Information processing method, information processing program, and information processing device

Abstract

To provide an information processing method, an information processing program, and an information processing apparatus configured to create and output comments that specialists would likely make for an unknown or new case.SOLUTION: An information processing method causes a computer to execute the processes of: acquiring an inquiry including the content of activity; acquiring, using a learning model, a comment on biodiversity for the acquired inquiry; and outputting the acquired comment.SELECTED DRAWING: Figure 13

Description

[Technical Field]

[0001] The present invention relates to an information processing method, an information processing program, and an information processing device for outputting comments. [Background technology]

[0002] In recent years, the idea that certain restrictions should be placed on human activities in order to conserve biodiversity has become widespread. In particular, development projects undertaken by companies can have a significant impact on the environment, depending on the content and location of the project. In such cases, it may be necessary to abandon the project from the perspective of biodiversity conservation.

[0003] In the past, to assess the impact of implementing a project, companies would first create a business plan and then use methods such as environmental assessment to predict the impact on biodiversity. However, this type of work requires time and money, and if the project is abandoned, it will result in a loss for the company. For this reason, there is a demand to gather opinions from academia at an early stage of business planning to determine the merits of the plan. In relation to this situation, Patent Document 1 proposes a system for collecting expert opinions on articles. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2020 / 136760 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in conventional systems, it takes time to collect opinions for unknown or new cases for which there is no accumulated information. The present invention has been made in consideration of this situation. Its purpose is to provide an information processing method, an information processing program, and an information processing device that create and output comments that experts would make for unknown or new cases. [Means for solving the problem]

[0006] An information processing method according to an aspect of the present application includes: acquiring a query including a content of an activity; A comment corresponding to article information having a content similar to the acquired inquiry is extracted from a database in which article information including the content of the activity, comments from pre-designated experts regarding biodiversity in relation to the activity, and evaluations of the comments are associated with each other, in accordance with the evaluations, and a prompt including the acquired inquiry and the extracted comment is input into a language model. Comments on biodiversity in response to inquiries From the language model The computer executes a process of acquiring the comments and outputting the acquired comments. [Effects of the Invention]

[0007] In one aspect of the present application, it is possible to output comments that an expert would make. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is an explanatory diagram illustrating an example of the configuration of a comment system. [Figure 2] FIG. 2 is a block diagram showing an example of the hardware configuration of a comment server. [Figure 3] FIG. 2 is a block diagram illustrating an example of the hardware configuration of a company terminal. [Figure 4] FIG. 2 is a block diagram showing an example of the hardware configuration of an academic terminal. [Figure 5] FIG. 2 is an explanatory diagram illustrating an example of a company DB. [Figure 6] FIG. 10 is an explanatory diagram showing an example of an academic database. [Figure 7] FIG. 10 is an explanatory diagram showing an example of a synonym DB. [Figure 8] FIG. 10 is an explanatory diagram showing an example of an article DB. [Figure 9] FIG. 10 is an explanatory diagram showing an example of a comment DB. [Figure 10] FIG. 2 is an explanatory diagram illustrating an example of a knowledge DB. [Figure 11] 10 is a flowchart illustrating an example of a procedure for request processing. [Figure 12] 10 is a flowchart illustrating an example of a procedure for a generation process. [Figure 13] 10 is a flowchart illustrating an example of a procedure for comment generation processing. [Figure 14] 10 is a flowchart illustrating an example of a procedure for performing fine tuning. [Figure 15] 10 is a flowchart illustrating another example of the procedure of the generation process. [Figure 16] FIG. 10 is an explanatory diagram showing an example of an article list screen. [Figure 17] FIG. 10 is an explanatory diagram showing an example of a comment input screen. [Figure 18] FIG. 10 is an explanatory diagram showing an example of a question and answer screen. [Figure 19] FIG. 10 is an explanatory diagram illustrating an example of a chatbot screen. [Figure 20] 10 is a flowchart illustrating an example of a procedure for a field-specific tabulation process. [Figure 21] FIG. 10 is an explanatory diagram showing another example of the question and answer screen. [Figure 22] FIG. 10 is an explanatory diagram showing an example of a configuration graph of a specialty field. [Figure 23] FIG. 10 is an explanatory diagram showing an example of a specialty field configuration table. DETAILED DESCRIPTION OF THE INVENTION

[0009] The following describes embodiments with reference to the drawings. In the following description, "academia" primarily refers to researchers at academic research institutions, etc. Note that academia is interpreted as not including researchers affiliated with companies that have a vested interest in a particular company, but here it also includes researchers at private research institutions. "Activities" primarily refers to business activities carried out by for-profit companies, but also includes social activities carried out by non-profit organizations, organizations without legal personality, and individuals. In the following description, activities are activities that have or may have an impact on biodiversity. A business plan is a plan for business activities and includes the content of the business. It is desirable for a business plan to include the location of the business.

[0010] Figure 1 is an explanatory diagram showing an example of the configuration of a comment system. The comment system 100 includes a comment server 1, a company terminal 2, and an academic terminal 3. The comment server 1, the company terminal 2, and the academic terminal 3 are connected to each other via a network N so that they can communicate with each other. In addition, the comment system 100 can use a dialogue system 4 via the network N. In Figure 1, one company terminal 2 and one academic terminal 3 are shown, but there may be two or more of each.

