Business support system, business support method, and data management program

Abstract

To provide an arrangement for supporting a work of replying to an inquiry from an authority.SOLUTION: The present invention is directed to a business support system having a data management apparatus for generating document data with utilizing a language model. The system has a document data acquiring unit for acquiring document data referred by the authority, and an agent unit for, if reference contents of a reference item is extracted from the document data, acquiring, from at least one retrieval destination, auxiliary information for use in generation of reply contents for the reference contents to transmit a prompt including the reference contents and the auxiliary information to the language model. The language model refers to the auxiliary information included in the prompt to generate the reply contents to the reference contents.SELECTED DRAWING: Figure 49B

Description

[Technical Field]

[0001] The present disclosure relates to a business support system, a business support method, and a data management program. [Background technology]

[0002] When manufacturing and selling pharmaceuticals, etc., it is necessary to first submit a pharmaceutical manufacturing and marketing approval application to the appropriate authorities and obtain approval from the Minister of Health, Labour and Welfare. Approval is also required overseas from authorities such as the U.S. Food and Drug Administration and the European Medicines Agency. Furthermore, if an inquiry (authority inquiry) is made by the authorities regarding the contents of the application during the drug review process, the applicant is required to respond appropriately to the inquiry. Furthermore, even outside of the drug review process, if a pharmaceutical development company receives an inquiry from the authorities (hereinafter referred to as a pharmaceutical affairs inquiry), the pharmaceutical development company is required to respond appropriately to the inquiry. In the following explanation, "authority inquiries" and their responses are considered to include pharmaceutical affairs inquiries and their responses. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2016 / 157467 [Non-patent literature]

[0004] [Non-Patent Document 1] Pharmaceuticals and Medical Devices Agency, "Approval Review Operations (Applications, Reviews, etc.)," ​​[online], [Retrieved October 21, 2022], Internet<URL:https: / / www.pmda.go.jp / review-services / drug-reviews / 0001.html> Summary of the Invention [Problem to be solved by the invention]

[0005] Here, the work of responding to inquiries from authorities has traditionally been carried out based on the experience of the worker (the person who actually prepares the response to each inquiry). Specifically, the worker would, for example, rely on their own memory to search for the responses to similar inquiries from the past, and then refer to them to prepare the response to each inquiry.

[0006] On the other hand, if a system could be provided for such business processes that allows workers to easily search for similar past cases of inquiries from authorities, it would improve the efficiency of workers' work and support their response work.

[0007] The present disclosure aims to provide a mechanism for supporting the process of responding to inquiries from authorities. [Means for solving the problem]

[0008] According to one aspect, the business support system includes: A business support system having a data management device that generates document data using a language model, a document data acquisition unit for acquiring document data of an inquiry from an authority; an agent unit that, when an inquiry content of an inquiry item is extracted from the document data, acquires auxiliary information used in generating an answer content to the inquiry content from at least one search source, and transmits a prompt including the inquiry content and the auxiliary information to the language model; The language model generates an answer to the query by referring to auxiliary information included in the prompt. [Effects of the Invention]

[0009] According to the present disclosure, a mechanism for supporting the work of responding to inquiries from authorities can be provided. [Brief explanation of the drawings]

