of documents that have the topic "Scanning function is popular." The summary text reads, "Regarding the topic of the scanning function being popular, there are posts like the following: - You can scan magazines cleanly..." Furthermore, the number of documents that have been assigned the "Scanning function is popular" trait is also output.

[0085] Step S313 performs summarization of documents that lack a viewpoint and topic. Details of the summarization process will be described later in Figure 6. We will refer to such summaries as summaries of unclassified documents. In this step, the list is narrowed down to documents that do not have a viewpoint and topic assigned, and the generating AI performs summaries of multiple documents. As a result, as shown in summary 1803 in Figure 18, the text of the unclassified summary is set as a summary with <no viewpoint><no topic>.

[0086] In other words, step S313 is a step that demonstrates an example of a process for generating a summary of information that was not classified into a topic.

[0087] Summary 1803 in Figure 18 shows an example of outputting summaries for documents in the document list in Figure 14 where the perspective and topic fields are blank, specifically for documents with "<No Perspective>" and "<No Topic>". The summary text reads, "The text without a perspective or topic contains various information about printer product A. Some of the content is shown below..." Furthermore, the number of documents without a perspective or topic is also output.

[0088] The summary generated in this step allows the analyst to identify noteworthy content within the document list, even if it doesn't fall into a predefined category.

[0089] In step S314, the topic list shown in Figure 12, the document list with assigned perspectives and topics shown in Figure 14, and the summary list shown in Figure 18 are combined into a single file to generate the final analysis report file. For example, these lists can be directly included in a file format for spreadsheet software. The format is not limited to this; any format that allows the analyst to confirm the perspectives and topics of the generated summaries is acceptable.

[0090] The resulting analysis report analyzes the topic according to predefined perspectives.

[0091] In this embodiment, a method was described in which topic extraction is performed in step S303, classification processing in steps S304 to S308, and summary generation in steps S309 to S313 in a stepwise manner. However, it is also possible to perform topic extraction, classification processing, and summary generation processing all at once with a single prompt. In that case, the generating AI is provided with the document, the list of viewpoints in Figure 8, and the data description in Figure 9. In addition, by instructing the prompt to "assign topics to viewpoints and extract them," "classify which of the extracted topics the document corresponds to," and "generate a summary for each topic or viewpoint," it is possible to obtain the summary result without outputting the topic list or classification results.

[0092] The flowchart in Figure 4 shows the topic extraction process performed by the analysis report generation unit 103. It is called from step S303 of the analysis report generation process.

[0093] The input consists of a document list, a list of viewpoints, and a description of the data.

[0094] The output is a topic list as shown in Figure 12. Each topic includes a topic name and is associated with one of the viewpoints in the viewpoint list.

[0095] In step S401, the document list is divided into document batches consisting of a fixed number of documents. The fixed number of documents is pre-set in the analysis report generation unit 103.

[0096] For example, if you have 1000 documents and want to divide them into batches of 20, they will be divided into 50 document batches. To mitigate bias due to the order of the document data, the document list may be rearranged in a random order before dividing it into document batches. Alternatively, the documents may be divided so that the number of characters per batch is equal. Another approach is to divide documents into batches when the number of characters or words is large, and process other documents in one go without batch division.

[0097] In other words, step S401 is a step that demonstrates an example of a process that divides information based on the number of characters or words contained in the information.

[0098] There may be a size limit on the prompt sent to the generation AI server 130. By performing the splitting process in step S401, it is possible to adjust the prompt so that it fits within the size limit.

[0099] In step S402, the batch topic list, which is a list of topic lists, is initialized with an empty list.

[0100] In step S403, the repetition of steps S404 to S406 is initiated for each document batch.

[0101] Step S404 creates prompts to extract topics from the document batch and the viewpoint list.

[0102] In step S405, the prompt created in step S404 is sent to the generation AI, and the generated text is retrieved.

[0103] In step S406, topics are extracted from the generated text and added to a list of batch topics.

[0104] Steps S404, S405, and S406 are processes for extracting topics from a document batch. This process will be explained using the example in Figure 10.

[0105] In Figure 10, the input consists of a document batch divided in step S401, the viewpoint list shown in Figure 8, and the data description shown in Figure 9. A prompt template for topic extraction is obtained from prompt template DB140, and prompt 1001, a string to be sent to the generating AI, is created by setting the input values ​​into variables.

[0106] Prompt 1001 includes instructions and output format definitions that provide a definition for the topic extraction task. The instructions and output format can include the following: ● Extract the main topics contained in the text data. ● Each extracted topic should be associated with one of the following perspectives. ● The output format must be one that allows for text analysis to extract topics and associate them with relevant viewpoints. Furthermore, prompt 1001 includes a list of viewpoints. The list of viewpoints contains the viewpoint names and definitions for all viewpoints from the input viewpoint list.

[0107] The prompt also includes a description of the data. The description of the entered data should be entered here exactly as it is.

[0108] The prompt also includes text data. This text data is created by taking the values ​​of text items from each document in the document batch and listing them. If the document schema is in the format shown in Figure 7, the string "Body" is used as the text.

