Matching system, matching method, and matching program

Abstract

The system matches multiple business entities based on potential elements related to the content of the input natural language text. [Solution] The matching system 1 includes: a first acquisition unit 40 that acquires response information R regarding the request from a language model 21 that responds about the content of taste T1 in response to the input of natural language text T1 containing first information I1 regarding the request of the first business entity X; a second acquisition unit 50 that acquires second information I2 containing at least one of the business content and the product content handled by each of a plurality of second business entities Y; a selection unit 80 that selects at least one partner candidate for the first business entity X based on a first similarity A1 between the response information R and at least one of the business content and the product content included in the second information I2; and an output unit 90 that outputs information about the partner included in the selected partner candidate C1.

Description

[Technical Field]

[0001] This invention relates to a matching system, a matching method, and a matching program. [Background technology]

[0002] In matching systems that present suitable combinations of companies for transactions, a technique is known that matches the first company and the second company based on the similarity between phrases extracted from natural language text containing the needs (requests) of the first company and the company information of the second company. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Patent No. 7586184 [Patent Document 2] Patent No. 7452701 [Patent Document 3] Patent No. 7094515 [Overview of the project] [Problems that the invention aims to solve]

[0004] However, when using the matching technology described above to search for potential partners, the potential partners that are matched are limited by the content of the natural language text entered by the user. Therefore, it is difficult to match multiple business entities based on potential factors.

[0005] In one aspect, the present invention aims to match multiple business entities based on potential elements relating to the content of input natural language text. [Means for solving the problem]

[0006] In one aspect, the matching system may be a matching system that matches multiple business entities with each other. The matching system may include a first acquisition unit that acquires at least one response piece of information regarding the requirements of a first business entity from a language model that responds to the content of a natural language text input containing first information regarding the requirements of a first business entity. The matching system may include a second acquisition unit that acquires second information including at least one of the business content and the product content handled by each of a plurality of second business entities. The matching system may include a selection unit that selects at least one candidate partner for the first business entity from among the plurality of second business entities based on a first similarity between the response information acquired by the first acquisition unit and at least one of the business content and the product content included in the second information acquired by the second acquisition unit. The matching system may include an output unit that outputs information about the partner included in the candidate partner selected by the selection unit. [Effects of the Invention]

[0007] In one aspect, the present invention can match multiple business entities based on potential factors. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows an example of the configuration of the matching system according to the first embodiment. [Figure 2] This figure shows another example of the configuration of the matching system according to the first embodiment. [Figure 3] This figure shows an example of the processing performed by the matching system of the first embodiment. [Figure 4] This is a flowchart illustrating an example of the operation of the matching system according to the first embodiment. [Figure 5] This figure shows an example of the processing content of the matching system according to the second embodiment. [Figure 6] This figure shows an example of the processing content of the matching system according to the third embodiment. [Figure 7] It is a diagram illustrating an example configuration of the matching system according to the fourth embodiment. [Figure 8] It is a diagram illustrating an example hardware configuration of the matching system according to the first to fourth embodiments. MODE FOR CARRYING OUT THE INVENTION

[0009] Hereinafter, for example, the present embodiment can be implemented with various modifications without departing from the scope of the spirit thereof. In the drawings used in the following description, portions denoted by the same reference numerals represent the same or similar portions unless otherwise specified.

[0010] [1] First Embodiment [1-1] Functional Configuration FIG. 1 is a diagram illustrating an example configuration of a matching system 1 according to the first embodiment. The matching system 1 is a business matching system that matches a plurality of business entities with one another. A business entity may be a company, a corporation, a university, a public institution, an individual business owner, or the like.

[0011] The matching system 1 of the present embodiment proposes candidates of business entities suitable as partners when a business entity forms a partnership with another business entity. As used herein, the term "partner" may include business alliance partners, capital alliance partners, providers of technology or products, and work request destinations.

[0012] In one example, the target business entity that wants to search for a partner may be a first business entity X, and the plurality of candidate business entities are second business entities Y1, ..., Y N (where N is a natural number of 2 or more). The first business entity X and the plurality of second business entities Y1, ..., Y N (hereinafter may be collectively referred to as a plurality of second business entities Y) are each an example of the first business entity and the second business entity.

[0013] When the matching system 1 receives natural language text T1 containing first information I1 regarding the requests of the first project entity X, it obtains response information R regarding the requests from a language model such as a large language model (LLM) 21. The matching system 1 then uses the response information R regarding the first project entity X and multiple second project entities Y1,...,Y N Based on the first similarity A1 between the business content (e.g., technical content) and product content of the second business entity Y1,...,Y N From among these candidates, we will select C1 as our first potential partner.

[0014] The matching system 1 has a control unit 10 that controls the matching system 1. The control unit 10 may be a processor 101 as shown in Figure 8, which will be described later. The control unit 10 includes a GUI (Graphical User Interface) 30, a first acquisition unit 40, a second acquisition unit 50, a selection unit 60, and an output unit 90. Furthermore, the control unit 10 may include a collection unit 70 and a second selection unit 80.

[0015] GUI30 is an example of an interface into which natural language text T1 is input. The natural language text T1 includes first information I1. The first information I1 concerns the requests of the first project entity X. The natural language text T1 to GUI30 may be input by a user. The user may be an affiliate of the first project entity X or another operator.

[0016] The natural language text T1 is input to the generative artificial intelligence model LLM21. The natural language text T1 may also be used as a type of prompt, instructing LLM21 on the tasks it should perform and the direction it should take.

[0017] LLM21 is an example of a language model that responds to questions about the content of natural language text in response to natural language text input. In this matching system 1, LLM21 may be a general-purpose large-scale language model such as ChatGPT® or BERT (Bidirectional Encoder Representations from Transformers). Since LLM21 itself is the same technology as before, a detailed explanation will be omitted.

[0018] The first acquisition unit 40 acquires at least one response information R related to the request from the LLM 21 in response to the input of natural language text T1 containing the first information I1. If there are multiple types of proposals with different perspectives to satisfy the request, multiple types of response information R may be acquired. The response information R may be text data, or it may be data that has been pre-vectored into a feature vector in the LLM 21 for the similarity calculation in the selection unit 60 described later. Vectorization is a method of converting text data into a feature vector as an array of numbers.