[0011] The comment server 1 outputs comments on the input business plan. The comment server 1 is configured with a server computer, a workstation, a PC (Personal Computer), etc. The comment server 1 may also be configured with a multi-computer consisting of multiple computers, a virtual machine virtually constructed by software, or a quantum computer. Furthermore, the functions of the comment server 1 may be realized by a cloud service.

[0012] 2 is a block diagram showing an example of the hardware configuration of the comment server 1. The comment server 1 includes a control unit 11, a main memory unit 12, an auxiliary memory unit 13, a communication unit 14, and a reading unit 15. The control unit 11, the main memory unit 12, the auxiliary memory unit 13, the communication unit 14, and the reading unit 15 are connected by a bus B.

[0013] The control unit 11 has one or more arithmetic processing devices such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), a GPU (Graphics Processing Unit), etc. The control unit 11 reads and executes a control program 1P (program, program product) stored in the auxiliary storage unit 13, thereby performing various information processing, control processing, etc., and realizing various functional units.

[0014] The main memory unit 12 is a static random access memory (SRAM), a dynamic random access memory (DRAM), a flash memory, etc. The main memory unit 12 mainly temporarily stores data required for the control unit 11 to execute arithmetic processing.

[0015] The auxiliary storage unit 13 is a hard disk or an SSD (Solid State Drive) or the like, and stores the control program 1P and various DBs (Databases) required for the control unit 11 to execute processing. The auxiliary storage unit 13 stores a company DB 131, an academia DB 132, a synonym DB 133, an article DB 134, a comment DB 135, and a knowledge DB 136. The auxiliary storage unit 13 may be separate from the comment server 1 and may be an external storage device connected externally. The various DBs and the like stored in the auxiliary storage unit 13 may be stored in a database server or cloud storage different from the comment server 1.

[0016] The communication unit 14 communicates with the corporate terminal 2, the academic terminal 3, and the dialogue system 4 via the network N. In addition, the control unit 11 may use the communication unit 14 to download the control program 1P from another computer via the network N, etc., and store it in the auxiliary storage unit 13.

[0017] The reading unit 15 reads the portable storage medium 1a including a CD (Compact Disc)-ROM and a DVD (Digital Versatile Disc)-ROM. The control unit 11 may read the control program 1P from the portable storage medium 1a via the reading unit 15 and store it in the auxiliary storage unit 13. The control unit 11 may also read the control program 1P from the semiconductor memory 1b.

[0018] The company terminal 2 is a terminal used by a company representative seeking comments on the business plan. The company terminal 2 may be a laptop computer, a panel computer, a tablet computer, a smartphone, or the like.

[0019] 3 is a block diagram showing an example of the hardware configuration of a company terminal 2. The company terminal 2 includes a control unit 21, a main memory unit 22, an auxiliary memory unit 23, a communication unit 24, an input unit 25, and a display unit 26. Each component is connected by a bus B.

[0020] The control unit 21 has one or more arithmetic processing units such as a CPU, an MPU, a GPU, etc. The control unit 21 provides various functions by reading and executing a control program 2P (program, program product) stored in the auxiliary storage unit 23.

[0021] The main memory unit 22 is an SRAM, a DRAM, a flash memory, etc. The main memory unit 22 mainly temporarily stores data necessary for the control unit 21 to execute arithmetic processing.

[0022] The auxiliary storage unit 23 is a hard disk or an SSD, etc., and stores various data necessary for the control unit 21 to execute processing. The auxiliary storage unit 23 may be an external storage device connected to the company terminal 2. The various DBs, etc. stored in the auxiliary storage unit 23 may be stored in a database server or cloud storage.

[0023] The communication unit 24 communicates with the comment server 1 via the network N. In addition, the control unit 21 may use the communication unit 24 to download the control program 2P from another computer via the network N or the like, and store it in the auxiliary storage unit 23.

[0024] The input unit 25 is a keyboard and a mouse. The display unit 26 includes a liquid crystal display panel or an organic EL (electroluminescence) display panel, etc. The display unit 26 displays comments output by the comment server 1. The input unit 25 and the display unit 26 may be integrated to form a touch panel display. The company terminal 2 may display on an external display device.

[0025] The academic terminal 3 is a terminal used by academia. The academic terminal 3 is composed of a laptop computer, a panel computer, a tablet computer, a smartphone, etc.

[0026] 4 is a block diagram showing an example of the hardware configuration of an academic terminal. The academic terminal 3 includes a control unit 31, a main memory unit 32, an auxiliary memory unit 33, a communication unit 34, an input unit 35, and a display unit 36. Each component is connected via a bus B. The control unit 31, the main memory unit 32, the auxiliary memory unit 33, the communication unit 34, the input unit 35, and the display unit 36 ​​have the same configuration as the control unit 21, the main memory unit 22, the auxiliary memory unit 23, the communication unit 24, the input unit 25, and the display unit 26 of the corporate terminal 2, respectively, and therefore their explanation will be omitted.

[0027] The dialogue system 4 includes a dialogue server 41 and an LLM 42 (learning model). Under the control of the dialogue server 41, the dialogue system 4 provides an interactive AI (Artificial Intelligence) service using LLM 42 (Large Language Models). LLM 42 is a natural language processing model trained using large amounts of text data. The dialogue system 4 is configured using Transformer, BERT, GPT, ChatGPT, BARD, etc. Note that the comment server 1 may be equipped with a large-scale language model and provide an interactive AI service.