[0109] For example, if you use a portion of the document list shown in Figure 7 as a document batch, and input the viewpoint list shown in Figure 8 and the item "SNS post data regarding printer product A" from the data description in Figure 9, prompt 1001 will be created.

[0110] The prompt 1001 created using the method described above is sent to the generating AI server 130 to extract the topic.

[0111] In Figure 10, the generating AI outputs text separated by line breaks, containing "Features: Scanning function is well-received, smartphone operation is comfortable," "Price: Opinions that the price is high," and "Bugs: ". When analyzing this output format, the part before the colon (:) on each line of text output is treated as the perspective name. The part after the colon (:) is treated as a list of topic names, and the strings separated by commas are used as topic names. As a result, for example, for the perspective name "Features," two topic names can be obtained: "Scanning function is well-received" and "Smartphone operation is comfortable." For the perspective name "Bugs," it can be analyzed that no topic has been extracted.

[0112] In the example using prompt 1001, it can be inferred that the generating AI extracted the abstract topic name "Scanning function is well-received" from the texts "I tried scanning a magazine with printer product A and it came out beautifully" and "Printer product A is too expensive for me to buy, but its scanning function is attractive." Furthermore, it can be inferred that the generating AI determined that this topic relates to the "function" aspect of the three perspectives: "function," "price," and "problems."

[0113] The analysis report generation unit 103 analyzes the text output of the generating AI and adds the analysis results as a batch topic to the list of batch topics.

[0114] In other words, steps S404 to S406 are steps that demonstrate an example of a process in which multiple pieces of information to be analyzed are divided into fixed-number units and output.

[0115] Step S407 creates a prompt to obtain the final topic list from the list of batch topics.

[0116] In step S408, the prompt created in step S407 is sent to the generation AI, and the generated text is retrieved.

[0117] In step S409, a list of topics is extracted from the generated text.

[0118] Steps S407-S409 described above are processes for aggregating the batch topic list. When topics are extracted for each document batch, there may be overlaps or variations in terminology among those topics. In addition, topics that can be aggregated if abstracted but have different content may be extracted, or topics that are infrequent in the overall document list may be extracted. To appropriately aggregate the extracted topics and make them easy for the analyst to understand, the generation AI performs the task of aggregating the topics for each document batch.

[0119] The aggregation process in steps S407 to S409 will be explained using Figure 11.

[0120] In Figure 11, the input consists of a list of batch topics, a list of perspectives shown in Figure 8, and a description of the data shown in Figure 9. A prompt template for topic aggregation is obtained from the prompt template DB140, and the input values ​​are set to variables to create prompt 1101, which is the string to be sent to the generating AI.

[0121] Prompt 1101 includes instructions and output format definitions that provide a definition for the topic aggregation task. The instructions and output format can include the following: ● The topic has been extracted for each document batch into which the document list has been divided. ● Consolidate topics from each document batch. ● Specify the number of topics to output (for example, 5 or fewer topics per perspective). ● The output format must be one that allows for text analysis to extract topics and associate them with relevant viewpoints. Furthermore, prompt 1101 includes a list of viewpoints. The list of viewpoints contains the viewpoint names and definitions for all viewpoints from the input viewpoint list.

[0122] Furthermore, prompt 1101 includes a description of the data. The description of the entered data should be entered here exactly as it was entered.

[0123] The prompt also includes a list of batch topics. This list contains the input batch topics, separated by delimiters and presented in the same format as the output.

[0124] For example, if you input the viewpoint list shown in Figure 8 and the data description shown in Figure 9, and then input multiple batch topics like the output in Figure 10, the prompt in Figure 11 will be created.

[0125] The prompts created using the method described above are sent to the generation AI server 130 to aggregate the topics.

[0126] In Figure 11, the output is divided into three lines separated by line breaks: "Features: Scan function is well-received, Feedback on smartphone operation," "Price: Feedback on high price, Sales information," and "Problems: Print quality issues."

[0127] In this example, it can be inferred that the topic "Function: Opinions on smartphone operation" was generated by the AI ​​by abstracting the topics "Function: Smartphone operation is comfortable" from BATCH001 and "Function: The UI for smartphone operation is difficult to understand" from BATCH002.

[0128] The aggregation method described here involves extracting topics in parallel for each document batch, and then aggregating all batch topics at once.

[0129] In addition, it is also possible to implement a method of sequential aggregation by creating dependencies between topic extraction processes. In sequential aggregation, the topic extraction prompt for each document batch is modified to include the topic list output in the previous document batch and an instruction to not output the previously outputted topic. However, this is not added to the prompt of the first document batch.

[0130] By performing either batch or sequential aggregation, efficient topic extraction that considers all documents becomes possible.

[0131] The analysis report generation unit 103 analyzes the text output of the generating AI and extracts topics. The analysis is performed in the same manner as in Figure 10.

[0132] In step S410, the topic list obtained in step S409 is returned, and finally a topic list categorized by viewpoint, as shown in Figure 12, is obtained.