[0019] The second acquisition unit 50 acquires second information I2 for each of the multiple second business entities Y. The second information I2 includes at least one of the business content and the product content handled by each of the multiple second business entities Y. The business content may include technical content.

[0020] The second information I2 may be stored in the business entity database (DB) 22. The second information I2 may, for example, be extracted from the technical section of a securities report and stored in the business entity DB, but is not limited to this case. The business entity DB 22 may be a commercial database. Furthermore, the second acquisition unit 50 may acquire the second information I2 from a website or the like via a network (NW) instead of the business entity DB 22. The network may be, for example, the internet. The second information I2 may be text data, or it may be pre-vectorized data for the similarity calculation in the selection unit 60.

[0021] The selection unit 60 may include a first similarity calculation unit 61 and a first ranking unit 62. The first similarity calculation unit 61 calculates a first similarity A1 between the response information R acquired by the first acquisition unit 40 and each business content and each product content included in the second information I2.

[0022] The first similarity calculation unit 61 vectorizes the response information R into a feature vector. Similarly, the first similarity calculation unit 61 vectorizes each business content and each product content included in the second information I2 of each second business entity Y into a feature vector. If the contents of the response information R and the second information I2 have already been vectorized, the vectorization process is omitted in the first similarity calculation unit 61. If there are multiple types of response information R, a feature vector may be obtained individually for each type of response information R.

[0023] The first similarity calculation unit 61 calculates the first similarity A1 based on the feature vector of the response information R and the feature vector of at least one of the business content and product content included in the second information I2. The first similarity A1 is calculated based on the feature vector of the response information R of the first business entity X and the feature vector of multiple second business entities Y1,...,Y N The first similarity A1 may be the Euclidean distance in the feature space between the feature vectors of each business activity and each product activity, at least one of each. The smaller the Euclidean distance, the higher the similarity. In addition, the first similarity A1 is the feature vector of the response information R of the first business entity X and the feature vectors of multiple second business entities Y1,···,Y N This could also be the cosine similarity between each feature vector and at least one of the business activities and product contents of each entity. Cosine similarity takes values ​​in the range of -1 to 1, with values ​​closer to 1 indicating higher similarity. The first similarity A1 can be any numerical representation of the degree of similarity and is not limited to Euclidean distance and cosine similarity. Multiple second business entities Y1,···,Y N For each of these, the first similarity score A1 is calculated.

[0024] When there are a plurality of types of response information R, the first similarity calculation unit 61 may individually calculate each first similarity A1 using the feature vectors calculated for each of the plurality of types of response information R.

[0025] The first ranking unit 62 is configured to process a plurality of second business entities Y1, ···, Y N based on the first similarity A1 calculated for each of said plurality of second business entities Y1, ···, Y N to calculate first ranking information B1 that ranks the plurality of second business entities. The first ranking unit 62 may set a higher ranking for the business entity as the value of the first similarity A1 increases. Alternatively, in addition to the first similarity A1, the first ranking unit 62 may calculate a matching score based on other factors such as the scale, transaction history, and region of the second business entity Y. In this case, the first ranking unit 62 ranks the plurality of second business entities Y using the matching score.

[0026] The first ranking unit 62 may select business entities ranked within a predetermined rank (a predetermined number of top-ranked business entities) as first partner candidates C1. The number of first partner candidates C1 to be selected may be appropriately set, for example, 3.

[0027] Note that the first similarity A1 and the first ranking information B1 calculated by the selection unit 60 may be output to a monitor (such as the IO unit 105 in FIG. 8 described later). In this case, the user can recognize the first similarity A1 and the first ranking information B1.

[0028] For reasons such as the response information R acquired by the first acquisition unit 40 from the LLM 21 being abstract, the first similarity A1 may not be high with respect to a plurality of second business entities Y. When the user recognizes the first similarity A1 and the like displayed on the monitor and notices that the first similarity A1 is not high, the user may input new natural language text T1 including more specific content.

[0029] Furthermore, if the first similarity A1 is below a threshold, the first acquisition unit 40 may execute a cycle 41 in which it inputs new natural language text based on the response information R into the LLM 21 to acquire new response information. The value of the first similarity A1 compared with the threshold may be the maximum value, average value, or median value of the first similarity A1 for each of the multiple second business entities Y.

[0030] In cycle 41, the first acquisition unit 40 may create a prompt that adds an instruction to the natural language text T1 to specify the content, and input the created prompt as new natural language text into LLM21. The first acquisition unit 40 may repeat cycle 41 without user processing until the first similarity A1 exceeds a threshold.

[0031] The collection unit 70 collects Web information W about each of the first partner candidate C1 selected by the selection unit 60 from the network. The network may be, for example, the internet. The Web information W may be collected from a WWW (World Wide Web) server 23.

[0032] The second selection unit 80 may include a second similarity calculation unit 81 and a second ranking unit 82. The processing of the second selection unit 80 may be a process of supplementing the second information I2 stored in the business entity DB 22 and confirming the consistency of the matching results based on the second information I2 using actual Web information W. The processing of the second selection unit 80 is called the supplemental consistency confirmation process of business entity information (company information).

[0033] The second similarity calculation unit 81 vectorizes the Web information W collected for each of the first partner candidate C1 into feature vectors. The second selection unit 80 obtains the feature vectors of the response information R from the selection unit 60, etc. If there are multiple types of response information R, the feature vectors of each type of response information R may be obtained.

[0034] The second similarity calculation unit 81 calculates the second similarity A2 based on the feature vector of the response information R and the feature vectors of the Web information W for each of the first partner candidate C1. The second similarity A2 may be calculated in the same way as the first similarity A1, except that the value used as the basis for calculation is the Web information W instead of the second information I2.

[0035] The second ranking unit 82 performs a re-ranking of multiple first partner candidate C1s based on the second similarity A2 calculated for each of the first partner candidate C1s, and calculates the second ranking information B2.