[0028] Next, the database will be explained. Fig. 5 is an explanatory diagram showing an example of a company DB. The company DB 131 stores information about companies. The company DB 131 includes a company ID column, a name column, and an industry column. The company ID column stores a company ID that identifies a company. The name column stores the name of the company. The industry column stores the industry of the business that the company is engaged in.

[0029] Figure 6 is an explanatory diagram showing an example of an academia DB. The academia DB 132 stores information about academia. The academia DB 132 includes an AID column, a name column, an affiliation category column, an affiliation column, a title column, and a field of expertise column. The AID column stores an AID that can uniquely identify an academia. The name column stores the name of the academia. The affiliation category column stores the category of the research institution to which the academia belongs. The category may be a university or a national or public research institute. The affiliation column stores the name of the research institution to which the academia belongs. The title column stores the title of the academia. The field of expertise column stores the field of expertise of the academia. Academia registered in the academia DB 132 are pre-designated experts who can comment on biodiversity.

[0030] FIG. 7 is an explanatory diagram showing an example of a thesaurus DB. The thesaurus DB 133 stores information about synonyms. The thesaurus DB 133 includes a synonym ID string, a representative name string, a synonym string, an associated word string, and a supplement string. The synonym ID string stores the synonym ID that identifies the synonym. The representative name string stores words used when standardizing notation. The synonym string stores multiple words that are similar to each other. The associated word string stores words associated with the representative name and synonym. The supplement string stores supplementary information and notes.

[0031] FIG. 8 is an explanatory diagram showing an example of an article DB. The article DB 134 stores news articles from newspapers, magazines, etc. The article DB 134 includes an article ID column, a title column, a body column, a location column, an author column, a reference column, a distribution date and time column, a deletion date and time column, a correction date and time column, a search index column, a key phrase column, a category column, a sentence vector column, and a summary column. The article ID column stores an article ID that identifies an article. The title column stores the title of the article. The body column stores the body of the article. The location column stores the location where the activity covered in the article took place. The location here is not limited to information indicating a geographical location. Information on places such as sandy beaches, riverbanks, and wetlands, which have different natural environments and are considered from different perspectives when considering their impact on biodiversity, may also be stored as the location. Furthermore, proper names of places whose natural environments can be understood, such as Tokyo Bay, Lake Kasumigaura, and Kamikochi, may also be stored as the location. Location determination is performed, for example, by syntactically analyzing the article text, extracting candidate words, and comparing the extracted words with a pre-prepared dictionary to determine whether they indicate a location. The author column stores the author's name if the article is a bylined article. The reference material column stores reference material information if there is reference material for the article. The distribution date and time column stores the date and time the article was distributed. The deletion date and time column stores the deletion date and time if the article has already been deleted. The correction date and time column stores the correction date and time if the content of the article was corrected after distribution. The search index column stores words that serve as an index when searching for an article. The main phrase column stores the main phrases that make up the article content. The category column stores the article category. The sentence vector column stores a feature vector that represents the content of the article. The summary column stores a summary of the article in three sentences.

[0032] FIG. 9 is an explanatory diagram showing an example of a comment DB. Comment DB 135 stores comments from academia on articles. Comment DB 135 includes a comment ID column, an article ID column, a main text column, an AID column, a reference material column, a creation date and time column, a deletion date and time column, a correction date and time column, a correction destination ID column, an evaluation column, and a parent ID column. The comment ID column stores a comment ID that identifies a comment. The article ID column stores the article ID of the article that the comment is the subject of. The main text column stores the main text of the comment. The AID column stores the AID of the academia that made the comment. The reference material column stores information about the material if there is material related to the comment. The creation date and time column stores the date and time the comment was created. The deletion date and time column stores the date and time the comment was deleted if the comment has been deleted. The correction date and time column stores the date and time the comment was corrected if the comment has been corrected. The correction destination ID column stores the comment ID of the corrected comment. The evaluation column stores the evaluation of the comment. When storing ratings as numerical values, the rating value may take a five-point scale, for example, from 1 to 5, with 5 being the highest rating and 1 being the lowest rating. Ratings may also be represented by symbols such as circles, triangles, or crosses, or by images that indicate "likes." If a comment is a comment in response to another comment, the parent ID column stores the comment ID of the comment to which the comment was added.

[0033] FIG. 10 is an explanatory diagram showing an example of a knowledge DB. The knowledge DB 136 stores article content and comments from academia in association with each other. The knowledge DB 136 includes a data ID column, an article ID column, an activity content column, an activity location column, an article vector column, a comment column, a comment vector column, an AID column, a distribution date and time column, a search index column, a key phrase column, and a category column. The data ID column stores data IDs that identify training data. The article ID column stores article IDs of related articles. The activity content column stores activity content included in an article. The activity content may be, for example, a company's business content. The activity location column stores information about the location where the activity content included in an article is performed. The location information is the value of the location column in the article DB 134. The article vector column stores feature vectors generated from the article. The comment column stores comments on the article, i.e., the content of the body column of the comment DB 135. The comment vector column stores feature vectors generated from the comments. The AID column stores the AID of the academia who made the comment. The distribution date and time column stores the date and time the article was distributed. The search index column stores the search index assigned to the article. The main phrase column stores the main phrases that make up the article. The category column stores the category of the activity content.

[0034] The feature vectors described above are vectors used in natural language processing. The process of vectorizing a sentence is a well-known technique, and therefore a detailed description thereof will be omitted.