[0133] The flowchart in Figure 5 shows the classification process performed by the analysis report generation unit 103. It is called from steps S304, S305, S307, and S308 of the analysis report generation process.

[0134] The input consists of the document list shown in Figure 7, the data description shown in Figure 9, and either the viewpoint list shown in Figure 8 or the topic list returned in step S410. In other words, when classifying by topic as in steps S304 and S307, the topic list returned in step S410 is given to the generating AI, and when classifying by viewpoint as in steps S305 and S308, the viewpoint list shown in Figure 8 is given to the generating AI.

[0135] As an output result, each document in the document list is assigned a perspective or topic, as shown in Figure 14.

[0136] In step S501, one of the perspective list or topic list given in the input is set as the classification target and stored in main memory.

[0137] In step S502, the document list is divided into document batches consisting of a fixed number of documents. The fixed number of documents is pre-set in the analysis report generation unit 103.

[0138] The document batching is performed in the same manner as described in the flowchart in Figure 4.

[0139] In step S503, the repeated processing of steps S504 to S506 is initiated for each document batch.

[0140] Step S504 creates a prompt for text classification from the document batch and the items to be classified.

[0141] In step S505, the prompt created in step S504 is sent to the generation AI, and the generated text is retrieved.

[0142] In step S506, the generated text is analyzed, and perspectives and topics are assigned to each document in the batch.

[0143] Steps S504 to S506 are processes for classifying documents within a document batch into categories. This process will be explained using Figure 13. Figure 13 shows an example where a topic list is the category. When performing steps S305 and S308, it is possible to assign perspectives in a similar manner to a perspective list as the category, so only the assignment of topics will be explained here.

[0144] In Figure 13, the input consists of a document batch, the topic list returned in step S410, and the data description shown in Figure 9. A prompt template for topic classification is obtained from the prompt template DB140, and the input values ​​are set to variables to create a prompt, which is a string to be sent to the generating AI.

[0145] To associate multiple documents with multiple topics, an ID is created for each document and topic immediately before creating the prompt. The IDs can be sequential numbers or random strings. These IDs are temporary IDs for each classification prompt, and a different ID may be created for each prompt.

[0146] Prompt 1301 includes instructions and output format definitions that provide a definition of the classification task. The instructions and output format may include the following: ● Classifying each text in the text data into topics. ● Classification involves assigning zero or more topics to each text. ● The output format must be one that allows for text analysis and the matching of topics with documents. Prompt 1301 also includes a topic list. The topic list contains the topic names and topic IDs of all topics from the input topic list. The perspective name is optional.

[0147] Prompt 1301 also includes a description of the data. The description of the entered data should be entered here exactly as it was entered.

[0148] Furthermore, prompt 1301 contains text data. Each text item in the text data is an extract of the text item value from each document in the document batch. In addition, each text item also includes the document ID of the original document.

[0149] For example, if you use a portion of the document list in Figure 7 as a document batch, and input the topic list in Figure 12 and the data description in Figure 9, the prompt in Figure 13 will be created.

[0150] The prompt 1301 created in this way is sent to the generating AI server 130 to assign a topic.

[0151] In Figure 13, the text output is separated by line breaks and shows "TOPIC001: TEXT101, TEXT102", "TOPIC002: ", and "TOPIC003: TEXT101".

[0152] When analyzing this output format, the part before the colon (:) in each line of text output is treated as the topic ID. The part after the colon (:) is treated as a list of document IDs, and each string separated by a comma (,) is treated as a document ID. The analysis report generation unit 103 maintains the correspondence between documents and document IDs, and the correspondence between topics and topic IDs, so it can determine the correspondence between topics and documents.

[0153] This section explains an example of parsing the line "TOPIC001: TEXT101, TEXT102".

[0154] Topic ID "TOPIC001" is a topic with the title "Scanning function is well-received". Text ID "TEXT101" is a document with the text "Printer product A is too expensive for me to buy, but the scanning function is attractive". Text ID "TEXT102" is a document with the text "I tried scanning a magazine with printer product A and it came out beautifully". Therefore, the topic "Scanning function is well-received" is assigned to these two documents.

[0155] In step S507, the results of assigning topics and perspectives to each document batch are combined as is, and a list of documents to which perspectives and topics have already been assigned is returned.

[0156] The flowchart in Figure 6 shows the summarization process performed by the analysis report generation unit 103. It is called from steps S310, S312, and S313 of the analysis report generation process.

[0157] This summarization process has three applications: "summarizing a list of documents with a specific perspective," "summarizing a list of documents with a specific topic," and "summarizing a list of documents without a specific perspective or topic." The third type of summarization will be referred to as the summarization of unclassified documents from now on.

[0158] The input consists of the document list shown in Figure 7 and the data description shown in Figure 9. Furthermore, when summarizing a specific viewpoint or topic, the viewpoint or topic to be summarized is also provided. On the other hand, when summarizing an unclassified document, a flag indicating it is unclassified, along with a list of all viewpoints and topics, is provided.

