JP7851042B2 - Information processing method, program, and information processing device - Google Patents

Description

Applicable under Article 30, Paragraph 2 of the Patent Act. Published on June 1, 2024, on the website of Sales Retriever Co., Ltd. https://www.sales-retriever.jp/

This invention relates to an information processing method, a program, and an information processing apparatus.

Patent Document 1 discloses a method for estimating and presenting customer challenges and needs based on dialogue with the customer.

Japanese Patent Publication No. 2024-068624

Companies and other organizations engage in various sales activities to promote their products and services. However, finding the right connection between their products and services and potential customers, and developing appropriate proposals that will resonate with those customers, is challenging.

One aspect of this disclosure is the provision of information processing methods, etc., that enable the creation of proposals for more efficiently and easily selling products to target companies.

An information processing method relating to one aspect of this disclosure is an information processing method in which an information processing device acquires a research theme and target company, outputs research results regarding the theme and target company using a language model, and outputs a draft proposal for the target company by providing the output research results and product information to the language model.

According to one aspect of this disclosure, the information processing method, program, and information processing device enable the creation of proposals for selling products to target companies more efficiently and easily.

This is an overview diagram of the system relating to one aspect of this disclosure. This is a block diagram showing an example of a server device configuration. This is a block diagram showing an example configuration of a terminal device. This is a block diagram showing an example configuration of a generation server device. This is a sequence diagram showing an example of processing in the system shown in Figure 1. This is a sequence diagram illustrating an example of the research result output process. This figure shows an example of a keyword acquisition prompt. This figure shows an example of an article review prompt. This figure shows an example of article review information. This figure shows an example of a research result prompt. This figure shows an example of research results. This figure shows an example of a screen displaying research results. This figure shows an example of a proposed draft prompt. This figure shows an example of a template. This figure shows an example of product information. This figure shows an example of a proposed draft. This is a sequence diagram showing examples of different processes in the system shown in Figure 1.

The following describes in detail, with reference to the drawings, an example of an information processing method, program, and information processing apparatus relating to one aspect of this disclosure. In the description, similar elements are denoted by the same reference numerals, and redundant explanations are omitted as appropriate.

Figure 1 is an overview diagram of a system relating to one aspect of this disclosure. The system relating to this disclosure includes an information processing device 1, an information processing terminal 2, and a generation server device 4. Each device transmits and receives information via a network 3 such as the Internet. The generation server device 4 can utilize language models 45 (Figure 4), such as a Large Language Model (LLM), stored in internal or external storage of the generation server device 4.

The information processing device 1 is an information processing device that processes, stores, and transmits various types of information. The information processing device 1 may be configured as, for example, a server device, a personal computer, a tablet terminal, or a smartphone. The information processing device 1 may be composed of multiple information processing devices, or it may be configured as one of multiple virtual devices (virtual machines) within a single information processing device. In this embodiment of the disclosure, to avoid complexity in the explanation, the information processing device 1 will be described as a server device 1. However, the information processing device 1 is not limited to a server device and may be configured as any of the above-described types of information processing devices.

The information processing terminal 2 receives information output from the server device 1 via the network 3, accepts user input from the information processing terminal 2, and transmits the information related to the input to the server device 1 via the network 3. The information processing terminal 2 can be, for example, a personal computer, smartphone, mobile phone, wearable device, or tablet. In this disclosure, to avoid complexity in explanation, the information processing terminal 2 will be referred to as the terminal device 2.

Figure 2 is a block diagram showing an example configuration of server device 1. Server device 1 can be an information processing device composed mainly of electronic circuits using semiconductor circuit elements. Server device 1 includes a control unit 11, a communication unit 12, a reading unit 13, and a storage unit 14. The control unit 11, communication unit 12, reading unit 13, and storage unit 14 are connected via a bus 19 or the like for communication. Note that server device 1 may also have other configurations, such as omitting the reading unit 13.

The control unit 11 can be configured to include one or more processing units such as a CPU (Central Processing Unit), MPU (Micro-Processing Unit), GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), DSP (Digital Signal Processor), and a quantum processor. The control unit 11 reads and executes the program (or program product) 16 stored in the storage unit 14.

The communication unit 12 is a communication module for performing communication-related processing and can send and receive information with terminal devices 2, etc., via the network 3. The reading unit 13 can read portable storage media 15 such as CD (Compact Disc)-ROM, DVD (Digital Versatile Disc)-ROM, and USB® memory. The control unit 11 may read the program 16 from the portable storage media 15 via the reading unit 13 and save it to the storage unit 14. Alternatively, the control unit 11 may download the program 16 from another computer via the network 3, etc., and save it to the storage unit 14.

The storage unit 14 includes volatile storage units such as RAM (Random Access Memory), and non-volatile storage units such as ROM (Read Only Memory), HDD (Hard Disk Drive), and flash memory. As shown in Figure 2, the storage unit 14 stores, for example, a program 16. The program 16 is a program executed by the control unit 11. In this embodiment, the program 16 may also be referred to as a program product 16.