[0010] [Figure 1]FIG. 1 is a first diagram illustrating an example of a system configuration of a business support system. [Figure 2] FIG. 2 is a second diagram illustrating an example of the system configuration of the business support system. [Figure 3] FIG. 3 is a diagram illustrating an example of the hardware configuration of the data management device and the server device. [Figure 4] FIG. 4 is a diagram showing an example of the authority inquiry document data. [Figure 5] FIG. 5 is a first diagram illustrating an example of the functional configuration of the management unit of the data management device. [Figure 6] FIG. 6 is a diagram showing a specific example of processing by the dividing unit included in the management unit of the data management device. [Figure 7] FIG. 7 is a diagram showing a specific example of the processing of the distribution unit and collection unit included in the management unit of the data management device. [Figure 8] FIG. 8 is a diagram showing a specific example of the process of the document data generation unit included in the management unit of the data management device. [Figure 9] FIG. 9 is a diagram showing a specific example of processing by the attribute information extraction unit included in the management unit of the data management device. [Figure 10] FIG. 10 is a diagram illustrating an example of the registration data. [Figure 11] FIG. 11 is a first diagram illustrating an example of a functional configuration of a search unit of a server device. [Figure 12] FIG. 12 is a first diagram illustrating an example of a search screen displayed on the worker terminal. [Figure 13] FIG. 13 is a first sequence diagram showing the flow of processing in the accumulation phase by the business support system. [Figure 14] FIG. 14 is a first sequence diagram showing the flow of processing in the search phase by the business support system. [Figure 15] FIG. 15 is a third diagram illustrating an example of the system configuration of the business support system. [Figure 16] FIG. 16 is a fourth diagram illustrating an example of the system configuration of the business support system. [Figure 17]FIG. 17 is a diagram illustrating an example of the functional configuration of the learning unit of the learning device. [Figure 18] FIG. 18 is a diagram illustrating an example of the functional configuration of the management unit of the data management device and the functional configuration of the prediction unit of the prediction device. [Figure 19] FIG. 19 is a sequence diagram showing the flow of processing in the learning phase by the business support system. [Figure 20] FIG. 20 is a second sequence diagram showing the flow of processing in the accumulation phase by the business support system. [Figure 21] FIG. 21 is a fifth diagram illustrating an example of the system configuration of the business support system. [Figure 22] FIG. 22 is a second diagram illustrating an example of the functional configuration of the search unit of the server device. [Figure 23] FIG. 23 is a second diagram illustrating an example of the search screen displayed on the worker terminal. [Figure 24] FIG. 24 is a first diagram illustrating a specific example of the processing of the first vectorization unit, the separation unit, and the second vectorization unit included in the search unit of the server device. [Figure 25] FIG. 25 is a second sequence diagram showing the flow of processing in the search phase by the business support system. [Figure 26] FIG. 26 is a third diagram illustrating an example of the functional configuration of the search unit of the server device. [Figure 27] FIG. 27 is a third diagram illustrating an example of a search screen displayed on the worker terminal. [Figure 28] FIG. 28 is a sixth diagram illustrating an example of the system configuration of the business support system. [Figure 29] FIG. 29 is a diagram for explaining an outline of the process in the feedback phase by the business support system. [Figure 30] FIG. 30 is a diagram illustrating an example of a functional configuration of an analysis unit of the analysis device. [Figure 31] FIG. 31 is a diagram showing a specific example of processing by a learning unit included in the analysis unit of the analysis device. [Figure 32]FIG. 32 is a second diagram illustrating a specific example of the processing of the first vectorization unit, the separation unit, and the second vectorization unit included in the search unit of the server device. [Figure 33] FIG. 33 is a sequence diagram showing the flow of processing in the feedback phase by the business support system. [Figure 34] FIG. 34 is a seventh diagram illustrating an example of the system configuration of the business support system. [Figure 35] FIG. 35 is a first diagram illustrating an example of an LM training dataset. [Figure 36] FIG. 36 is a first sequence diagram showing the flow of processing in the LM learning phase by the business support system. [Figure 37A] FIG. 37A is an eighth diagram illustrating an example of the system configuration of the business support system. [Figure 37B] FIG. 37B is a ninth diagram illustrating an example of the system configuration of the business support system. [Figure 38] FIG. 38 is a second diagram illustrating an example of the LM training dataset. [Figure 39A] FIG. 39A is a second sequence diagram showing the flow of processing in the LM learning phase by the business support system. [Figure 39B] FIG. 39B is a third sequence diagram showing the flow of processing in the LM learning phase by the business support system. [Figure 40] FIG. 40 is a tenth diagram illustrating an example of the system configuration of the business support system. [Figure 41] FIG. 41 is a first sequence diagram showing the flow of processing in the generation phase by the business support system. [Figure 42] FIG. 42 is an eleventh diagram illustrating an example of the system configuration of the business support system. [Figure 43] FIG. 43 is a third diagram illustrating an example of the LM training dataset. [Figure 44] FIG. 44 is a fourth sequence diagram showing the flow of processing in the LM learning phase by the business support system. [Figure 45] FIG. 45 is a twelfth diagram illustrating an example of the system configuration of the business support system. [Figure 46] FIG. 46 is a sequence diagram showing the flow of processing in the pharmaceutical application phase by the business support system. [Figure 47] FIG. 47 is a thirteenth diagram illustrating an example of the system configuration of the business support system. [Figure 48A] FIG. 48A is a first sequence diagram showing the flow of processing in the pharmaceutical affairs inquiry phase by the business support system. [Figure 48B] FIG. 48B is a second sequence diagram showing the flow of processing in the pharmaceutical affairs inquiry phase by the business support system. [Figure 49A] FIG. 49A is a fourteenth diagram illustrating an example of the system configuration of the business support system. [Figure 49B] FIG. 49B is a fifteenth diagram illustrating an example of the system configuration of the business support system. [Figure 50A] FIG. 50A is a first sequence diagram showing the flow of processing in the search phase and the generation phase by the business support system. [Figure 50B] FIG. 50B is a second sequence diagram showing the flow of processing in the search phase and the generation phase by the business support system. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, each embodiment will be described with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant description will be omitted.