[0035] Next, the information processing performed by the comment system 100 will be described. FIG. 11 is a flowchart showing an example of the procedure for request processing. The request processing is a process of requesting academia to submit comments on article information. The request processing is periodically started as a batch process, or is started when a robot program such as a crawler detects that article information that can be collected has been distributed. The control unit 11 of the comment server 1 collects articles (step S1). The control unit 11 collects articles from pre-set news sites, etc. If articles have already been collected by a crawler, etc., the control unit 11 collects the articles from the crawler, etc. The control unit 11 selects one article to process (step S2). The control unit 11 creates article data (step S3). The control unit 11 extracts titles, selects key words and phrases, assigns categories, creates feature vectors, creates summaries, etc. These processes can be realized using known technologies, so explanations will be omitted. Note that part of the creation work may be performed manually. The control unit 11 stores the article data in the article DB 134 (step S4). The control unit 11 creates and edits a request for comments on the article (step S5). The control unit 11 selects the academia to which the request is made as necessary. It is assumed that all academia will be requested, but depending on the article category, the request may be made only to academia that specialize in a pre-set academic field. If a request for the academia to which the request is made has already been made, the control unit 11 edits the request and adds the request content. If a request has not yet been made, the control unit 11 creates a new request. The control unit 11 determines whether there is unprocessed article data (step S6). If the control unit 11 determines that there is unprocessed article data (YES in step S6), the process returns to step S2 and processes the unprocessed article data. If the control unit 11 determines that there is no unprocessed article data (NO in step S6), the control unit 11 sends a request to the academia to make a comment (step S7) and ends the process. The transmission method is email, push notification, etc.

[0036] FIG. 12 is a flowchart showing an example of the procedure for the generation process. The generation process is a process for generating knowledge data that associates article information with comments from academia in response to the article. The generation process is initiated, for example, by academia selecting a hyperlink included in a comment request. The control unit 31 of the academic terminal 3 sends a request for a comment input form to the comment server 1 (step S11). The control unit 11 of the comment server 1 receives the request (step S12). The control unit 11 determines whether comments are being accepted for the target article (step S13). If the control unit 11 determines that comments are being accepted (YES in step S13), it creates a comment input screen (step S14). The control unit 11 sends the created input screen to the academic terminal 3 (step S15). The control unit 31 of the academic terminal 3 receives and displays the input screen (step S16). The academia enters a comment on the input screen and issues a transmission instruction. The control unit 31 accepts the comment and sends it to the comment server 1 (step S17). The control unit 11 of the comment server 1 receives the comment (step S18). The control unit 11 generates knowledge data that associates article information with the comment (step S19). The control unit 11 stores the knowledge data in the knowledge DB 136 (step S20). The control unit 11 sends a notification of acceptance completion to the academic terminal 3 (step S21). The control unit 31 of the academic terminal 3 receives and displays the completion (step S22). The control unit 31 ends the processing. If the control unit 11 determines that comments are not being accepted (NO in step S13), it sends a message to the academic terminal 3 indicating that comments cannot be accepted (step S23). The control unit 31 of the academic terminal 3 receives the message (step S24). The control unit 31 displays the message (step S25) and ends the processing.

[0037] Whether or not comments are being accepted is stored in the article DB 134, etc. Comments are accepted for a predetermined period after the article is distributed, or until a predetermined number of comments are received. Regardless of the period or number of comments, comments may be accepted until the article is deleted.

[0038] FIG. 13 is a flowchart showing an example of the procedure for the comment generation process. The comment generation process uses knowledge data to generate comments that are likely to be submitted by academia regarding a company's new business, etc. A company employee inputs a business plan, including the business content and location, into the company terminal 2. The control unit 21 of the company terminal 2 accepts the business plan (step S41). When the employee operates a button requesting comments, the control unit 21 transmits the accepted business plan to the comment server 1 (step S42). The control unit 11 of the comment server 1 receives the business plan (step S43). The control unit 11 vectorizes the sentences that indicate the business content included in the business plan (step S44). At this time, the control unit 11 refers to the synonym DB 133 and replaces used terms with representative names, etc. The control unit 11 then creates a feature vector from the sentence. Using the created feature vector, the control unit 11 refers to the knowledge DB 136 to extract articles with similar content and comments on them (step S45). Similarity is determined using Euclidean distance, cosine similarity, etc. The control unit 11 sets a role (step S46). The control unit 11 sends a prompt including the role, an inquiry about the business plan, and the extracted comments to the dialogue system 4 (step S47). A role is a sentence that instructs the dialogue system 4 on how to behave. For example, "You are an editor who collects comments evaluating the impact of corporate activities on biodiversity from multiple academics and creates representative comments." is set as the role. The dialogue system 4 sends the role, the inquiry about the business plan, and responses to the comments to the comment server 1. The number of comments to be created by the dialogue system 4 may be specified. The control unit 11 of the comment server 1 receives the responses (step S48). The control unit 11 creates a screen including the received responses (step S49). The control unit 11 sends the created screen to the corporate terminal 2 (step S50). The control unit 21 of the corporate terminal 2 receives the screen (step S51). The control unit 21 displays the screen (step S52) and ends the process.

[0039] (Example prompt) Below is an example of a prompt sent to the dialogue system 4 in the comment generation process.

[0040] You are an editor responsible for collecting comments from multiple academics assessing the impact of corporate activities on biodiversity and compiling a representative set of comments. {User's inquiry} Please refer to the following article information when creating this. {Article ID...} {Article ID...} {Article ID...} The output format should be separated by researcher, listing the researcher's name, comments, and evaluation value.