Figure 3 is a block diagram showing an example configuration of terminal device 2. Here, terminal device 2 can be an information processing device composed mainly of electronic circuits using semiconductor circuit elements. Terminal device 2 may include a control unit 21, a communication unit 22, a reading unit 23, a storage unit 24, a display unit 26, and an input unit 27. The components of the control unit 21, communication unit 22, reading unit 23, storage unit 24, display unit 26, and input unit 27 are connected to each other via a bus 29 or the like, enabling communication between them.

The control unit 21 can be configured to include one or more processing units such as a CPU (Central Processing Unit), MPU (Micro-Processing Unit), GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), DSP (Digital Signal Processor), and a quantum processor. The control unit 21 reads and executes the program stored in the storage unit 24.

The communication unit 22 is a communication module for performing communication-related processing and can send and receive information with the server device 1, etc., via the network 3. The reading unit 23 can read portable storage media 25 such as CD (Compact Disc)-ROM, DVD (Digital Versatile Disc)-ROM, and USB® memory. The control unit 21 can read programs and/or data from the portable storage media 25 via the reading unit 23 and save them to the storage unit 24. Furthermore, the control unit 21 can download programs from other computers via the network 3, etc., and save them to the storage unit 24.

The memory unit 24 includes volatile memory devices such as RAM (Random Access Memory), and non-volatile memory devices such as ROM (Read Only Memory), HDD (Hard Disk Drive), and flash memory. The display unit 26 can be a liquid crystal display or an organic EL (Electro-Luminescence) display, etc. The input unit 27 can be an input device such as a keyboard, mouse, touch panel, and camera.

Figure 4 is a block diagram showing an example configuration of the generation server device 4. The generation server device 4 can be an information processing device composed mainly of electronic circuits using semiconductor circuit elements. The generation server device 4 includes a control unit 41, a communication unit 42, a reading unit 43, and a storage unit 44. The control unit 41, communication unit 42, reading unit 43, and storage unit 44 are connected to each other via a bus 49 or the like for communication. Except for the contents stored in the storage unit 44, these components can be configured similarly to the information processing device (server device) 1 in Figure 2, so redundant explanations are omitted. Note that, as with server device 1, the generation server device 4 may have other configurations, such as not having a reading unit 43.

The memory unit 44 can store the language model 45. The language model 45 consists of a program and training data, and the control unit 41 executes the program by referring to the input data and training data received from an external source. The language model 45 may also be stored in an external storage device, such as an external storage device, rather than in the memory unit 44.

The generation server device 4 is an information processing device that handles language models 45, such as large-scale language models (LLMs) like GPT (Generative Pre-trained Transformer, registered trademark), BERT (Bidirectional Encoder Representations from Transformer), and Gemini (registered trademark). The language models 45 may be stored in the storage unit 44 within the generation server device 4, or they may be stored in storage connected to the generation server device 4 via communication.

The generation server device 4 receives input data such as images, audio, and text from the language model 45 and generates a response sentence. The language model 45 is a pre-trained machine learning model and can be a large-scale language model such as GPT (Generative Pre-trained Transformer) or BERT (Bidirectional Encoder Representations from Transformer), but other language models may also be used.

Server device 1 may be connected to generation server device 4 via network 3, or it may be connected to generation server device 4 directly or via a local network without using network 3. Since generation server device 4 is a device that handles language model 45, from the perspective of server device 1, sending prompts, etc., to generation server device 4 is expressed as server device 1 sending (inputting) to language model 45.

Figure 5 is a sequence diagram showing an example of processing in the system of Figure 1. In this sequence diagram, the processing of the server device 1 is executed based on the instructions of the program 16 stored in the storage unit 14. As shown in this sequence diagram, first, the terminal device 2 transmits the theme and target company information to the server device 1 (step S11). Here, the target company can be a company to be researched or a potential customer from whom a proposal draft for sales activities is to be obtained. In this embodiment, the target company is described as "ABC Company." The theme can be the content to be researched or information to obtain a proposal draft, as described later. To gather all information about the target company, the theme can consist of multiple themes, including general information about the target company.

For example, themes can include a wide range of topics, from basic company information such as "Company Overview/Business Activities," "Goals/Challenges/Initiatives of the Medium-Term Management Plan," "Human Resources/Talent Development Policy," "Human Capital Management," "Trends in Number of Hires and Employees," "Contact Information/Initiatives/Manager/Recruitment Information for the Human Resources Department," "Talent Management," "System Implementation for the Human Resources Department," "DX Promotion/Contact Information/Initiatives/Manager/Recruitment Information for the Systems Department," and "Labor Management," to matters related to the products or services you wish to promote.

Server device 1, having received the theme and target company information, executes research result output processing P30. Research result output processing P30 can involve server device 1 receiving research results from generating server device 4. Research result output processing P30 will be described in detail using Figure 6.