[0159] Figure 18 shows an example of the output results of the summaries. The summary by viewpoint 1801 is the summary within the <entire viewpoint> entry for that viewpoint. The summary by topic 1802 is the summary within the entry for viewpoints related to that topic. The summary for unclassified documents 1803 is the summary of the <no viewpoint> and <no topic> entries.

[0160] Step S601 sets the target for summarization. If a specific viewpoint or topic is provided in the input, that viewpoint or topic will be the target for summarization. The user can also configure the system to summarize by viewpoint or topic. If a flag indicating that it is a summary of unclassified documents is provided, documents that have not been assigned a classification will be the target for summarization.

[0161] In step S602, the document list is divided into document batches consisting of a fixed number of documents. The fixed number of documents is pre-set in the analysis report generation unit 103.

[0162] The document batching is performed using the same method as described in the flowchart in Figure 4.

[0163] In step S603, the list of batch summaries, which is a list of strings, is initialized with an empty list.

[0164] In step S604, the repetition of steps S605 to S607 is initiated for each document batch.

[0165] Step S605 creates a prompt to generate a summary from the document batch and the items to be summarized.

[0166] In step S606, the prompt created in step S605 is sent to the generation AI, and the generated text is retrieved.

[0167] In step S607, a summary is extracted from the generated text and added to a list of batch summaries.

[0168] Steps S605 to S607 involve summarizing a document batch as input. This process will be explained using Figures 15 and 16.

[0169] Figure 15 shows an example of the output of the AI ​​for prompt generation and generation of perspective summaries and topic summaries.

[0170] Figure 15 shows the operation when summarizing from a specific viewpoint or topic, and Figure 16 shows the operation when summarizing without classification. Although Figure 15 specifically shows an example where a topic is the target of summarization, a similar method can be used to generate summaries when viewpoints are the target, so only the case of summarizing by topic will be explained.

[0171] In Figure 15, a specific viewpoint or topic, a document batch, and a description of the data shown in Figure 9 are given as input. A prompt template for topic summarization is obtained from prompt template DB140, and prompt 1501, a string to be sent to the generating AI, is created by setting the input values ​​to variables.

[0172] Prompt 1501 includes instructions and output format definitions that provide a definition for the summarization task. A template can also be set up to include the relevant perspective / topic in the instructions. The instructions and output format can include the following: ● Topic to be summarized ● Summary output format and character limit Furthermore, prompt 1501 includes the name of the topic to be summarized. The name of the topic to be summarized should be entered here exactly as it is.

[0173] Prompt 1501 also includes a description of the data. The description of the entered data should be entered here exactly as it is.

[0174] Furthermore, prompt 1501 contains text data. Each piece of text data is an extract of the text item value from each document in the document batch.

[0175] As an example, for the topic "Scanning function is well-received," we will use a document batch from the document list in Figure 14 that has the tag "Scanning function is well-received." In this case, the selected documents are "I tried scanning a magazine with printer product A and it came out beautifully" and "Printer product A is too expensive for me to buy, but the scanning function is attractive." Furthermore, we will input the data description shown in Figure 9. As a result, prompt 1501 will be created.

[0176] The prompt 1501 created in this way is sent to the generation AI server 130 to perform text generation.

[0177] Figure 15 shows that the AI-generated text output reads, "Regarding the topic of the popular scanning function, there are posts like the following: - It can scan magazines beautifully..." This text output can be used directly as a batch summary.

[0178] Figure 16 shows an example of prompt generation and output from an AI for unclassified summaries.

[0179] In the unclassified summary shown in Figure 16, the input consists of a list of all perspectives, a list of topics, a description of the data shown in Figure 9, and a batch of documents that do not have topics or perspectives assigned. The prompt template for the unclassified summary is obtained from the prompt template DB140, and prompt 1601, which is the string to be sent to the generating AI, is created by setting the input values ​​to variables.

[0180] Prompt 1601 includes instructions and output format definitions that provide a definition for the summarization task. A template can also be set up to include a list of viewpoints and a list of topics in the instructions. The instructions and output format can include the following: ● Information indicating that the text data is unrelated to any of the viewpoints in the viewpoint list and unrelated to any of the topics in the topic list. ● Summary output format and character limit ● An instruction to output information about the content of documents that have not been assigned a viewpoint or topic. ● Instructions to suggest additional topics or perspectives. Including information that is unrelated to any particular viewpoint or topic can be expected to allow analysts to include unexpected information in their summaries that they had not anticipated beforehand.

[0181] Prompt 1601 also includes a description of the data. The description of the entered data should be entered here exactly as it is.

[0182] Furthermore, prompt 1601 contains text data. Each piece of text data is an extract of the text item value from each document in the document batch.

[0183] As an example, we will use a portion of the documents in the document list in Figure 14 that do not have a topic and perspective assigned to them as a document batch. Here, the documents selected are "I'm undecided whether to buy printer product A from company X or printer product B from company Y" and "I use printer product A at home." Furthermore, we will input the data description shown in Figure 9. As a result, prompt 1601 in Figure 16 will be created.