[0041] When sending the above prompt, the control unit 11 creates a virtual table (View) based on the academic DB 132, article DB 134, and comment DB 135, which contains the article ID, the body of the article, the comment ID, the name of the researcher who made the comment, the body of the comment, and the evaluation of the comment, and sets it up so that the dialogue system 4 can refer to it.

[0042] (Narrow down the target articles) In the comment generation process, when extracting articles that contain content similar to the business content, it is assumed that similarity is calculated using a brute force method. However, since a brute force method can result in a huge amount of calculation, it is also possible to narrow down the target articles. For example, if the business content entered by the person in charge includes location information, only articles with the same or similar location information will be extracted.

[0043] The attributes of the comments on articles can also be used as filtering conditions. The degree (Master's degree or below, Doctorate) obtained by the academia who made the comment can be used as a condition. Another condition is the academia's affiliation (university, national or public research institute, private research institute). The search can also be narrowed down by the period in which the comments were made. Comments that meet the conditions are selected, and articles corresponding to the selected comments are extracted. In addition, comments can be selected based on the evaluation value of the comment, as described below, or the evaluation value of the academia who made the comment, and articles corresponding to the selected comments are extracted. Note that the conditions for comments are also used when selecting comments that are associated with similar articles.

[0044] The process of narrowing down comments is performed after extracting comments in step S45 of FIG. 13. In the case of academia-related conditions, the control unit 11 of the comment server 1 acquires the AID of the academia that created the extracted comment. The control unit 11 searches the academia DB 132 using the acquired AID as a search key to acquire information about the academia. The control unit 11 compares the acquired academia information with the narrowing down conditions. If the narrowing down conditions are met, the control unit 11 keeps the comment as a processing target, and if the narrowing down conditions are not met, the control unit 11 excludes the comment from the processing target. In the case of comment-related conditions, the control unit 11 compares the attributes of the extracted comment (creation date, rating, etc.) with the narrowing down conditions. If the narrowing down conditions are met, the control unit 11 keeps the comment as a processing target, and if the narrowing down conditions are not met, the control unit 11 excludes the comment from the processing target.

[0045] (Variation) 13, multiple past articles similar to the content of the input business and comments on those articles are sent to the dialogue system 4 to generate a comment, but this is not limiting. Fine-tuning of the LLM 42 of the dialogue system 4 may be performed in advance, and the content of the input business may be sent to the dialogue system 4 and comments on it may be received.

[0046] FIG. 14 is a flowchart showing an example of the procedure for executing fine tuning. The control unit 11 of the comment server 1 creates a dataset (training data) by referring to the article DB 134 and the comment DB 135 (step S61). Each piece of data in the dataset consists of "system," "user," and "assistant." For example, "You can comment as a biodiversity expert" is set in the system. "Please write a comment on the following article as a biodiversity expert" is set in the user, which is a combination of the article text from the article DB 134. The comment text from the comment DB 135 is set in the assistant. Such data is created for all articles, and used as a dataset. The control unit 11 stores the dataset in the auxiliary storage unit 13 or the like (step S62). The control unit 11 transmits the dataset to the dialogue system 4 (step S63). The dialogue server 41 of the dialogue system 4 receives the dataset (step S64). The dialogue server 41 uses the dataset to train the LLM 42 (step S65). The dialogue server 41 transmits completion to the comment server 1 (step S66). The control unit 11 of the comment server 1 receives the completion notification (step S67) and ends the process. By the above process, some of the parameters of the LLM 42, which have already been optimized to some extent, are updated, thereby causing the LLM 42 to learn new knowledge.

[0047] FIG. 15 is a flowchart showing another example of the steps of the generation process. A company representative inputs a business plan, including the business content and location, into the company terminal 2. The control unit 21 of the company terminal 2 accepts the business plan (step S81). The control unit 21 transmits the accepted business plan to the comment server 1 (step S82). The control unit 11 of the comment server 1 receives the business plan (step S83). The control unit 11 creates a query (step S84). For example, the control unit 11 creates data by combining a question (inquiry) such as, "Please comment on the impact of the following business plan on biodiversity." with the received business plan. The control unit 11 sets a role (step S85). For example, the control unit 11 sets a role such as, "You are an expert on biodiversity. You can answer questions such as the impact of social activities on biodiversity." The control unit 11 then transmits the created role and query to the dialogue system 4 (step S86). The dialogue server 41 of the dialogue system 4 receives the role and query. The dialogue server 41 provides the role and query to the LLM 42. The LLM 42 outputs the comment. The dialogue server 41 transmits the comment to the comment server 1. The control unit 11 of the comment server 1 receives the comment (step S87). The control unit 11 creates a screen including the comment (step S88). The control unit 11 transmits the screen to the company terminal 2 (step S89). The control unit 21 of the company terminal 2 receives the screen (step S90). The control unit 21 displays the screen (step S91) and ends the processing.