In the research result output processing P30, the server device 1, having acquired the research results, transmits them to the terminal device 2 (step S12), and the terminal device 2 can display the research results on the display unit 26 (Figure 12) (step S13). Note that the research results do not necessarily have to be displayed on the terminal device 2, nor do they necessarily have to be transmitted to the terminal device 2. Next, the user of the terminal device 2 transmits their company's product information (Figure 15) to the server device 1 by operating the terminal device 2 (step S14).

The server device 1, having received the product information, executes a proposal draft prompt creation process based on the research results and product information (step S15). The proposal draft prompt created in the proposal draft prompt creation process (e.g., Figure 13) is transmitted to the generation server device 4 (step S16). The generation server device 4 then uses the language model 45 to execute the proposal draft prompt processing (step S17) and generates a proposal draft (step S21).

The generated proposal draft is transmitted from the generation server device 4 to the server device 1 (step S21). Upon receiving the proposal draft, the server device 1 transmits it to the terminal device 2 (step S22). The terminal device 2, upon receiving the proposal draft, can then provide it to the user by displaying it on the display unit 26, etc.

Figure 6 is a sequence diagram showing an example of the research result output processing P30. The research result output processing P30 can be a process performed by the server device 1 together with the generation server device 4. The processing of the server device 1 in this sequence diagram may be executed based on instructions in the program 16 stored in the storage unit 14.

In the research result output processing P30, the server device 1 first executes the keyword acquisition prompt creation process (step S31). This keyword acquisition prompt creation process creates a keyword acquisition prompt (Figure 7) to be input to the language model 45. The keyword acquisition prompt may request the extraction of keywords suitable for finding the theme's content through website searches. Multiple themes can be combined into a single prompt.

The generated keyword acquisition prompt is sent to the generation server device 4 (step S32), and the generation server device 4 executes the keyword acquisition prompt processing, which is the processing described in the received keyword processing prompt (step S33). In the keyword acquisition prompt processing, one or more keywords suitable for finding the content of the requested theme (or each theme if there are multiple themes) are extracted for website searches.

The generation server device 4 transmits keywords to the server device 1 (step S35), and the server device 1 receives the keywords. Subsequently, the server device 1 scrapes articles from a group of websites 31 on the network 3 (such as the Internet) based on the received keywords (step S36). In other words, it sequentially browses websites on the network 3 and extracts and collects articles related to the keywords from their content (steps S37, S38). When collecting information on multiple themes, the server device 1 can perform scraping to collect articles for each theme.

As described above, the server device 1 can generate keywords for searching a theme using the language model 45. Furthermore, it can collect articles related to the theme based on the generated keywords. This allows for the acquisition of more necessary information related to the target company.

The server device 1, which has collected the articles, executes an article review prompt creation process (step S41) to create an article review prompt (Figure 8), which is a prompt for reviewing the articles. Once the article review prompt is created, the server device 1 sends it to the generating server device 4 (step S42). The generating server device 4, upon receiving the article review prompt, executes the article review prompt process, which is the process described in the received article review prompt (step S43).

The article review prompt processing can be a process that outputs article review information. In the article review prompt processing, the generation server device 4 uses the language model 45 to extract sentences related to the target company and theme as article review information for each collected article. This makes it easier to perform processing such as searching and extraction on the collected articles. Alternatively, based on prompts that extract whether each collected article is related to the theme, sentences within the article related to the theme, and the company name that is the subject of those sentences, the article review information may be extracted for each article.

Here, "sentences from articles related to the theme" can be direct quotes from the original articles. By objectively outputting one or more pieces of information corresponding to the theme in this way, it becomes easy to grasp the target company's approach or direction regarding the theme. Furthermore, by checking the "company name that is the subject of the sentence," it is possible to confirm whether the information pertains to the target company. This allows for a more accurate draft proposal, as described later.

The extracted article review information is transmitted from the generation server device 4 to server device 1 (step S44). Server device 1, having acquired the article review information, can then execute the research result prompt creation process (step S46). The research result prompt creation process creates a research result prompt (Figure 10). The research result prompt can include the title and URL of the article from which the article review information originated, the target company, and the headline theme, in addition to the acquired article review information.

Server device 1 transmits a research result prompt to generation server device 4 (step S47). Upon receiving the research result prompt, generation server device 4 executes research result prompt processing (step S48). The research result prompt processing extracts sentences from articles that are relevant to the theme and whose subject is the target company. This allows for the acquisition of only the necessary information related to the target company. Alternatively, the sentences from each article may be fed to the language model 45 for summarization to generate the research results (Figure 12). The summary may include links to the original sentences from which the summary was created. Even with lengthy articles, this method can produce more easily understandable research results.

The generation server device 4 transmits the generated research results to the server device 1 (step S49). The server device 1 then transmits the research results to the terminal device 2 (Figure 5, step S12), where they can be displayed (Figure 5, step S13). Furthermore, by transmitting multiple generated keywords and links to articles related to the target company, the display can include research results containing summaries of multiple texts and links to each text, multiple generated keywords, and information containing links to articles related to the target company (Figure 12). This allows the information to be displayed in a more easily understandable format for the user. Note that the research results do not necessarily have to be displayed on the terminal device 2, nor do they necessarily have to be transmitted to the terminal device 2.