[0036] The second ranking unit 82 may select business entities ranked within a predetermined rank (a predetermined number of top business entities, for example, the highest-ranking business entity) as second partner candidate C2. The number of second partner candidate C2 to be selected may be set as appropriate. The second ranking unit 82 is an example of a ranking unit that ranks at least some of the partner candidates based on a second similarity A2 between the collected information and the response information R acquired by the first acquisition unit 40.

[0037] The output unit 90 outputs information about the partners included in the first partner candidate C1 selected by the selection unit 60. The partners whose information is output by the output unit 90 may be at least one of the first partner candidate C1, and may also be a second partner candidate C2 further selected from the first partner candidate C1, as shown in Figure 1. The output unit 90 may output first generated information O1a which includes the requesting company and the reason for the request. The requesting company is an example of a partner, and in this embodiment, it is the second partner candidate C2. The output first generated information O1a may be presented to the user by a monitor of the IO unit shown in Figure 8, which will be described later.

[0038] Furthermore, the functions in the second selection unit 80, including vectorization to feature vectors for calculating the second similarity, the second ranking function, and the generation of reasons for making a request, may be implemented using LLM24, a general-purpose large-scale language model.

[0039] In the example shown in Figure 1, the case where cycle 41 is executed until the natural language text T1 is materialized, and the re-selection process using Web information W is performed is described, but these functions can be omitted. In this case, as shown in Figure 2, cycle 41, the collection unit 70, and the second selection unit 80 may be omitted. In this case, the output unit 90 outputs the first partner candidate C1 selected by the selection unit 60 as the first generated information O1a. The other configurations are the same as the matching system 1 shown in Figure 1, so a detailed explanation is omitted.

[0040] [1-2] Regarding the processing details Figure 3 shows an example of the processing performed by the matching system 1 of the first embodiment. The matching system 1 of the first embodiment is used by a business entity facing a problem to find other business entities (companies, etc.) that have the ability to solve that problem.

[0041] In this embodiment, the first information I1 relating to the request includes information about a problem that the first business entity X has, and the response information R includes information about a solution to the problem that the first business entity X has. The solution to the problem may be a problem-solving technology.

[0042] When a user inputs natural language text T1 containing a problem, LLM21 generates information about problem-solving techniques. The generated information about problem-solving techniques is obtained as response information R.

[0043] For example, if the first information I1 includes the problem that "data recovery will be difficult in the event of an earthquake because the data is stored in a storage device within a single building," then response information R from multiple perspectives may be obtained as solutions to this problem. For example, the first acquisition unit 40 may obtain multiple types of response information from LLM21 from different perspectives, such as response information for a first solution, "acquire a building with a seismic isolation structure," and response information for a second solution, "use a cloud service."

[0044] Selection Section 60 is a technology for solving problems and multiple second business entities Y1,...,YN The first similarity A1 (see Figure 1) is calculated between each of the provided technologies. The provided technologies may be any technologies that can be provided. The provided technologies are examples of the second information I2 stored in the business entity DB22 (for example, the company DB).

[0045] If the first similarity score A1 is low due to reasons such as the information about the problem-solving technology obtained from LLM21 being abstract, cycle 41 for concretizing the problem-solving technology may be executed. Cycle 41 may be repeated without user instruction until the first similarity score A1 exceeds a threshold.

[0046] The selection unit 60 ranks the second project entity Y based on the calculated first similarity score A1.

[0047] The selection unit 60 selects, for example, several business entities (companies) as candidates for business entities that have the ability to solve the problems of the first business entity X. One example of a business entity that has the ability to solve the problems is the first partner candidate C1.

[0048] The second information I2 stored in the business entity DB22 is reliable, but it does not contain any new information about potential partners that has been added since it was stored in the business entity DB22. Therefore, LLM24 (second selection unit 80) obtains Web information W from the Web server 23 on the network to secure new information. Using the obtained Web information W, LLM24 performs a supplemental consistency check process for the business entity information (company information) as described above. Furthermore, LLM24 performs another ranking. Based on the results of the second ranking, LLM24 may select a company to request. Based on the first information I1, the second information I2, and Web information W, LLM24 may create a prompt and generate reasons for requesting the company to make a request based on the created prompt. LLM24 generates first generated information O1a. First generated information O1a may include the company to request and the reasons for requesting. The company to request included in first generated information O1a is an example of a partner included in the potential partners selected by the selection unit 60.

[0049] [1-3] About operation Figure 4 is a flowchart showing an example of the operation of the matching system 1 in the first embodiment.

[0050] The first acquisition unit 40 acquires at least one response piece of information R from the LLM 21 regarding the request of the first business entity X in response to the input of a natural language text T1 containing the first information I1 (step S1).

[0051] The second acquisition unit 50 acquires the second information I2 (step S2). The second information I2 may include the business content (technical content) and the product content handled by each of the multiple second business entities Y.

[0052] The first similarity calculation unit 61 calculates the first similarity A1 between the acquired response information R and each business content and each product content included in the second information I2 (step S3).

[0053] The first acquisition unit 40 determines whether the maximum value, average value, or median value of the first similarity A1 for each of the multiple second business entities Y is below a threshold (step S4). If the maximum value, average value, or median value of the first similarity A1 is below the threshold (see the YES route in step S4), new natural language text based on the response information R is input to the LLM 21 to obtain the new response information R (step S5). After step S5, the process returns to step S3. If the value is greater than the threshold (see the NO route in step S4), the first ranking unit 62 ranks each of the multiple second business entities Y (step S6).

[0054] The first ranking unit 62 selects business entities ranked within a predetermined rank as the first partner candidate C1 (step S7).

[0055] The collection unit 70 collects Web information W about each of the first candidate partners C1 on the network (step S8).

[0056] The second ranking unit 82 reranks at least a portion of the first partner candidate C1 based on the second similarity A2 between the Web information W collected for the first partner candidate C1 and the response information R (step S9).

[0057] The second ranking unit 82 selects a business entity that has been re-ranked within a predetermined rank as the second partner candidate C2 (step S10). In this case, the second partner candidate C2 is a partner that is included in the partner candidate selected from among multiple second business entities Y.

[0058] The output unit 90 outputs first generated information O1a, which includes the second partner candidate C2 and the reason for its selection (step S11), and the process ends.