[0048] FIG. 16 is an explanatory diagram showing an example of an article list screen. The article list screen d01 is a screen that displays a list of collected articles. The article list screen d01 includes a hot news list d011, the number of hot comments d012, a new news list d013, the number of comments d014, comments d015, the number of favorable ratings d016, and handle names d017. The hot news list d011 displays the titles of the top few articles in terms of the number of comments. The number of hot comments d012 displays the number of comments for each article displayed in the hot news list d011. The new news list d013 displays several headlines of newly collected articles. The number of comments d014 displays the number of comments for each article displayed in the new news list d013. Comments d015 displays comments for the selected article. Comments d015 can be displayed or hidden using an accordion UI (User Interface). In the example of FIG. 16, when one of the comment counts d014 is selected and a mouse click or the like is performed, the comment d015 for the corresponding article is displayed. When the comment d015 is displayed, if the corresponding comment count d014 is selected and a mouse click or the like is performed, the displayed comment d015 will be hidden. The number of favorable ratings d016 indicates the number of favorable ratings, commonly known as "likes." Operating the icon to the left of the number with a mouse click or the like will input a "like" for the comment. The handle name d017 indicates the handle name of the academia that input the comment. The handle name is entered each time a comment is input, or it is stored in advance in the academia DB 132.

[0049] The number of comments for each article displayed on the article list screen d01 may be calculated each time by referring to the comment DB 135, but is not limited to this. A column for storing the number of comments may be provided in the article DB 134, and the count value of the comment may be stored in that column each time a comment is made. The number of favorable ratings for each comment may be stored in the comment DB 135.

[0050] FIG. 17 is an explanatory diagram showing an example of a comment input screen. The comment input screen d02 is a screen for academia to input comments on articles. When an article headline is selected on the article list screen d01 by clicking the mouse or the like, a comment input screen d02 is displayed on which a comment on the selected article can be input. The comment input screen d02 includes a new news list d021, a hot news list d022, article content d023, a comment field d024, a registration button d025, and a handle name d026. The new news list d021 displays several headlines of newly collected articles. The hot news list d022 displays the titles of the top few articles in terms of the number of comments. The article content d023 displays the content of the selected article. The comment field d024 is a field in which academia inputs comments. When the registration button d025 is selected, the comment entered in the comment field d024 is sent from the academia terminal 3 to the comment server 1 and stored in the comment DB 135. The handle name d026 is the handle name of Academia. Here, it is displayed assuming that it has been set in advance.

[0051] The article list screen d01 shown in Fig. 16 may be a home screen that is displayed after an academia logs in to the comment system 100. The comment input screen d02 shown in Fig. 17 is an example of an input screen that is displayed on the academic terminal 3 in step S16 of the generation process shown in Fig. 12.

[0052] FIG. 18 is an explanatory diagram showing an example of a question and answer screen. The question and answer screen d03 is a screen on which a company inputs inquiries about its business and receives responses from the comment system 100. The question and answer screen d03 is intended to be used in the comment generation process shown in FIG. 13. The question and answer screen d03 includes a question input field d031, an execute button d032, option settings d033, a question d034, a comment d035, an article list d036, a comment d037, a number of favorable ratings d038, and a handle name d039. The question input field d031 is a field for inputting a question for which a comment is requested. When the execute button d032 is selected, the question entered in the question input field d031 is sent from the company terminal 2 to the comment server 1. The option settings d033 set extraction conditions when the comment server 1 extracts comments. These extraction conditions are used in step S45 of the comment generation process shown in FIG. 13. In other words, articles associated with comments that satisfy the extraction conditions are extracted. Question d034 indicates a question sent to the comment server 1. Comment d035 indicates a comment sent from the comment server 1 as an answer to the question. Article list d036 indicates a list of articles extracted as articles similar to the question. Comment d037 displays comments on an article selected in article list d036. The article list d036 can be displayed or hidden using an accordion UI (User Interface), as in FIG. 16 . Comments on the extracted article are sent from the comment server 1 to the dialogue system 4, and a summary of the comments returned from the dialogue system 4 is displayed as the above-mentioned comment d035. Only comments that satisfy the conditions set in option setting d033 are sent to the dialogue system 4. Number of favorable ratings d038 indicates the number of favorable ratings for the comment. Handle name d039 indicates the handle name of the academia who entered the comment. A rating for comment d035 displayed on the question and answer screen d03 may be entered, for example, on a three-point scale. The evaluation may be reflected in the evaluation of the referenced comment d037.

[0053] FIG. 19 is an explanatory diagram showing an example of a chatbot screen. The chatbot screen d04 is a screen provided by the comment server 1. The comment server 1 provides a chatbot in cooperation with the dialogue system 4, for example, using an API (Application Programming Interface) provided by the dialogue system 4. Providing the chatbot is premised on the fact that the LLM 42 of the dialogue system 4 has been fine-tuned by the fine-tuning execution process shown in FIG. 14. The chatbot screen d04 includes a question input field d041, an execute button d042, a question d043, an answer d044, and an end button d045. The question input field d041 is a field for inputting content for which a comment is requested as a question. When the execute button d042 is selected, the question entered in the question input field d041 is sent from the company terminal 2 to the comment server 1. The comment server 1 sends the question and role to the dialogue system 4. The role does not need to be sent when sending the second or subsequent questions. A question d043 indicates the question sent to the comment server 1. The dialogue server 41 of the dialogue system 4 uses the LLM 42 to generate an answer to the question and transmits it to the comment server 1. The comment server 1 transmits the answer to the company terminal 2. The answer is displayed on the company terminal 2. This answer is answer d044. When the end button d045 is selected, the chatbot screen d04 is closed and the chat ends.