[0184] The prompt 1601 created in this way is sent to the generation AI server 130 to perform text generation.

[0185] Figure 16 shows that the AI-generated text output reads, "The text without a specific viewpoint or topic contains various information about printer product A. Some of the content is shown below: - Posts comparing it with other printer products for purchase consideration..." This text output can be used directly as a batch summary.

[0186] Step S608 creates a prompt to obtain the final summary from the list of batch summaries.

[0187] In step S609, the prompt created in step S608 is sent to the generation AI, and the generated text is treated as a summary.

[0188] Steps S608 and S609 are processes for aggregating batch summaries.

[0189] When summarizing topics for each document batch, there may be overlaps in the summaries, or the summaries may include content that is infrequent in the overall document list. To appropriately aggregate the summaries and make them easy for analysts to understand, a generative AI is used to aggregate the summaries from each document batch.

[0190] Figure 17 shows an example of prompt generation and output from the batch summary aggregation AI.

[0191] Figure 17 shows an example where the subject of the summary is a topic, but since summarizing viewpoints and uncategorized summaries can be handled similarly, we will only explain the case of topics.

[0192] In Figure 17, the input consists of a list of batch summaries, a description of the data, and the data to be summarized.

[0193] The prompt template for summarizing and aggregating is obtained from prompt template DB140, and prompt 1701, which is the string to be sent to the generating AI, is created by setting the input values ​​to variables.

[0194] Prompt 1701 includes instructions and output format definitions that provide a task definition for summarization. A template can also be set up that includes the topic to be summarized in the instructions. The instructions and output format can include the following: ● A summary is generated for each document batch into which the document list has been divided. ● Aggregate summaries for each document batch. ● Format and word count of the output summary Furthermore, prompt 1701 contains the topic to be summarized. The topic name entered here should be written exactly as it is.

[0195] Furthermore, prompt 1701 includes a description of the data. The description of the entered data should be entered here exactly as it was entered.

[0196] Prompt 1701 also includes a list of batch summaries. The entered list of batch summaries should be listed here, separated by delimiters.

[0197] For example, if you input a description of the data shown in Figure 9 for the topic "Scan function is well-received" and then input multiple batch summaries like the output in Figure 15, prompt 1701 in Figure 17 will be created.

[0198] The prompts thus created are sent to the generation AI server 130 for summary collection.

[0199] Figure 17 shows a summarized output regarding the topic of the scanning function being well-received, which includes comments such as: - It can scan magazines cleanly - The scanning speed is fast...

[0200] In this example, it can be inferred that the generating AI merged the content from the first BATCH001, "It can scan magazines cleanly," with the content from BATCH002, "There are opinions that the scanning speed is fast." This text output can be used directly as a summary of the topic.

[0201] The aggregation method described here involves summarizing each document batch in parallel without dependencies, and then aggregating all batch summaries at once.

[0202] In addition, it is also possible to implement a sequential aggregation method by creating dependencies between summarization processes. In sequential aggregation, the summary prompt for each document batch is modified to include the text of the summary output from the previous document batch and an instruction to not output the previously outputted summary. However, this is not added to the prompt for the first document batch.

[0203] By aggregating the content of summaries either all at once or sequentially, it becomes possible to generate summaries that take all documents into consideration.

[0204] In step S610, the summary obtained through aggregation is returned, and the process ends.

[0205] As a result, it becomes possible to generate summaries for each topic based on different perspectives, providing a mechanism that allows for an appropriate understanding of the content of the information being analyzed.

[0206] (Second Embodiment) In the first embodiment, a method for performing VoC analysis on relatively short texts, such as SNS posts related to a product, was described. In the second embodiment, which will be described later, a method for performing VoC analysis on relatively long texts, such as conversational texts like call logs to a call center, will be described.

[0207] In the second embodiment, an example is described using a document database as shown in Figure 22, in which documents of call logs to a call center are registered. This document database includes text fields such as "Document ID 2201", "Inquiry Content 2202", "Response Content 2203", and "Title 2204".

[0208] For example, the fields "Inquiry Details 2202" and "Response Details 2203" are expected to be automatically registered with text transcribed from the audio of the call log sent to the call center. Therefore, the registered text may sometimes be quite long.

[0209] Now, let's consider the case where a pay-per-use service (a system that charges based on the amount of data or the number of interactions with a text generation system such as a text generation AI) is used. In this case, a fee is incurred each time a request is sent. In particular, with text generation AIs, a fee is often incurred for each word (token) input (or output by the generation AI). In other words, there was a problem in that the fee increased as the number of input characters increased, and the processing load also increased.

[0210] Therefore, while it is possible to use long sentences like those registered in inquiry content 2202 and response content 2203 directly in the topic extraction process in Figure 4 and the classification process in Figure 5, for longer sentences, it is desirable to reduce the cost and processing load.

[0211] Therefore, in the second embodiment, we propose a method in which a generating AI generates a short sentence (for example, a title 2204) that explains the content of an item from an item containing a long sentence such as the inquiry content 2202 and the response content 2203, and then uses the generated short sentence to efficiently perform topic extraction processing and classification processing.