[0012] [First embodiment] <System configuration of business support system> First, a system configuration of a business support system according to the first embodiment will be described. In the first embodiment, the business support system executes processing in a storage phase and processing in a search phase, and therefore, the system configuration will be described here for each phase.

[0013] (1) Business support system for the accumulation phase Figure 1 is the first diagram showing an example of the system configuration of the business support system, and shows the system configuration in the accumulation phase. The accumulation phase is a phase that is executed in parallel with the work of responding to inquiries from authorities, and is a phase in which a database is built to make it easy to search for past cases of similar inquiries from authorities.

[0014] 1, the business support system 100 in the accumulation phase includes an authority terminal 110, a data management device 120, worker terminals 130 and 140, and a server device 150. In the business support system 100, the authority terminal 110 and the data management device 120 are communicatively connected via an external network 160. In addition, in the business support system 100, the data management device 120, the worker terminals 130 and 140, and the server device 150 are communicatively connected via an internal network 170.

[0015] The authority terminal 110 is a terminal operated by the person in charge 111 who has accepted an application for approval to manufacture and sell a pharmaceutical product, etc., when making an inquiry (authority inquiry) to the applicant 122 regarding the contents of the application. When operated by the person in charge 111, the authority terminal 110 transmits, for example, authority inquiry document data (document data sent by the authority when making an authority inquiry) to the data management device 120 via the external network 160. Note that in this embodiment, the authority inquiry by the person in charge 111 is described as being made by sending the authority inquiry document data in the form of electronic data, but the method of the authority inquiry by the person in charge 111 is not limited to this. For example, the authority inquiry may also be made by facsimile transmission. In this case, however, the applicant 122 digitizes the authority inquiry document, and the data management device 120 acquires the digitized authority inquiry document data.

[0016] In addition, in response to sending the authority inquiry document data to the data management device 120, the authority terminal 110 receives input document data (document data in which the response content has been input in response to the authority inquiry document data) from the data management device 120 via the external network 160.

[0017] A data management program is installed in the data management device 120, and the data management device 120 functions as a management unit 121 by executing the program in the accumulation phase.

[0018] The management unit 121 operates based on operational instructions from the applicant 122. Specifically, the management unit 121 receives authority inquiry document data transmitted from the authority terminal 110, and generates multiple pieces of divided document data by dividing the received authority inquiry document data by inquiry subject. The management unit 121 also transmits the generated multiple pieces of divided document data to the corresponding worker terminals 130, 140, etc.