[0054] This embodiment provides the following advantages. In this embodiment, collected newspaper and magazine articles related to biodiversity are presented to biodiversity experts (academia) to collect comments. A collection of combinations of articles and comments constitutes the knowledge of the comment system 100. The comment system 100 uses this knowledge to generate comments that are likely to be used in response to inquiries from academia. This enables companies and other organizations to determine the degree of impact on biodiversity when considering new businesses. Companies and other organizations can refer to the comments obtained from the comment system 100 to decide whether to continue with their business plans, revise their plans by changing the implementation location, or freeze their business plans. Furthermore, to confirm the validity of the content of the comments, they can also seek advice from academia as necessary. The content of the business for which inquiries are made is not limited to new businesses; it can also be about businesses that the company is already conducting.

[0055] It is expected that the comment system 100 will indirectly deepen communication between companies and other entities that have little knowledge about the impact of social activities such as business on biodiversity and academia, which conducts daily research on the impact of social activities on biodiversity. As a result, it is believed that direct communication between companies and academia will be carried out appropriately.

[0056] (Comment rating) The key to the comment system 100 creating appropriate comments is the selection of the comments entered by the academics that form the basis for the comments. Therefore, the system selects which comments to use based on the evaluation value of each comment. Specifically, when extracting comments in step S45 of the comment generation process shown in FIG. 13, comments with evaluation values ​​below a predetermined threshold are excluded. Similarly, when creating a dataset in step S61 of the fine-tuning execution process shown in FIG. 14, comments with evaluation values ​​below a predetermined threshold are not included in the dataset.

[0057] Various evaluation values ​​are possible, and examples are shown below. The number of favorable ratings, or the so-called "likes," as described above, may be used as the evaluation value. Alternatively, the percentage of the number of "likes" may be calculated assuming that the number of academics participating in the comment system 100 is 100, and the obtained value may be used to classify comments and to use this as the evaluation value for each comment. In addition to rating comments by giving "likes," each comment may also be rated on a multi-level scale. For example, a three-level rating may be given, with either "good," "average," or "bad."

[0058] Furthermore, for comments that have many further comments, the evaluation value is relatively high or relatively low. If a comment with many further comments is deemed to be an important comment that deserves attention, the evaluation value is high. On the other hand, if it is deemed to be a so-called flame war and its inclusion would reduce the accuracy of the comment, the evaluation value is low. Since a computer cannot determine which of these two categories it is, such comments can be excluded from extraction in the first place, or the decision can be made by an administrator or other person.

[0059] In addition, academia is evaluated, and comments created by highly rated academia are given a relatively higher evaluation value, while comments created by low-rated academia are given a relatively lower evaluation value. Academia is evaluated, for example, by the average number of "likes" per comment, calculated by dividing the total number of "likes" received by past comments by the number of comments.

[0060] The evaluation value of the comment may be used when displaying the comment. In Figures 16 and 18, when comments for a selected article are displayed, comments with higher evaluation values ​​are displayed at the top.

[0061] (Statistical value display function) The following describes a function that presents statistical values ​​related to comments created by the comment system 100 to end users. The statistical values ​​displayed include, for example, the total number of articles stored in the article DB 134, the number of articles sent to the dialogue system 4 in the generation process shown in FIG. 13, the total number of comments added to the articles, and the average number of comments per article. In the fine-tuning execution process shown in FIG. 14, the number of articles and the total number of comments included in the dataset also apply. The number of articles can be considered as a real number or a total number. If one article has four comments, the total number is 4 and the real number is 1.

[0062] Statistical values ​​about the academia participating in the comment system 100 may also be displayed, such as the number of participating academia, the number of people by specialty, and the percentage of people by specialty. It is possible to display the percentage of academia by specialty that made comments, the percentage of academia by specialty that corresponds to comments linked to articles sent to the dialogue system 4, and the percentage of academia by specialty for comments used in fine tuning and included in the dataset.

[0063] FIG. 20 is a flowchart showing an example of the procedure for the field-specific aggregation process. The field-specific aggregation process is a process for calculating the proportion of each specialized field of academia that commented on an article. The control unit 11 of the comment server 1 resets the aggregation counter to 0 (step S101). The control unit 11 acquires the article ID to be processed (step S102). The control unit 11 searches the comment DB 135 and acquires the comments associated with the article ID (step S103). The control unit 11 selects one comment to be processed (step S104). The control unit 11 acquires the AID assigned to the academia that created the selected comment (step S105). The control unit 11 stores the acquired AID in a temporary storage area provided in the main memory unit 12, the auxiliary memory unit 13, etc. (step S106). The control unit 11 determines whether there are any unprocessed comments (step S107). If the control unit 11 determines that there is an unprocessed comment (YES in step S107), it returns the process to step S104 and processes the unprocessed comment. If the control unit 11 determines that there is no unprocessed comment (NO in step S107), it deletes duplicate AIDs from among those stored in the temporary storage area (step S108). The control unit 11 selects one AID (step S109). The control unit 11 refers to the academia DB 132 and acquires the academic specialty to which the selected AID is assigned (step S110). The control unit 11 counts a counter corresponding to the acquired specialty (step S111). If there are multiple specialty fields, it counts each counter. The control unit 11 determines whether there is an unprocessed AID (step S112). If the control unit 11 determines that there is an unprocessed AID (YES in step S112), it returns the process to step S109 and processes the unprocessed AID. If the control unit 11 determines that there are no unprocessed AIDs (NO in step S112), it outputs the number of academics and the ratio of the number of people in each field (step S113) and ends the process. The number of academics and the ratio of the number of people in each field may be updated each time a comment is added to each article and stored in the article DB 134.