Furthermore, the research result output process P30 may be performed for one theme at a time, and the process may be repeated for each theme. Alternatively, the research result output process P30 may be executed only once, with the keyword acquisition prompt process, keyword-based scraping, article refinement prompt process, and research result prompt process included in P30 each repeated for each theme.

As described above, the server device 1 acquires the research theme and target company, outputs research results regarding the theme and target company using the language model 45, and can output a proposal draft for the target company by providing the output research results and product information to the language model 45. This makes it easy to create proposal content for introducing the product to the target company.

As described above, the server device 1 can generate keywords for searching a theme using the language model 45 and output research results based on those keywords. Alternatively, it may collect articles related to the theme based on the generated keywords and output research results based on those articles. This allows for the acquisition of more necessary information related to the target company.

In this way, the server device 1 provides the language model 45 with prompts to extract, for each collected article, whether or not it is related to the theme, the sentences within the article related to the theme, and the company name that is the subject of the sentences. This allows the server device 1 to extract the article sentences that are related to the theme and whose subject is the target company.

This allows for the objective acquisition of one or more pieces of information related to the theme, making it easy to understand the target company's approach to the theme or its direction. Furthermore, by checking the "company name that is the subject of the text," it is possible to confirm whether the information pertains to the target company. This allows for the acquisition of more necessary information related to the target company, resulting in a more accurate proposal draft.

Figure 7 shows an example of a keyword acquisition prompt. The keyword acquisition prompt is created, for example, in the keyword acquisition prompt creation process shown in the sequence diagram of Figure 6, by server device 1. Furthermore, the keyword acquisition prompt is executed, for example, in the keyword acquisition prompt processing shown in the sequence diagram of Figure 6, by generation server device 4.

The keyword acquisition prompt in this example requests the language model 45 to output 1 to 4 search keywords for the target company that match the theme "Goals/Challenges/Initiatives of the Medium-Term Management Plan." Specifically, the keyword acquisition prompt can include the "theme," "keyword generation instructions for searching the theme," and "sample queries and answers."

In Figure 7, the "theme" is listed in the theme column 111, and the theme ("query") for which search keywords are requested is "Goals/Issues/Initiatives of the Medium-Term Management Plan." The "keyword generation command for searching the theme" is listed in the command column 112, and requests that one to four search keywords be extracted from the website based on the theme ("query") and output in JSON (JavaScript® Object Notation) format.

The "Sample Query and Answer" is described in Sample Column 113. As a sample answer format, when the theme ("query") is "Regarding the Goals and Status of Real Estate Investment Trusts," the keywords "Real Estate Investment Trust," "Real Estate Investment Trust Goals," and "Real Estate Investment Trust Status" are extracted and output in JSON format.

The keywords obtained by executing the keyword prompt in Figure 7 were, for example, four types of keywords: "Medium-Term Management Plan," "Medium-Term Management Plan Objectives," "Medium-Term Management Plan Challenges," and "Medium-Term Management Plan Initiatives."

Note that while Figure 7 shows only one theme, it is also possible to process multiple themes at once. To allow for a wider range of proposal possibilities when creating draft proposals, multiple themes covering diverse fields may be set.

In this way, the keyword acquisition prompt can generate keywords by providing the language model 45 with a prompt containing a theme, a command to generate keywords for searching the theme, and sample queries and answers. By using the keywords generated in this way, articles related to the theme can be collected more accurately. Furthermore, research results, as described later, can be output based on the generated keywords.

Figure 8 shows an example of an article review prompt. The article review prompt is created, for example, in the article review prompt creation process shown in the sequence diagram of Figure 6, by server device 1. Furthermore, the article review prompt is executed, for example, in the article review prompt processing shown in the sequence diagram of Figure 6, by generation server device 4.

As shown in this diagram, the article review prompt allows you to check each article collected through scraping to see if it contains content relevant to the theme and if it is an article that uses the target company as the subject. The results can then be displayed along with the original text of the source material.

Specifically, the article review prompt extracts information such as "whether each collected article is relevant to the theme," "sentences within the article related to the theme," and "the company name that is the subject of the sentence." The article review prompt may also include a "summary of the content related to the theme." The extracted content should be output in the specified "output format."

In Figure 8, the "Output Format" is listed as "return_format" in the format field 126. In the format field 126, the output format ("return_format") is in JSON format and has three elements: "Relevance to the theme of each collected article" ("related"), "Summary of the content about the theme" ("summary"), and "Extraced texts within the article related to the theme" ("extraced_texts"). Furthermore, "Extraced texts within the article related to the theme" ("extraced_texts") has the elements of the article's text ID ("id"), the text of that text ("text"), and the "Company name corresponding to the subject of the text" ("subject_company"). Here, "id" can correspond to subscripts 1 to 6, indicated by symbols A1 to A6, in the example of article analysis information shown in Figure 9.