[0212] Figure 20 shows an example of the system configuration of the document addition information automatic registration device 2000 in an embodiment of the present invention.

[0213] The document addition information automatic registration device 2000 can retrieve documents from the document DB 120 and update the documents in the document DB 120. It can also send text-formatted prompts to the generation AI server 130 and receive replies in text format. Furthermore, the document addition information automatic registration device 2000 can retrieve prompt templates from the prompt template DB 140.

[0214] The document DB120, generation AI server 130, and prompt template DB140 are the same as in Figure 1, so their explanations are omitted.

[0215] Next, the process performed by the document addition information automatic registration device 2000 in an embodiment of the present invention will be explained using the flowchart in Figure 21.

[0216] Figure 21 shows an example of the automatic document addition information registration process performed by the automatic document addition information registration device 2000 in the second embodiment.

[0217] From here on, we will explain using an example in which call log documents for a call center are registered in the document database, as shown in Figure 22. This document database includes the following text fields: "Document ID 2201", "Inquiry Details 2202", "Response Details 2203", and "Title 2204".

[0218] When registering data to the document database, the values ​​for inquiry content 2202 and response content 2203 can be entered manually. Alternatively, the inquiry content and response content fields may be automatically registered using text transcribed from the call log audio.

[0219] At the time of data registration, the value for Title 2204 will be blank, but Title 2204 is an item that is automatically registered, and its value will be automatically registered by the document addition information automatic registration process.

[0220] The automatic registration process for additional document information may be performed immediately after data registration, or it may be performed asynchronously through nightly batch processing or similar means.

[0221] Note that the item names in the document database and the items subject to automatic registration are not limited to this example. You may register the inquiry content and the response content without separating them, or you may include information such as the name of the customer who made the inquiry and the person in charge of the response.

[0222] In step S2101, the input is information that identifies a document in the document database and the name of the item in the document database that is to be automatically registered. For example, the document ID 2201 or the line number can be used as information that identifies a document.

[0223] In step S2102, the document is retrieved from the document database using information that identifies the document, such as the document ID 2201 mentioned above.

[0224] In step S2103, a prompt for generating additional document information is created from the values ​​of the document's text items. In this embodiment, a method is described in which the AI ​​generates a short sentence (title 2204) that will serve as a title to explain the item content, based on the inquiry content 2202 and response content 2203 items, as additional document information.

[0225] First, the default prompt template registered in the settings of the Document Addition Information Automatic Registration Device 2000 is retrieved from the prompt template DB140. The variables in the prompt template are set to the item values ​​of the document, and the correspondence between which variable corresponds to which item value is also registered in the settings of the Document Addition Information Automatic Registration Device 2000. A prompt is then created by setting the values ​​of the variables from there.

[0226] In step S2104, the prompt created in step S2103 is sent to the generation AI, and the generated text is retrieved.

[0227] The processes in steps S2103 and S2104 will be explained using Figure 23.

[0228] Figure 23 shows an example of prompt creation and output of the generation AI for generating additional document information in a second embodiment of the present invention. The example describes generating an item with the document title 2204 as additional information, using a document as input. It is assumed that "Title 2204" is specified as the item name to be automatically registered.

[0229] Let's explain using document 2205 shown in Figure 22 as an example. The inquiry states, "There is a paper jam. When I tried to send a fax, an A4 sheet of paper jammed. This is disrupting my work, so I would like to address this as soon as possible." The action taken is described as "Provide instructions on how to perform manual maintenance."

[0230] The prompt template for title generation is obtained from prompt template DB140, and prompt 2301, a string to be sent to the generation AI, is created by setting the input values ​​to variables. Prompt 2301 includes instructions that define the title generation task and a definition of the output format. The instructions and output format can include the following: ● Output the document title. ● Title character limit Furthermore, prompt 2301 includes text items from the document. In the example in Figure 23, it includes the inquiry and response from the document. However, it is not limited to this, and the prompt may include only the inquiry or the response.

[0231] The prompt thus created is sent to the generation AI server 130 to generate the title. The generated string "Regarding paper jams during fax transmission" can be used as the title string as is.

[0232] In other words, step S2104 is a step that shows an example of a process that outputs information and an instruction to generate a title that describes the information, and retrieves the title generated based on the instruction.

[0233] In S2105, the acquired text is used to update the values ​​of the items in the original document's database that are subject to automatic registration.

[0234] For example, in document 2205 shown in Figure 22, the value of "Title" is updated, and the string "Regarding paper jams during fax transmission" is newly registered. In this way, a string suitable as a title can be registered in the title field, which was blank when the document was registered in the document database.

[0235] In the analysis report generation process, the prompt templates and variable values ​​used can be specified according to the analysis report generation instructions sent by the client terminal 150 to the analysis report generation device 100. The specification of prompt templates and variables is described for each step in which a prompt is sent to the generation AI.

[0236] If no prompt template is specified in the analysis report creation instruction, the default settings are used, but these default settings can be changed by the system administrator who manages the analysis report generation device 100.