[0064] FIG. 21 is an explanatory diagram showing another example of the question and answer screen. The question and answer screen d03 shown in FIG. 21 is a modified example of the question and answer screen d03 shown in FIG. 18. In FIG. 21, the same components as those in FIG. 18 are given the same reference numerals, and detailed description will be omitted. The question and answer screen d03 shown in FIG. 21 has added an article number d03A, an academic information button d03B, and a statistical information link d03C. The article number d03A indicates the number of articles referenced when the comment d035 was created. Selecting the academic information button d03B displays information about the academia participating in the comment system 100. Selecting the statistical information link d03C displays information about the academia that made the comment.

[0065] Figure 22 is an explanatory diagram showing an example of a specialty composition graph. This is an example of the screen displayed when the Academia Information button d03B in Figure 21 is selected. The graph in Figure 22 shows the specialty composition of academia participating in the comment system 100. In the example in Figure 22, biology is the most prevalent, accounting for 30% of the total. This is followed by forest science at 20% and science at 15%. It is also possible to display the number of participants instead of the composition percentage. When calculating the composition percentage, the calculation is limited to academia that has commented on an article or evaluated comments made by other academia. Researchers not involved in the learning model M are not counted. This is because we believe that showing the number of experts involved in developing the functions provided by the comment system 100, rather than simply the number of experts registered in the comment system 100, will help to gain trust in the comment system 100.

[0066] Figure 23 is an explanatory diagram showing an example of a specialty composition table. This is an example of the screen that is displayed when the statistical information link d03C is selected in Figure 21. The specialty composition table shows the composition of the specialty fields of the academics who commented on the article linked to the statistical information link d03C. In the example of Figure 23, there are three comments, which indicates that one academic specializing in each of ecology, fisheries science, and environmental science has commented.

[0067] By referring to the statistical values ​​explained above, company personnel can numerically estimate the reliability of the comment system 100.

[0068] In the above explanation, it is assumed that the comment server 1 uses an external dialogue system 4. Here, "external" means that the operator of the dialogue system 4 is a separate organization that has no capital or collaborative relationship with the operator of the comment server 1. The operator of the comment server 1 may also operate the dialogue system. For example, the learning model M of the comment server 1 may be a large-scale language model, and the comment server 1 may provide a dialogue system 4 that uses the learning model M.

[0069] The technical features (constituent elements) described in each embodiment can be combined with each other, and by combining them, new technical features can be formed. The embodiments disclosed herein are to be considered as illustrative in all respects and not restrictive. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the meaning and scope of the claims. In addition, the claims are written in a format in which a claim cites two or more other claims (multiple claim format), but this is not limited to this. Multiple claims that cite at least one other claim (multi-multi claim format) may also be written. [Explanation of symbols]

[0070] 100: Comment System 1: Comment server 11: Control section 12: Main memory 13: Auxiliary storage section 131: Corporate DB 132: Academia DB 133: Thesaurus DB 134: Article DB 135: Comment DB 136: Knowledge DB 14: Communications Department 15: Reading unit 1P: Control program 1a: Portable storage medium 1b: Semiconductor memory 2: Corporate devices 21: Control unit 22: Main memory 23: Auxiliary storage section 24: Communications Department 25: Input section 26:Display section 2P: Control program 3: Academia terminal 31: Control unit 32: Main memory 33:Auxiliary storage section 34: Communications Department 35: Input section 36:Display section 4: Dialogue system 41: Interactive server 42:LLM B: Bus N: Network

Claims

1. Obtain inquiries including activity details, extracting comments corresponding to article information having content similar to the acquired inquiry from a database in which article information including the content of the activity, comments from pre-designated experts regarding biodiversity in relation to the activity, and evaluations of the comments are associated with each other, in accordance with the evaluations; inputting a prompt including the acquired query and the extracted comment into a language model, thereby acquiring a comment, which is a sentence about biodiversity in response to the acquired query, from the language model; Output the retrieved comments An information processing method in which processing is performed by a computer.

2. The language model is trained using training data in which article information including the content of an activity is associated with comments from pre-designated experts on biodiversity regarding the activity based on the article information. The information processing method according to claim 1 .

3. The training data includes ratings for the comments. The information processing method according to claim 2 .

4. Obtaining a query that includes a description of the activity and a location of the activity. The information processing method according to any one of claims 1 to 3.

5. The number of article information used in learning the language model, and the actual number or percentage of experts who evaluated the comments or experts who created the comments are output for each field. The information processing method according to claim 2 .

6. Obtain inquiries including activity details, extracting comments corresponding to article information having content similar to the acquired inquiry from a database in which article information including the content of the activity, comments from pre-designated experts regarding biodiversity in relation to the activity, and evaluations of the comments are associated with each other, in accordance with the evaluations; inputting a prompt including the acquired query and the extracted comment into a language model, thereby acquiring a comment, which is a sentence about biodiversity in response to the acquired query, from the language model; Output the retrieved comments An information processing program that causes a computer to execute a process.

7. An information processing device including a control unit, The control unit Obtain inquiries including activity details, extracting comments corresponding to article information having content similar to the acquired inquiry from a database in which article information including the content of the activity, comments from pre-designated experts regarding biodiversity in relation to the activity, and evaluations of the comments are associated with each other, in accordance with the evaluations; inputting a prompt including the acquired query and the extracted comment into a language model, thereby acquiring a comment, which is a sentence about biodiversity in response to the acquired query, from the language model; Output the retrieved comments An information processing device that executes processing.

Images