The article review prompt allows for further specification of the requirements for each element. In the example in Figure 8, regarding "relevance to the theme of each collected article" ("related"), the prompt requires users to enter "yes" in the relevance field 124 if there are sentences related to the theme ("query"), and "no" if there are no related sentences. For "sentences within the article related to the theme" ("extraced_texts"), the prompt requires users to enter sentences directly related to the theme in their original form in the related sentences field 122.

Regarding the "subject_company" field, the system requires you to enter the name of the company that is the subject of the text directly related to the theme ("query") in the company name field 123. Regarding the "summary" field, the system requires you to enter a summary of the response (result) regarding the theme ("query") in the summary field 121.

Furthermore, in Figure 8, the theme field 125 of the article review prompt is written with the theme ("query") as "Goals/Challenges/Initiatives of the Medium-Term Management Plan," followed by each article ("content") to be extracted. Here, "query" may also be expressed as a question prompting an answer regarding the theme.

The article review prompt may request, for each collected article, the extraction of one or more of the following: whether or not it is relevant to the theme, sentences within the article related to the theme, and the company name that is the subject of the sentence.

This allows for the objective output of multiple pieces of information related to the theme, making it easy to grasp the target company's direction regarding the theme. Furthermore, by checking the "company name that serves as the subject of the text," it's possible to verify whether the information pertains to the target company. This allows for the creation of a proposal draft based on more accurate information.

Figure 9 shows an example of article review information. Article review information is generated, for example, by the generation server device 4 during the article review prompt processing in the sequence diagram of Figure 6. This example of article review information is output in JSON format. This article review information includes fields for the presence or absence of text related to the theme ("related"), a summary of the content on the theme ("summary"), and a description of text directly related to the theme ("extraced_texts"). Furthermore, the "extracted_texts" field includes a field for the text itself ("text") and a field for the company name that is the subject of the text ("subject_company").

Here, if an article contains multiple sentences directly related to the theme, multiple lists of these directly related sentences ("extraced_texts") can be created. In Figure 9, six sentences, A1 to A6, are extracted from article A as sentences directly related to the theme. Article A can be multiple articles, and multiple articles collected in step S36 may be targeted. Furthermore, multiple sentences directly related to the theme can be extracted from each of these multiple articles.

In this example, the "related" field indicates that there are relevant sentences to the theme, and a summary of the content on the theme is provided in the "summary" field. Additionally, six sentences directly related to the theme are listed in the "extraced_texts" field, and the subject of each sentence is "ABC Company."

Figure 10 shows an example of a research result prompt. The research result prompt is created, for example, in the research result prompt creation process shown in the sequence diagram of Figure 6, by server device 1. Furthermore, the research result prompt is executed, for example, in the research result prompt processing shown in the sequence diagram of Figure 6, by generation server device 4.

Here, the research results prompt can request the extraction of articles that are "relevant to the theme" and whose "subject is the target company." Regarding "relevance to the theme," the relevance column 131 in Figure 10 indicates that information unrelated to the theme ("query") should be excluded. Regarding "subject being the target company," the target company column 132 in Figure 10, etc., requires that the company name ("subject_company") is the same company as the target company ("company_name").

Furthermore, the research results prompt compiles the acquired article analysis information, along with the titles and URLs of the articles from which the analysis information originated, into a document ("doc"), and asks the user to answer the question ("query") based on this document ("doc").

Furthermore, an example answer ("<example>") is provided, and in addition to the content of the answer, the "text" of the article that served as the basis is indicated in the format of citation ID [item_X_X]. Here, the first "X" in [item_X_X] indicates the number assigned to the article, and the last "X" may indicate which sentence (or extracted sentence) it is. Also, if there is no information related to the theme ("query"), it is requested that "No relevant information was found" be written and that irrelevant information is not included.

Here, the article review information entered is not limited to one; multiple entries can be entered. Furthermore, since the system aims to exclude information unrelated to the theme, for example, by checking the "presence or absence of theme-related text ("related")" item in the article review information within a document ("doc"), if it is "no", the entire article will not be reviewed, thereby improving processing speed. Note that the articles within the document ("doc") included in the prompt may have already been reviewed by the server device 1 before being input to the language model 45, so that only articles where the "presence or absence of theme-related text ("related")" item is "yes" are included.

Figure 11 shows an example of research results. The research results are generated, for example, by the generation server device 4 during the research result prompt processing in the sequence diagram in Figure 6. The research results in this figure show the themes "Basic Policies and Goals," "Management Challenges," "Specific Initiatives," and "Strengthening the Management Foundation" for the target company, "ABC Company." The information in Figure 11 is compiled from articles collected through scraping, filtered through the research result prompt processing to remove irrelevant information, ensure articles are subject to the target company, and then summarized. This allows users to obtain more accurate information.