[0237] This structure allows for specifying how to change the items used in the document at each step of the analysis report creation process, depending on the purpose of the analysis and the budget.

[0238] For example, when creating an analysis report from the document shown in Figure 22, we can assume a scenario where the goal is to reduce the cost and processing load associated with using generation AI by performing topic extraction and classification using only the title. On the other hand, in the summarization process, there are cases where it is desirable to generate an accurate summary using information from the inquiry content and response content items, rather than just the short sentence of the title.

[0239] In the above case, the analyst sends an instruction to create an analysis report, as shown in Figure 24, to the analysis report generation device 100 via the client terminal.

[0240] Figure 24 shows an example of an analysis report creation instruction including prompt template settings in a second embodiment of the present invention.

[0241] In the analysis report creation instructions in Figure 24, the topic extraction step is described as using a prompt template with the value 2401 and template ID "TEMPLATE001". This prompt template has three variables: VAR01, VAR02, and VAR03. Of these, only VAR01 is used to input document values.

[0242] The instructions state that VAR01 should be assigned the titles extracted from the document list, VAR02 should be assigned the list of perspectives included in the analysis report creation instructions (e.g., Figure 8), and VAR03 should be assigned the data description included in the analysis report creation instructions (e.g., Figure 9).

[0243] This type of analysis report creation instruction makes it possible to include only the generated title 2204 from the items, rather than prompting for the entire document to be entered in the topic extraction step. The same applies to the topic classification step and the perspective classification step.

[0244] In other words, this is an example of a process that outputs a title, a set of perspectives for analyzing pre-defined information, and instructions to extract topics based on those perspectives, and then retrieves topics based on the perspectives extracted according to the instructions. Alternatively, it is an example of a process that outputs a title, topics based on the extracted perspectives, and instructions to classify the topics based on the titles, and then retrieves titles classified according to the instructions.

[0245] On the other hand, in the step of summarizing the viewpoint, it is stated that a prompt template with the value 2402 and template ID "TEMPLATE202" should be used.

[0246] This prompt template has four types of variables: VAR01, VAR02, VAR03, and VAR04. Of these, VAR01 and VAR02 are the two types used to input document values.

[0247] VAR01 is assigned the value extracted from the document list for the inquiry item, and VAR02 is assigned the value extracted from the document list for the response item. Furthermore, VAR03 is assigned the perspective name, which is the input data for the process, and VAR04 is assigned the data description included in the instruction to create the analysis report.

[0248] In other words, when creating summaries for each perspective, it is possible to include the inquiry content 2202 and the response content 2203 in the prompt and perform the summarization process. The same applies to the steps for summarizing topics and summarizing unclassified documents.

[0249] Through the processing described above, for the topic extraction and classification task, only the title 2204, a short sentence generated by the generation AI, is used, while for the summarization task, the inquiry content 2202 and response content 2203 are used. This allows for the arrangement of input content according to the processing, enabling the efficient and accurate generation of summaries. As a result, the report generation desired by the user can be achieved.

[0250] In this embodiment, the input content to the prompt was changed according to the processing based on the analysis report creation instructions in Figure 24. In addition, for example, when performing topic extraction and classification tasks, it may be possible to automatically select whether to use the entire document, as in the first embodiment, or a part of the document, as in the second embodiment, depending on the number of characters or words contained in the document. For example, if the number of characters or words exceeds a certain number, it is conceivable to perform topic extraction and classification using the method of the second embodiment. This makes it possible to perform classification in an efficient manner according to the volume of the document. That is, this is an example of a process that outputs an instruction to generate a title that describes the information when the information satisfies predetermined conditions.

[0251] Although embodiments have been described above, the present invention can take the form of, for example, a system, apparatus, method, program, or recording medium. Specifically, it may be applied to a system consisting of multiple devices, or to an apparatus consisting of a single device.

[0252] Furthermore, the program in this invention is a program that allows a computer to execute the processing method shown in the flowchart in Figure 3, and the storage medium of this invention stores a program that allows a computer to execute the processing method shown in Figure 3. Note that the program in this invention may also be a program for each processing method of each device shown in Figure 3.

[0253] As described above, it goes without saying that the object of the present invention can also be achieved by supplying a recording medium containing a program that realizes the functions of the embodiments described above to a system or device, and by having the computer (or CPU or MPU) of that system or device read and execute the program stored on the recording medium.

[0254] In this case, the program read from the recording medium itself realizes the novel function of the present invention, and the recording medium on which that program is recorded constitutes the present invention.

[0255] For recording media used to supply programs, examples include flexible disks, hard disks, optical disks, magneto-optical disks, CD-ROMs, CD-Rs, DVD-ROMs, magnetic tapes, non-volatile memory cards, ROMs, EEPROMs, silicon disks, and the like.

[0256] Furthermore, it goes without saying that the functions of the aforementioned embodiments are realized not only by the computer executing the program it has read, but also by the operating system (OS) running on the computer performing some or all of the actual processing based on the instructions of that program, thereby realizing the functions of the aforementioned embodiments.