[0059] Step S1 is an example of a process that obtains at least one response information R regarding a request from a language model that responds to the content of a natural language text T1 in response to an input of a natural language text T1 containing first information I1 regarding the request of a first business entity. Step S2 is an example of a process for obtaining second information I2, which includes at least one of the business activities and the products handled by each of several second business entities. Steps S3, S6, and S7 are examples of processes for selecting at least one potential partner for the first business entity from among multiple second business entities, based on a first similarity A1 between the response information R and at least one of the business content and product content included in the second information I2. Step S11 is an example of a process that outputs information about the partner companies included in the selected candidate partners.

[0060] According to the matching system 1 of the first embodiment, multiple business entities can be matched with each other based on potential elements related to the input content of the challenges of the first business entity X, which is entered by the user.

[0061] [2] Second Embodiment In the first embodiment, a matching system 1 was described that a business entity facing a problem uses to find other business entities (companies, etc.) that have the ability to solve that problem. However, the embodiments of the present invention are not limited to this.

[0062] [2-1] Functional Configuration This embodiment may have the same configuration as the matching system 1 of the first embodiment shown in Figure 1, except for the difference in the first information I1 and the response information R. Therefore, repeated explanations will be omitted, and the same reference numerals will be used to describe similar components.

[0063] [2-2] Regarding the processing details Figure 5 shows an example of the processing performed by the matching system 1 in the second embodiment. The matching system 1 in the second embodiment is used by the first business entity X to find business entities (companies, etc.) to which it can offer its selling points and products (target customers).

[0064] In this embodiment, the natural language text T1 includes first information I1. The first information I1 includes product details (selling points) or technical details provided by the first business entity X. The response information R includes information about business proposals (utilization business proposals) that utilize the product details or technical details provided by the first business entity X.

[0065] When a user inputs natural language text T1 containing sales points or product information, LLM21 generates a business plan. The generated business plan is obtained as response information R.

[0066] For example, if the first information I1 includes a product called "a tough, novel fiber material," then response information R from multiple perspectives regarding business applications for this product may be considered. For instance, response information R for a first business application in the general fiber industry may be obtained. Furthermore, response information R for a second business application in the aerospace industry may be obtained. In this way, multiple types of response information R from different perspectives may be obtained from LLM21.

[0067] As shown in Figure 1, the second acquisition unit 50 acquires multiple second business entities Y1,...,Y N Obtain second information I2 for each of the following: Multiple second business entities Y1,...,Y N The details of each of these businesses are examples of those described in Information 2, Section I2.

[0068] The selection unit 60 calculates the first similarity A1 between the proposed utilization project and the respective business content or product content of multiple second business entities.

[0069] If the first similarity score A1 is low due to reasons such as the information on the proposed business use obtained from LLM21 being abstract, cycle 41 of concretizing and refining the proposed business use may be executed. Cycle 41 may be repeated without user intervention until the first similarity score A1 exceeds a threshold.

[0070] The selection unit 60, based on the calculated first similarity A1, selects the second project entity Y1,...,Y N To rank them.

[0071] The selection unit 60 selects, for example, several business entities (companies) as candidates for business entities (recipient companies, etc.) to whom the first business entity X will provide its sales points and products.

[0072] LLM24 (Second Selection Unit 80) acquires Web information W from Web server 23 on the network. Using the acquired Web information W, LLM24 performs supplemental consistency verification processing of business entity information (company information) as described above. Furthermore, LLM24 performs ranking again. Based on the results of the re-ranking, LLM24 selects target companies to whom products and sales points should be provided. Based on the first information I1, second information I2, and Web information W, LLM24 may create prompts and output the reasons for providing the products to target companies and the sales strategy for selling to target companies based on the created prompts. The first generated information O1b, which includes target companies, reasons for providing the products, and sales strategies, is output by the output unit 90. The first generated information O1b is presented to the user through a monitor or the like. The target companies included in the first generated information O1b are examples of partner companies included in the partner candidate selected by the selection unit.

[0073] [2-3] Operation The operation example of the matching system 1 in the second embodiment is the same as in the first embodiment shown in Figure 4, except that the content of the first information I1 regarding the request is the product details (sales points) or technology details provided by the first business entity X, and the content of the response information R is information regarding a business plan (utilization business plan) that utilizes the product details or technology details provided by the first business entity X. Therefore, repeated explanations are omitted.

[0074] According to the matching system 1 of the second embodiment, multiple business entities can be matched with each other based on potential business proposals regarding the selling points or products of the first business entity X, which are entered by the user.

[0075] [3] Third Embodiment [3-1] Functional Configuration This embodiment may have the same configuration as the matching system 1 of the first embodiment shown in Figure 1, except that the matching system 1 has a RAG unit 29, an LLM 24b, and a recipient information DB 25. Therefore, repeated explanations will be omitted. The LLM 24b and the recipient information DB 25 will be explained together with the explanation of the processing content in [3-2].

[0076] [3-2] Regarding the processing details Figure 6 shows an example of the processing content of the matching system in the third embodiment. The matching system 1 of this embodiment has a process for receiving requests RQ from users who desire a more detailed sales strategy, a process for acquiring Web text WT, and a process for generating a sales strategy using RAG (Retrieval-Augmented Generation). The matching system 1 of this embodiment has a RAG unit 29, an LLM 24b, and a provider information DB 25 for storing Web text. RAG is a technology that improves the accuracy of responses by combining text generation using a large-scale language model with searching for external information.

[0077] Aside from these points, the matching system 1 of this embodiment is the same as that of the second embodiment, so repeated explanations will be omitted, and it will be described using the same reference numerals as in Figures 1 and 5.

[0078] The user learns about the first generated information O1b, which includes information about the recipient company, the reason for providing the information, and the sales strategy, through a monitor or similar means. If the user wishes to obtain a more specific sales strategy, they enter a request RQ regarding the sales strategy. The request RQ may be entered using GUI30 or similar in Figure 1.