Specifically, in one example of the displayed research results, ABC Company's medium-term management plan is summarized under the headings of "Basic Policy and Goals," "Management Challenges," "Specific Initiatives," and "Strengthening the Management Foundation." Furthermore, the "Specific Initiatives" section is further divided into "Business Strategy," "Global Expansion," "Investment Plan," and "ESG Management and Sustainability." Each sentence includes a link to the cited article. In the example in Figure 11, under [item_2_1], the initial number "2" indicates the article number, and the following "1" may indicate the sentence number (or extracted sentence). Here, the link is not limited to an internet website; it can also be a link to the location where the acquired article is stored.

Figure 12 shows an example of a screen 100 displaying research results. This screen can, for example, be displayed on the display unit 26 of the terminal device 2 by the server device 1 that acquired the research results (Figure 5, step S12). The screen 100 displaying the research results may output "research results including summaries of multiple texts and links to each text," "multiple generated keywords," and "information including links to articles related to the target company."

In Figure 12, specifically, the server device 1 can display "research results including summaries of multiple texts and links to each text" in the research results display area 101, "generated keywords" in the search keyword display area 102, and "information including links to articles related to the target company" in the article link area 103.

The research results display area 101, similar to the research results in Figure 11, displays summarized information and links to each document regarding the themes "Basic Policies and Goals," "Management Challenges," "Specific Initiatives," and "Strengthening the Management Foundation" for the target company, "ABC Company."

In Figure 12, the research results display area 101 has some information omitted due to space limitations in the drawing. The search keyword display area 102 can display keywords extracted through keyword prompt processing. The article link area 103 can display links to articles used in the research results in the research results display area 101. Note that the links are not limited to links to internet websites; they can also be links to the storage location of the retrieved articles.

Figure 13 shows an example of a proposed draft prompt. The proposed draft prompt is created, for example, in the proposed draft prompt creation process shown in the sequence diagram of Figure 5, by server device 1. Furthermore, the proposed draft prompt is executed, for example, in the proposed draft prompt processing shown in the sequence diagram of Figure 5, by generation server device 4.

Here, the proposed draft prompt can include "multiple research results," "product information," and "templates." In Figure 13, the "multiple research results" are referenced in the research results column 144 by "researched_information." The referenced information can be, for example, the research results shown in Figure 11. While the research results shown in Figure 11 cover a single theme, the research results used in the proposed draft prompt are not limited to a single theme.

For example, research results can be used for all the following topics: "Company Overview/Business Overview," "Goals/Challenges/Initiatives of the Medium-Term Management Plan," "Human Resources/Talent Development Policy," "Human Capital Management," "Trends in Number of Hires and Employees," "Contact Information/Initiatives/Responsible Person/Recruitment Information for the Human Resources Department," "Talent Management," "System Implementation for the Human Resources Department," "DX Promotion/Contact Information/Initiatives/Responsible Person/Recruitment Information for the Systems Department," and "Labor Management." By creating a proposal draft based on such a large amount of research results, Server Device 1 can create a more persuasive proposal draft.

Furthermore, in Figure 13, "Product Information" is referenced in the Product Information field 143 by "product_information". The specific details of "Product Information" will be described in detail in Figure 13 and the subsequent explanation using Figure 13. "Template" is referenced in the Template field 142 by "proposal_format". The specific details of "Template" will be described in detail in Figure 14 and the subsequent explanation using Figure 14.

Furthermore, the proposal draft prompt, in instruction field 141, requests the language model 45 to respond in the role of an excellent sales assistant and to output a proposal or optimal approach to the customer. It also requests a proposal that is as specific as possible and tailored to the customer's situation.

Figure 14 shows an example of a template. The template is referenced or substituted, for example, as "proposal_format" in the proposal draft prompt in Figure 13. As shown in this figure, the template may include fields for outputting "the target company's challenges," "the supporting documents for the target company's challenges," and "information for accessing the sources of the supporting documents for the target company's challenges."

Specifically, in Figure 14, "Challenges of the Target Company" are output in the Challenges column 152 as challenges related to the company's products, based on the referenced texts. "Texts supporting the Target Company's Challenges" are output in the Reference Texts column 151 as texts referenced in listing the challenges. "Information for accessing the source of the texts supporting the Target Company's Challenges" are output in the Links column 153 as the reference ID. Note that links are not limited to internet websites; they can also be links to the location where the acquired articles are stored, etc.

Furthermore, the "Referenced Texts" section will include a "Reference ID" for each text, which allows for the output of links and other information to access the websites or other sources from which the referenced texts were quoted.

Thus, the template, in the "Proposal Direction" section, lists, for example, "Challenges of the Target Company," and in the "References" section, outputs "Documents that support the claims about the target company's challenges" and "Information to access the sources of those documents." The template is incorporated into the proposal draft prompt creation process, along with prompts requesting users to fill in the template's fields.