[0257] Furthermore, it goes without saying that this also includes cases where, after a program read from a recording medium is written to the memory of a function expansion board inserted into a computer or a function expansion unit connected to a computer, the CPU or other components of the function expansion board or function expansion unit perform some or all of the actual processing based on the instructions of the program code, and the functions of the aforementioned embodiments are realized through that processing.

[0258] Furthermore, the present invention may be applied to a system consisting of multiple devices or to a device consisting of a single device. It goes without saying that the present invention can also be applied when the results are achieved by supplying a program to a system or device. In this case, by reading a recording medium containing a program for achieving the present invention into the system or device, the system or device can enjoy the effects of the present invention.

[0259] Furthermore, by downloading and reading the program for achieving the present invention from a server, database, etc. on a network using a communication program, the system or device can enjoy the effects of the present invention. It should be noted that configurations combining the above-described embodiments and their variations are all included in the present invention. [Explanation of Symbols]

[0260] 100 Analysis Report Generator 120 Document Database 130 AI Generator Server 140 Prompt Template DB 150 client terminals

Claims

1. A prompt generation means that generates prompts corresponding to each of the following instructions using a template: a first instruction to generate topics from the information to be analyzed based on two or more pre-set viewpoints for generating topics to be used for classifying the information; a second instruction to classify the information to be analyzed into the topics; and a third instruction to generate a summary of the information to be analyzed based on the viewpoints or topics. Output means for outputting the aforementioned prompt to the generating AI, Acquisition means for acquiring summary information generated by the generating AI based on a prompt corresponding to the third instruction, An information processing device characterized by comprising:

2. The acquisition means further acquires the topic generated by the generating AI based on the prompt corresponding to the first instruction. The second instruction is an instruction to classify the information to be analyzed into topics acquired by the acquisition means, The acquisition means further acquires a classification result based on a prompt corresponding to the second instruction, The information processing apparatus according to claim 1, characterized in that the third instruction is an instruction to generate a summary of the information to be analyzed based on the classification result obtained by the acquisition means.

3. The prompt generation means further generates a prompt that includes information not classified into the topic, the generated topic, and a fourth instruction to classify the information not classified into the topic into the topic. The information processing apparatus according to claim 2, characterized in that the acquisition means acquires the classification result generated by the generating AI based on the fourth instruction.

4. The information processing apparatus according to claim 1, characterized in that the third instruction for generating the summary is an instruction to generate a summary of the information classified into the said viewpoint or topic for each viewpoint or topic.

5. The information processing apparatus according to claim 1, characterized in that the prompt generation means generates a prompt including a fifth instruction to generate information and a summary that were not classified as topics.

6. The information processing apparatus according to claim 1, characterized in that the prompt generating means generates a prompt that includes an explanatory text for explaining the information together with the first instruction, the second instruction, or the third instruction.

7. The information processing apparatus according to claim 1, characterized in that the prompt generation means generates a prompt including information relating to the definition of the viewpoint together with the first instruction or the second instruction.

8. The information to be analyzed consists of multiple pieces of information, The information processing apparatus according to claim 2, characterized in that the prompt generation means divides the information into fixed numbers and includes them in the prompt.

9. The information processing apparatus according to claim 8, characterized in that the prompt generation means divides the information based on the number of characters or words contained in the information and includes it in the prompt.

10. The prompt generation means further generates a prompt including a sixth instruction that generates the information and a title describing the information, The acquisition means acquires the title generated by the generating AI based on the sixth instruction. The information processing apparatus according to claim 1, characterized by the following:

11. The prompt generation means further generates a prompt that includes the title, a pre-set viewpoint for analyzing the information, and a seventh instruction for generating a topic based on the viewpoint. The information processing apparatus according to claim 10, characterized in that the acquisition means acquires the topic generated by the generating AI based on the seventh instruction.

12. The prompt generation means further generates a prompt that includes the title, the topic, and an eighth instruction that classifies the topic based on the title. The information processing apparatus according to claim 11, characterized in that the acquisition means acquires the title classified by the generating AI based on the eighth instruction.

13. The information processing apparatus according to claim 10, characterized in that the prompt generation means generates a prompt including a sixth instruction for generating a title describing the information when the information satisfies predetermined conditions.

14. The information processing apparatus according to claim 13, characterized in that the predetermined conditions are conditions based on the number of characters or words in the information.

15. A prompt generation step in which the prompt generation means of an information processing device generates prompts corresponding to each of the following instructions using a template: a first instruction that generates topics from the information to be analyzed based on two or more perspectives set in advance for generating topics to be used for classifying information; a second instruction that classifies the information to be analyzed into the topics; and a third instruction that generates a summary of the information to be analyzed based on the perspectives or topics. The output means of the information processing device outputs the prompt to the generating AI in an output step, The acquisition means of the information processing device acquires summary information generated by the generating AI based on a prompt corresponding to the third instruction, A control method for an information processing device, characterized by comprising the following:

16. A program for causing at least one computer to function as one of the means of an information processing apparatus described in any one of claims 1 to 14.

Images