[0079] The RAG unit 29 acquires Web text WTs from the network in response to a user request RQ. For example, the RAG unit 29 may vectorize at least one of the following: T1 containing the first information I1, the second information I2, and the first generated information O1b acquired by LLM 24, and acquire related documents as Web text WTs whose similarity to the vector is greater than or equal to a predetermined value. The RAG unit 29 may function as an external information collection unit that acquires Web text WTs. The acquired Web text WTs may be stored in the recipient information DB 25 for each business entity (company, etc.). Web text WTs are an example of external information, which is information from outside the matching system 1. The RAG unit 29 is an example of a text collection unit that collects text information on the network about the second partner candidate C2, which is a partner included in the first partner candidate C1 selected by the selection unit 60.

[0080] The RAG unit 29 may create a prompt by adding the retrieved web text (related document) to a query that includes at least one of the following: T1 containing the first information I1, the second information I2, and the first generated information O1b obtained by LLM24. LLM24b generates a sales strategy that is made concrete by the sales strategy contained in the first generated information O1b, while referring to the information in the web text and based on the newly obtained web text WT.

[0081] In one example, LLM24b uses at least one of the first information I1 and the second information L2 as a query (keyword, etc.) and extracts relevant portions from the web text according to the query. Based on the prompt (instructional information) created by adding the relevant portions to the first information I1, the second information I2, and the web information W, LLM24b may concretize the sales strategy.

[0082] Furthermore, LLM24b may use the RAG function to generate not only sales strategies but also reasons for making a request. In other words, LLM24b may generate second generated information Ob2, which is a concrete expression of the content of the first generated information O1b. The second generated information Ob2 is presented to the user through a monitor or the like. LLM24b is an example of a generation unit that generates at least one of a sales strategy for a potential partner (C1 or C2) and reasons for partnering, based on natural language text T1, second information I2, and text information (WT) collected by the text collection unit.

[0083] LLM24b may be the same LLM as LLM24, or it may be a separate LLM. In the example shown in Figure 6, the second generated information is shown as a trigger for a user request RQ, but this embodiment is not limited to this case. Regardless of whether or not there is a user request RQ, for example, LLM24 may concretize the sales strategy based on the Web text WT. In this case, the Web text WT acquired by LLM24 may be acquired as part of the Web information W. Also, if LLM24b itself has the function of the RAG unit 29, the RAG unit 29 does not need to be provided separately from LLM24b.

[0084] In the embodiment shown in Figure 6, a case is shown in which the content of the first generated information O1b in the second embodiment is materialized based on a Web text WT or the like to generate the second generated information Ob2. However, this embodiment is not limited to this case. The generation unit (i.e., LLM24b, etc.) may materialize the first generated information O1a in the first embodiment based on a Web text WT or the like to generate the second generated information O2a. [3-3] About operation Except for matters concerning the process explained using Figure 6, the operation of this embodiment is the same as in the case of the first embodiment shown in Figure 4, so the explanation of the operation example using a flowchart will be omitted.

[0085] According to the matching system 1 of the third embodiment, users can learn about sales strategies and other information, including information that has not been collected in advance.

[0086] [4] Fourth Embodiment In the first to third embodiments, a case was described in which natural language text containing first information I1 about the first business entity X is input to a large-scale language model (LLM21), and the first acquisition unit 40 acquires response information R from the LLM21. In the fourth embodiment, a case is shown in which natural language text containing second information I2 is also input to the large-scale language model, and response information is acquired from the large-scale language model.

[0087] [4-1] Functional Configuration Figure 7 shows an example of the configuration of the matching system 2 of the fourth embodiment. The matching system 2 is a matching system that matches multiple business entities with each other. The matching system 2 has a control unit 12 for controlling the matching system 2. The control unit 12 may be a processor. The control unit 12 includes a first GUI 30a, a second GUI 30b, a first acquisition unit 40a, a second acquisition unit 40b, and a selection unit 63. Furthermore, the control unit 12 may include a collection unit 71 and a second selection unit 83.

[0088] The first GUI30a is an example of an interface into which the first natural language text T1 is input. The first natural language text T1 contains the first information I1. The first information I1 contains multiple first business entities X1,...X M This includes information about the challenges each of (where M is a natural number greater than or equal to 2) faces.

[0089] The second GUI30b is an example of an interface into which the second natural language text T2 is input. The second natural language text T2 contains the second information I2. The second information I2 contains multiple second business entities Y1,...Y N This includes information about at least one of the product details (selling points) and technical details offered by each of the (N is a natural number greater than or equal to 2) entities.

[0090] The first natural language text T1 is input into LLM21a, a generative artificial intelligence model. Similarly, the second natural language text T2 is input into LLM21b. LLM21a and LLM21b may be the same large-scale language model, or they may be separate large-scale language models.

[0091] The first acquisition unit 40a acquires at least one first response information R1 relating to the issue from LLM21a in response to the input of a first natural language text T1 containing first information I1. The second acquisition unit 40b acquires at least one second response information R2 relating to at least one of product details and technical details from LLM21b in response to the input of a second natural language text T2 containing second information I2. Multiple first business entities X1,...X M The first response information R1 obtained for each of the following (hereinafter, sometimes simply referred to as multiple first business entities X) is: First business entity X1, ...X M It may be stored in the first DB27 in correspondence with multiple second business entities Y1,...,Y N The second response information R2 obtained for each of the following is for the second project entity Y1,...,Y N It may be stored in the second DB28 in correspondence with the first DB27. The first DB27 and the second DB28 may be used as the business entity DB22 in the first to third embodiments. The first natural language text T1 and the second natural language text T2 are examples of the first natural language text and the second natural language text, respectively, and the first response information R1 and the second response information R2 are examples of the first response information and the second response information, respectively.

[0092] The selection unit 63 includes a first similarity calculation unit 64 and a first ranking unit 65. The first similarity calculation unit 64 calculates a first similarity A1 for each pair of first response information R1 acquired by the first acquisition unit 40a and second response information R2 acquired by the second acquisition unit 40b. The first similarity calculation unit 64 is the same as the first similarity calculation unit 61 in the first embodiment.