The "Information for Accessing the Source" (Reference ID) can be a link to a website, a link to a saved article, or other links. The template may also include a "Possible Approach" section, which describes the optimal product usage method based on challenges related to your company's products that can be identified from the referenced text.

Thus, the server device 1 inputs the template, along with the product information and research results, into the language model 45. This template can include fields for outputting the target company's challenges, the supporting documents for those challenges, and information for accessing the sources of those supporting documents.

A prompt input to the language model 45, including such a template, can request that the model output a proposal draft according to the template. In this way, the user can have the language model 45 output the target company's challenges, allowing them to verify the supporting documents and sources, thus enabling them to utilize the proposal draft in a more convincing manner. Furthermore, the template may include general company information such as the company name and number of employees, such as in a "Company Overview" section.

Figure 15 shows an example of product information. Product information is referenced or substituted, for example, as "product_information" in the proposed draft prompt in Figure 13. In the example in Figure 15, the product is a talent management system software named "TalentSoft".

Product information may include any of the following: "Product details," "Past proposals related to the product," and "Information on company conditions that may be suitable for the product." The example product information in Figure 15 includes a "Product Features" item under "Product Details." The "Product Features" item in Figure 15 indicates features such as a user interface that makes it easy to find the most suitable personnel even with over 10,000 employees.

Furthermore, the example product information in Figure 15 includes a "Past Cases" section under "Past Proposals Related to the Product." The "Past Cases" section in Figure 15 shows a successful example of developing individualized training plans for digital talent at DEF Company, Ltd. Additionally, the example product information in Figure 15 includes a "Company Conditions Suitable for the Product" section under "Company Conditions Suitable for the Product." The "Company Conditions Suitable for the Product" section lists companies that have established talent development goals or implemented human resources systems. This allows the server device 1 to efficiently apply the user's anticipated proposal content to the language model 45, resulting in a user-friendly proposal draft.

Figure 16 shows an example of a proposal draft. The proposal draft is generated by the generation server device 4 during the proposal draft prompt processing, as shown in the sequence diagram in Figure 5. As shown in the proposal draft in Figure 16, the proposal draft includes the template items "Company Overview" and "Proposal Direction." Furthermore, the "Proposal Direction" section includes the items "Referenced Texts," "Issues Related to Our Products Based on the Referenced Texts," and "Expected Approach Methods," and is output in a format that allows for the completion of each of these items.

Furthermore, each entry in the "Referenced Texts" section will display a four-digit "Reference ID," and by embedding a link to the cited document, website, or original document indicated by this number, users can access the source of the referenced text. Note that the links are not limited to internet websites; they can also be links to the location where the retrieved article is stored.

Thus, in the proposal draft, server device 1 can output the "Target Company's Challenges" as shown in the "Proposal Direction" column, and in the "Referenced Documents" column, it can output the "Documents that Support the Target Company's Challenges" and "Information for Accessing the Sources of the Documents that Support the Target Company's Challenges." Here, "Information for Accessing the Sources" can be links to websites, saved articles, or other links. Furthermore, server device 1 may also output general company information such as the company name and number of employees in the "Company Overview" column of the proposal draft.

In this way, the server device 1 outputs multiple research results for each of several different themes using the language model 45, and by providing the language model 45 with the output research results and product information, it outputs a proposal draft for the target company. This allows for a proposal draft based on more accurate information.

Furthermore, the server device 1 may output a proposal draft that includes the target company's challenges, the supporting documents for those challenges, and information for accessing the sources of those supporting documents. This allows users to verify the supporting documents and sources for the challenges output by the language model 45, enabling them to utilize the proposal draft in a more convincing way.

As shown in the above configuration, the server device 1 acquires the research theme and target company, outputs research results regarding the theme and target company using the language model 45, and can output a draft proposal to the target company by providing the outputted research results and product information to the language model 45. This makes it easy to create proposal content for introducing the product to the target company. Furthermore, since it can acquire not only information related to the product but also basic information about the target company, it can also explore the problems that the product solves from information unrelated to the product.

Figure 17 is a sequence diagram showing an example of a different process in the system of Figure 1. The sequence diagram in Figure 17 differs in that it performs the proposal draft prompt creation process without performing the processes in steps S12 and S13 of Figure 5. Here, the transmission of product information (step S14) is performed at the same time as the transmission of the theme and target company (step S11) in the sequence diagram of Figure 17. Even with this processing, the server device 1 can perform the proposal draft prompt creation process using the research results and product information. Therefore, it can output a proposal draft (steps S21 and S22).

Therefore, even in a system like the one shown in Figure 17, the server device 1 acquires the research theme and target company, outputs research results regarding the theme and target company using the language model 45, and outputs a draft proposal to the target company by providing the outputted research results and product information to the language model 45. This makes it easy to create proposal content for introducing the product to the target company. Furthermore, since it can acquire not only information related to the product but also basic information about the target company, it can explore the problems that the product solves even from information unrelated to the product.