[0019] In addition, in response to sending the generated multiple divided document data to the corresponding worker terminals 130, 140, etc., the management unit 121 collects input divided document data from the worker terminals 130, 140, etc., in which the response content to each of the multiple inquiry items has been input.

[0020] Furthermore, the management unit 121 generates input document data by converting the collected input divided document data into a single file format, and transmits the generated input document data to the authority terminal 110 via the external network 160. Note that, in this embodiment, the response to the authority's inquiry is described as being made by transmitting the input document data, which is electronic data, but the method of response by the applicant 122 is not limited to this. For example, there may be cases where the response to the authority's inquiry is made by facsimile transmission.

[0021] Furthermore, management unit 121 stores registration data including divided document data, input divided document data, input document data, etc. in database 151 of server device 150. When storing the registration data, management unit 121 stores the registration data in association with keywords assigned by applicant 122 to facilitate search. Keywords are words or phrases that are associated with the inquiry content and response content included in the input divided document data.

[0022] Worker terminals 130, 140 receive the divided document data transmitted from data management device 120 and display the inquiry content. In addition, in response to displaying the inquiry content, worker terminals 130, 140 accept the response content input by workers 131, 141 and generate input divided document data. Furthermore, worker terminals 130, 140 transmit the generated input divided document data to data management device 120.

[0023] The server device 150 receives the registration data transmitted from the data management device 120 and stores it in the database 151. In the accumulation phase, the server device 150 stores the registration data in the database 151 every time the data management device 120 transmits input document data to the authority terminal 110.

[0024] (2) Business support system for the search phase Figure 2 is a second diagram showing an example of the system configuration of the business support system, illustrating the system configuration in the search phase. The search phase is executed in parallel with the work of responding to inquiries from authorities, and is the phase in which workers search a database for past cases of similar inquiries from authorities in order to create a response to the inquiry.

[0025] The difference from the system configuration in the accumulation phase is that in the case of the business support system 200 in the search phase, the server device 150 functions as a search unit 201.

[0026] A search program is installed in the server device 150, and the server device 150 functions as a search unit 201 by executing the program in the search phase.

[0027] The search unit 201 provides a search screen to the worker terminals 130 and 140. In response to providing the search screen, when a search request is received from the worker terminals 130 and 140, the search unit 201 searches the database 151 and acquires registered data that matches the search conditions included in the search request. The acquired registered data is displayed on the search screen as a search result.

[0028] 2, the system configuration differs from that in the accumulation phase in that the worker terminals 130 and 140 in the search phase of the task support system 200 function as search terminals.

[0029] Specifically, worker terminals 130, 140 access server device 150 based on operation instructions from workers 131, 141, and display a search screen provided by server device 150. Furthermore, worker terminals 130, 140 accept input of search conditions by workers 131, 141, and when a search instruction is input by workers 131, 141, send a search request including the search conditions to server device 150. Furthermore, in response to sending the search request, worker terminals 130, 140 display a search screen including search results provided by server device 150.

[0030] <Data management device and server device hardware> Next, a description will be given of the hardware configurations of the data management device 120 and the server device 150 among the devices constituting the task support systems 100 and 200. Fig. 3 is a diagram showing an example of the hardware configurations of the data management device and the server device.

[0031] 3, the data management device 120 includes a processor 301, a memory 302, an auxiliary storage device 303, an operation device 304, a display device 305, a communication device 306, and a drive device 307. The hardware components included in the data management device 120 are connected to each other via a bus 308.

[0032] The processor 301 has various arithmetic devices such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), etc. The processor 301 reads various programs (for example, a data management program, etc.) into the memory 302 and executes them.

[0033] The memory 302 has a main storage device such as a ROM (Read Only Memory), a RAM (Random Access Memory), etc. The processor 301 and the memory 302 form a so-called computer, and the processor 301 executes various programs read onto the memory 302, thereby enabling the computer to realize various functions.