[0093] The first ranking unit 65 ranks each pair of first response information R1 and each second response information R2 based on the first similarity A1 and calculates the first ranking information E1. Based on the ranking results, the first ranking unit 65 ranks multiple first business entities X1,...X M and multiple second business entities Y1,...,Y N From among these, we select a first potential partner pair P1 that is a good match for each other. The first potential partner pair P1 is just one example of a potential partner pair.

[0094] Furthermore, if the first similarity A1 between a given first response information R1 and a plurality of second response information R2 is below a threshold, the first acquisition unit 40a may execute cycle 41a, in which it inputs new natural language text based on the first response information R1 into the LLM 21a to acquire new first response information R1. Similarly, if the first similarity A1 between a given second response information R2 and a plurality of first response information R1 is below a threshold, the second acquisition unit 40b may execute cycle 41b, in which it inputs new natural language text based on the second response information R2 into the LLM 21b to acquire new second response information R2. The processing of each cycle 41a and 41b may be the same as in the first and second embodiments. Therefore, repeated explanations are omitted.

[0095] The collection unit 71 collects first Web information W1 and second Web information W2 on the network concerning the first business entity X and the second business entity Y, which are the first partner candidate pair P1 selected by the selection unit 63. The first Web information W1 and second Web information W2 may be collected from the Web server 23 on the network.

[0096] The second selection unit 83 may include a second similarity calculation unit 84 and a second ranking unit 85. The second similarity calculation unit 84 calculates the second similarity A2. The second similarity A2 may be one or more combinations of the similarity A2_1 between the first Web information W1 and the second Web information W2, the similarity A2_2 between the first Web information W1 and the second response information R2, and the similarity A2_3 between the second Web information W2 and the first response information R1. Note that the illustrations of similarity A2_1, A2_2, and A2_3 are omitted.

[0097] The second ranking unit 82 may select pairs of business entities ranked within a predetermined rank as second partner candidate pairs P2. The number of second partner candidate pairs P2 to be selected may be set as appropriate. The second selection unit 83 may generate first generated information O1c, which includes information about the second partner candidate pairs P2 and the reasons why the second partner candidate pairs P2 were selected.

[0098] The output unit 90 may output first generated information O1c, which includes information about the second candidate partner pair P2, as information about the partner pair included in the first candidate partner pair P1. The second candidate partner pair P2 is an example of a candidate partner pair. If the matching system 2 does not have a second selection unit 83, the output unit 90 may output information about the first candidate partner pair P1.

[0099] In the example shown in Figure 7, the case is shown in which the first acquisition unit 40a and the second acquisition unit 40b acquire both the first response information R1 and the second response information R2. However, the acquisition unit (first acquisition unit 40a or second acquisition unit 40b) may be configured to acquire at least one of the first response information R1 and the second response information R2. In this case, the configuration will be the same as in the first to third embodiments. Therefore, repeated explanations will be omitted.

[0100] [4-2] Regarding the processing details Since the processing details were explained in conjunction with the explanation of the functional configuration in Figure 7 [4-1], a separate explanation of the processing details using other diagrams will be omitted.

[0101] [4-3] Operation Since the operation has been explained in conjunction with the explanation of the functional configuration in [4-1] using Figure 7, the explanation of the operation example using a separate flowchart will be omitted.

[0102] According to the fourth embodiment, multiple first business entities X and multiple second business entities Y each register with the matching system 2 and can receive introductions to promising business entities as potential partners.

[0103] [5] Hardware Configuration Example Each process in the matching system 1 of the first to third embodiments and the matching system 2 of the fourth embodiment can be performed by executing a computer program on a computer.

[0104] Figure 8 shows examples of hardware configurations for Matching System 1 in the first to third embodiments and Matching System 2 in the fourth embodiment. Matching System 1 and Matching System 2 (hereinafter collectively referred to as Matching System 1) may be virtual servers (VMs) or physical servers. Furthermore, the functions of Matching System 1 may be implemented by one computer or by two or more computers. Accordingly, the matching program executed by the computer may be divided into multiple program modules. In addition, at least a part of the functions of Matching System 1 may be implemented using HW (Hardware) resources and NW (Network) resources provided by a cloud environment.

[0105] Figure 8 is a block diagram showing an example of the hardware (HW) configuration of a computer 100 that implements the functions of the matching system 1 according to the first to fourth embodiments. If multiple computers are used as HW resources to implement the functions of the matching system 1, each computer may have the HW configuration illustrated in Figure 1.

[0106] As shown in Figure 8, the computer 100 may, as an example of its hardware configuration, include a processor 101, memory 102, storage unit 103, IF (Interface) unit 104, IO (Input / Output) unit 105, and read unit 106.

[0107] The processor 101 is an example of an arithmetic processing unit that performs various control and calculations. The processor 101 may be connected to each block in the computer 100 via a bus 107 so as to be able to communicate with each other. The processor 101 may be a multiprocessor containing multiple processors, a multicore processor having multiple processor cores, or a configuration having multiple multicore processors.

[0108] Examples of processors 101 include integrated circuits (ICs) such as CPUs, MPUs, GPUs, APUs, DSPs, ASICs, and FPGAs. Note that two or more combinations of these integrated circuits may be used as processor 101. CPU stands for Central Processing Unit, MPU for Micro Processing Unit, GPU for Graphics Processing Unit, and APU for Accelerated Processing Unit. DSP stands for Digital Signal Processor, ASIC for Application Specific IC, and FPGA for Field-Programmable Gate Array.

[0109] For example, if the matching system 1 performs machine learning processing in addition to the matching processing according to one embodiment, the processor 101 may be a combination of a processing unit such as a CPU that performs the matching processing and an accelerator that performs the machine learning processing. Examples of accelerators include the GPU, APU, DSP, ASIC, or FPGA mentioned above.

[0110] Memory 102 is an example of hardware that stores various data and program information. Examples of memory 102 include volatile memory such as DRAM (Dynamic Random Access Memory) and non-volatile memory such as PM (Persistent Memory), or both.

[0111] The memory unit 103 is an example of hardware that stores various data and program information. Examples of memory units 103 include magnetic disk devices such as HDDs (Hard Disk Drives), semiconductor drive devices such as SSDs (Solid State Drives), and various storage devices such as non-volatile memory. Examples of non-volatile memory include flash memory, SCM (Storage Class Memory), and ROM (Read Only Memory).