In the above configuration, program 16 is stored in the storage unit 14 of server device 1, and the control unit 11 of server device 1 executes the processing of program 16. However, for example, program 16 may be stored in the storage unit 24 of terminal device 2, and the control unit 21 of terminal device 2 may directly access the generation server device 4 to execute the processing of program 16. Alternatively, part of program 16 may be executed in terminal device 2, and the other part of program 16 may be executed in server device 1, allowing terminal device 2 and server device 1 to operate collaboratively. Furthermore, while the above configuration stores the language model 45 in the generation server device 4, the language model 45 may be stored in server device 1 or terminal device 2, and the language model 45 may operate on either terminal device 2 or server device 1, or both, without accessing the generation server device 4.

Furthermore, the program in program 16 may be referred to as a program product, software, or software product, and these may be provided on a recording medium or distributed via a communication network.

The forms of this disclosure are illustrative in all respects and not restrictive. The scope of the invention is not limited to those shown in the above disclosure but is expressed by the claims, and all modifications within the meaning and scope equivalent to the claims are intended.

Furthermore, the sequences shown in each embodiment described above are not limiting, and within a consistent scope, the order of each processing step may be changed, and multiple processes may be executed in parallel. The processing entity for each process is not limiting, and within a consistent scope, the processing of each device may be executed by other devices.

The matters described in each embodiment can be combined with each other. Furthermore, the independent and dependent claims described in the claims can be combined with each other in any combination, regardless of the form of reference. In addition, although the claims do not use the form of a multi-claim (multi-multi-claim) which further references a multi-claim that references two or more claims, a combination using the form of a multi-multi-claim that references all higher-level claims is also possible.

1. Information processing equipment (server equipment)
11 Control unit 12 Communication unit 13 Reading unit 14 Storage unit 15 Portable storage medium 16 Program (program product)
19 Bus 2 Information Processing Terminal (Terminal Device)
21 Control Unit 22 Communication Unit 23 Reading Unit 24 Storage Unit 25 Portable Storage Medium 26 Display Unit 27 Input Unit 29 Bus 3 Network 4 Generation Server Device 41 Control Unit 42 Communication Unit 43 Reading Unit 44 Storage Unit 45 Language Model 49 Bus

Claims (5)

Information processing device,
The research theme and target companies are acquired using the aforementioned information processing device .
Based on keywords for searching the theme generated by the language model, articles related to the theme are collected.
By providing the language model with prompts to extract, for each collected article, whether or not it is related to the theme, the sentences within the article related to the theme, and the company name that is the subject of the sentences, the language model extracts, for each article, whether or not it is related to the theme, the sentences within the article related to the theme, and the company name that is the subject of the sentences.
Based on the presence or absence of extracted relevance and the company name that constitutes the subject, a language model is used to extract research results that are relevant to the aforementioned theme and that include sentences from the aforementioned articles of the aforementioned target company.
An information processing method that outputs a draft proposal to the target company by providing the language model with prompts including extracted research results, product information of the target company, and response commands for the content of the proposal to the customer . The aforementioned language model extracts multiple research results for each of the multiple different themes,
The information processing method according to claim 1, wherein the extracted research results and product information are provided to the language model to output the draft proposal for the target company. The aforementioned product information is acquired by the information processing device,
The information processing method according to claim 1 or 2 , wherein the product information includes any of the following: the content of the product, past proposals relating to the product, and information on company conditions that may be suitable for the product. The research theme and target companies are acquired using the aforementioned information processing device .
Based on keywords for searching the theme generated by the language model, articles related to the theme are collected.
By providing the language model with prompts to extract, for each collected article, whether or not it is related to the theme, the sentences within the article related to the theme, and the company name that is the subject of the sentences, the language model extracts, for each article, whether or not it is related to the theme, the sentences within the article related to the theme, and the company name that is the subject of the sentences.
Based on the presence or absence of extracted relevance and the company name that constitutes the subject, a language model is used to extract research results that are relevant to the aforementioned theme and that include sentences from the aforementioned articles of the aforementioned target company.
A program that causes an information processing device to execute a process that outputs a draft proposal to the target company by providing the language model with prompts including extracted research results, product information of the target company, and response commands for the content of the proposal to the customer. An information processing device comprising a control unit,
The control unit,
The research theme and target companies are acquired using the aforementioned information processing device .
Based on keywords for searching the theme generated by the language model, articles related to the theme are collected.
By providing the language model with prompts to extract, for each collected article, whether or not it is related to the theme, the sentences within the article related to the theme, and the company name that is the subject of the sentences, the language model extracts, for each article, whether or not it is related to the theme, the sentences within the article related to the theme, and the company name that is the subject of the sentences.
Based on the presence or absence of extracted relevance and the company name that constitutes the subject, a language model is used to extract research results that are relevant to the aforementioned theme and that include sentences from the aforementioned articles of the aforementioned target company.
An information processing device that outputs a draft proposal to the target company by providing the language model with prompts including extracted research results, product information of the target company, and response commands for the content of the proposal to the customer .