[0112] The memory unit 103 may store one or more programs 108 (matching programs) that implement all or part of the various functions of the computer 100. The programs 108 may be program products.

[0113] For example, the processor 101 of the matching system 1 can realize the functions of the matching system 1 (the functions of the control unit 10 as illustrated in Figures 1 to 6, and the functions of the control unit 12 as illustrated in Figure 7) by loading the program 108 stored in the memory 102 and executing it. In other words, the program 108 may cause the computer to execute each process as shown in Figure 4, or it may cause the computer to execute each process of the matching method shown in the first to fourth embodiments.

[0114] The IF unit 104 is an example of a communication IF that controls connections and communications between the matching system 1 and various networks, including a network between the matching system 1 and devices not shown. Examples of devices include a computer such as a user terminal or server that provides data to the matching system 1, and a computer such as a server that outputs response information R as LLMs 21, 21a, 21b, 24, 24b, and 26 based on the data output from the matching system 1.

[0115] For example, the IF section 104 may include an adapter compliant with LAN (Local Area Network) such as Ethernet (registered trademark), or optical communication such as FC (Fibre Channel). The adapter may support wireless, wired, or both communication methods.

[0116] The program 108 may be downloaded from the network to the computer 100 via the communication interface and stored in the storage unit 103.

[0117] The I / O unit 105 may include either an input device or an output device, or both. Examples of input devices include keyboards, mice, and touch panels. Examples of output devices include monitors, projectors, and printers. The I / O unit 105 may also include a touch panel or the like that integrates an input device and a display device. The output device of the I / O unit 105 may be used to present information from partners to the user.

[0118] The reading unit 106 is an example of a reader that reads data and program information recorded on the recording medium 109. The reading unit 106 may include a connection terminal or device to which the recording medium 109 can be connected or inserted. Examples of the reading unit 106 include an adapter compliant with USB (Universal Serial Bus), a drive device for accessing a recording disk, and a card reader for accessing flash memory such as an SD card. The recording medium 109 may store a program 108, and the reading unit 106 may read the program 108 from the recording medium 109 and store it in the storage unit 103.

[0119] Examples of recording media 109 include non-temporary computer-readable recording media such as magnetic / optical discs and flash memory. Examples of magnetic / optical discs include flexible discs, CDs (Compact Discs), DVDs (Digital Versatile Discs), Blu-ray discs, and HVDs (Holographic Versatile Discs). Examples of flash memory include semiconductor memory such as USB memory and SD cards.

[0120] The hardware configuration of computer 100 described above is illustrative. Therefore, the addition or deletion of hardware within computer 100 (for example, adding or deleting arbitrary blocks), division, integration in any combination, or addition or deletion of buses may be performed as appropriate.

[0121] [5] Effects of the first to fourth embodiments According to the methods of the first to fourth embodiments, the matching system 1 for matching multiple business entities includes a first acquisition unit 40 that acquires at least one response information R regarding the request from a language model (LLM21) that responds to the content of natural language text T1 in response to the input of natural language text T1 containing first information regarding the request of the first business entity X, and a plurality of second business entities Y1,...,Y NA second acquisition unit 50 acquires second information I2 which includes at least one of the business activities and product contents of each of the first business entities X, and based on the first similarity between the response information R acquired by the first acquisition unit 40 and at least one of the business activities and product contents of each of the first business entities X acquired by the second acquisition unit 50, at least one candidate partner (first candidate partner C1) with multiple second business entities Y1,...,Y N The system includes a selection unit 60 that selects from among the candidates, and an output unit 90 that outputs information about the partner included in the candidate partner C1 selected by the selection unit 60.

[0122] This allows for matching multiple business entities based on potential elements related to the content of natural language text T1 entered by the user. Furthermore, it enables the selection of potential partners from a wide range of perspectives based on response information R from multiple viewpoints.

[0123] As shown in the first embodiment, the first information I1 may include information about a problem faced by the first business entity X, and the response information R may include information about solutions to the problem. This allows multiple business entities to be matched based on potential solutions related to the content of the problem information. The first business entity X, facing a problem, can search for potential clients capable of solving that problem in a single search, based on multiple perspectives.

[0124] As shown in the second embodiment, the first information I1 includes at least one of the product content and technology content provided by the first business entity X, and the response information R includes information on business proposals that utilize at least one of the product content and technology content provided by the first business entity X. This allows multiple business entities to be matched based on potential business proposals that utilize at least one of the product content and technology content provided. The first business entity X, which has selling points and products, can search for potential customers to whom it provides selling points and products based on multiple perspectives in a single search.

[0125] The selection unit 60 selected multiple potential partners from multiple second business entities Y1,...,Y N The system selects from among these, and further includes a collection unit 70 that collects Web information W on the network regarding each of the multiple first partner candidate C1 selected based on the first similarity A1, and a second selection unit that selects a partner (second partner candidate C2) from among multiple second business entities based on the second similarity A2 between the information collected by the collection unit 70 and the response information R acquired by the first acquisition unit 40. This makes it possible to acquire new information that has arisen after the storage of the second information I2, supplement the business entity information, and confirm consistency.

[0126] Furthermore, as shown in the third embodiment, the matching system 1 includes a text collection unit that collects text information on the network about the partner candidates selected by the selection units (60, 80) (for example, second partner candidate C2), and a generation unit that generates at least one of a sales strategy for the partner candidate and reasons for partnering based on natural language text, second information, and the text information collected by the text collection unit. This allows the user to learn about sales strategies, etc., including information that has not been collected in advance. Therefore, the user can learn about sales strategies, etc., in more depth.

[0127] The first acquisition unit 40 inputs new natural language text based on response information R into the LLM 21 to acquire new response information when the first similarity A1 is below a threshold. The selection unit 60 selects a first partner candidate C1 for the first business entity X from among the second business entities Y based on the first similarity A1 between the new response information acquired by the first acquisition unit 40 and the respective business content and product content included in the second information I2. In this way, the matching accuracy between multiple business entities can be improved by acquiring new response information R so that appropriate matching processing is possible when the first similarity A1 is below a threshold.

[0128] As shown in the fourth embodiment, the matching system 2 for matching multiple business entities is a multiple first business entity X1,...,X MA first acquisition unit 40a acquires at least one first response information R1 concerning the issues from an LLM 21a that responds to the content of the first natural language text T1 in response to the input of a first natural language text T1 containing first information I1 concerning the issues each of the second entities Y1,...,Y N A second acquisition unit 40b acquires at least one second response information R2 relating to at least one of the product content and technical content from a language model that responds to the content of the second natural language text T2 in response to the input of a second natural language text T2 containing second information I2 relating to at least one of the product content and technical content provided by each of the first acquisition unit 40a and the second response information R2 acquired by the second acquisition unit 40b, based on a first similarity between the first response information R1 acquired by the first acquisition unit 40a and the second response information R2 acquired by the second acquisition unit 40b. M and multiple second business entities Y1,...,Y N The system includes a selection unit (selection unit 63 and second selection unit 83) that selects a first partner candidate pair P1 that matches each other from among them, and an output unit 90 that outputs information about the partnership pairs included in the partner candidate pair selected by the selection unit. This allows for multiple first business entities X1,...,X M and multiple second business entities Y1,...,Y N However, each can register with Matching System 2 and receive introductions to promising business entities that could become partners.

[0129] It should be noted that the embodiments described above are merely illustrative, and various modifications are possible. The response information R from LLM21 in the first embodiment may be stored in the business entity DB22 of the second embodiment as second information I2 associated with its first business entity X. Conversely, the response information R (proposed utilization business) from LLM21 in the second embodiment may be stored in the business entity DB22 of the first embodiment as second information I2 associated with its first business entity X. The business entity DB22 may be updated in conjunction with these processes. [Explanation of Symbols]

[0130] 1,2 Matching System 10,12 Control Unit 21,21a,21b,24,24b,26 LLM 22 Business entity DB 23 Web Servers 25 Provided information DB 27. 1st DB 28 2nd DB 29 RAG section 30, 30a, 30b GUI 40,40a 1st acquisition part 40b,50 2nd acquisition part 60,63 Selection Department 61,64 First similarity calculation unit 62,65,81,84 Second similarity calculation unit 70, 71 Collection Department 80,83 2nd selection section 82,85 Second Ranking Section 90 Output section

Claims

1. A matching system that connects multiple business entities, A first acquisition unit acquires at least one response piece of information relating to a request from a language model that responds to the content of a natural language text in response to the input of a natural language text containing first information relating to a request from a first business entity. A second acquisition unit that acquires second information including at least one of the business content and the product content handled by each of several second business entities, A selection unit selects at least one candidate partner for the first business entity from among the plurality of second business entities based on a first similarity between the response information obtained by the first acquisition unit and at least one of the business content and product content included in the second information obtained by the second acquisition unit, A matching system comprising: an output unit that outputs information about partner candidates selected by the selection unit; and an output unit that outputs information about partner candidates included in the partner candidate selected by the selection unit.

2. The aforementioned first information includes information relating to the challenges faced by the aforementioned first business entity, The response information includes information relating to a solution to the problem, The matching system according to claim 1.

3. The first information includes at least one of the product details and technical details provided by the first business entity. The response information includes information relating to a business plan that utilizes at least one of the product content and the technology content provided by the first business entity. The matching system according to claim 1.

4. The aforementioned selection unit selects multiple potential partners from among the aforementioned multiple second business entities. Furthermore, a collection unit collects information on each of the multiple potential partners selected based on the first similarity score over the network, A matching system according to any one of claims 1 to 3, comprising: a second selection unit that selects the partner from among the plurality of partner candidates based on a second similarity between the information collected by the collection unit and the response information acquired by the first acquisition unit.

5. Furthermore, a text collection unit collects text information about the aforementioned partner over the network, A generation unit generates at least one of a sales strategy for the potential partner and reasons for forming a partnership, based on the natural language text, the second information, and the text information collected by the text collection unit. The matching system according to claim 4, comprising:

6. The first acquisition unit, when the first similarity is below a threshold, inputs new natural language text based on the response information into the language model to acquire new response information. The selection unit selects a candidate partner for the first business entity from among the second business entities based on the new response information obtained by the first acquisition unit and the first similarity between each of the business contents and each of the product contents included in the second information. The matching system according to any one of claims 1 to 3.

7. A matching system that connects multiple business entities, A first acquisition unit acquires at least one first response information relating to the issues from a language model that responds to the content of the first natural language text in response to the input of a first natural language text containing first information relating to the issues each of a plurality of first business entities, A second acquisition unit acquires at least one second response information relating to at least one of the product content and technical content from a language model that responds to the content of the second natural language text in response to the input of a second natural language text containing second information relating to at least one of the product content and technical content provided by each of a plurality of second business entities, A selection unit selects at least one candidate partner pair from among a plurality of first business entities and a plurality of second business entities based on the similarity between the first response information acquired by the first acquisition unit and the second response information acquired by the second acquisition unit, A matching system comprising: an output unit that outputs information about partner pairs included in the partner candidate pairs selected by the selection unit.

8. A matching method that connects multiple business entities, From a language model that responds to the content of a natural language text in response to the input of a natural language text containing first information relating to the requests of a first business entity, at least one response piece of information relating to the requests is obtained. We obtain second information that includes at least one of the business activities and the products handled by each of several second business entities. Based on the first similarity between the response information and at least one of the business content and product content included in the second information, at least one candidate partner for the first business entity is selected from among the plurality of second business entities. Output information about the partners included in the selected candidate partners. A matching method in which a computer performs the processing.

9. This is a matching program that connects multiple business entities, From a language model that responds to the content of a natural language text in response to the input of a natural language text containing first information relating to the requests of a first business entity, at least one response piece of information relating to the requests is obtained. We obtain second information that includes at least one of the business activities and the products handled by each of several second business entities. Based on the first similarity between the response information and at least one of the business content and product content included in the second information, at least one candidate partner for the first business entity is selected from among the plurality of second business entities. Output information about the partners included in the selected candidate partners. A matching program that has a computer perform a process.

Images