[0034] The auxiliary storage device 303 stores various programs and various information used when the processor 301 executes the various programs.

[0035] The operation device 304 is used when the applicant 122 inputs various operation instructions to the data management device 120. The display device 305 displays the results of processing in the data management device 120.

[0036] The communication device 306 is a communication device for communicating with the authority terminal 110 via the external network 160 or for communicating with the worker terminals 130 and 140 and the server device 150 via the internal network 170 .

[0037] The drive device 307 is a device for loading a storage medium 310. The storage medium 310 here includes media that record information optically, electrically, or magnetically, such as a CD-ROM, a flexible disk, or a magneto-optical disk. The storage medium 310 may also include semiconductor memory that records information electrically, such as a ROM or flash memory.

[0038] The various programs to be installed in the auxiliary storage device 303 are installed, for example, by setting the distributed storage medium 310 in the drive device 307 and reading the various programs recorded on the storage medium 310 by the drive device 307. Alternatively, the various programs to be installed in the auxiliary storage device 303 may be installed by being downloaded from the external network 160 via the communication device 306.

[0039] 3, the server device 150 includes a processor 321, a memory 322, an auxiliary storage device 323, a connection device 324, a communication device 325, and a drive device 326. The hardware components included in the server device 150 are connected to each other via a bus 327.

[0040] The hardware components of the server device 150 are generally the same as those of the data management device 120. The storage medium 333 set in the drive device 326 of the server device 150 is also the same as the storage medium 310 set in the drive device 307 of the data management device 120. The difference from the hardware configuration of the data management device 120 is that the operation device 331 and the display device 332 of the server device 150 are connected to the server device 150 via the connection device 324. The processor 321 of the server device 150 reads and executes a search program on the memory 322. The database 151 of the server device 150 is also implemented in the auxiliary storage device 323.

[0041] <Explanation of government inquiry document data> Next, a specific example of the authority inquiry document data transmitted from the authority terminal 110 will be described. FIG. 4 is a diagram showing an example of the authority inquiry document data. As shown in FIG. 4, the authority inquiry document data 400 includes a plurality of pairs of an inquiry item and an inquiry content from the authority in response to the application content of the pharmaceutical manufacturing and sales approval application submitted by the applicant 122. The example in FIG. 4 shows that the authority inquiry document data 400 includes three pairs of an inquiry item and an inquiry content (three pairs of inquiry items 1 to 3 and an inquiry content indicated by reference numerals 410, 420, and 430). The number of pairs of an inquiry item and an inquiry content included in the authority inquiry document data 400 is arbitrary, and in this embodiment, the description will be given assuming that the data includes a plurality of pairs.

[0042] <Functional configuration of the management unit of the data management device> Next, the functional configuration of the management unit 121 of the data management device 120 will be described. FIG. 5 is a first diagram showing an example of the functional configuration of the management unit of the data management device. As described above, in the accumulation phase, the data management device 120 functions as the management unit 121. Furthermore, as shown in FIG. 5, the management unit 121 further Document data acquisition unit 501, A user interface unit 502 (an example of a third acquisition unit), ·Divided part 503, ·Distribution section 504, ·Attribute information extraction unit 505, a collection unit 506 (an example of a first acquisition unit); a document data generation unit 507 (an example of a second acquisition unit); Document data transmission unit 508, Storage unit 509, Includes.

[0043] The document data acquisition unit 501 acquires the authority inquiry document data sent from the authority terminal 110 (or, if the authority inquiry is made by facsimile transmission, the digitized authority inquiry document data) and notifies the division unit 503.

[0044] The division unit 503 divides the authority inquiry document data notified by the document data acquisition unit 501 into inquiry items to generate multiple pieces of divided document data. The division unit 503 also notifies the distribution unit 504 of the multiple pieces of divided document data that have been generated.

[0045] The distribution unit 504 transmits the multiple pieces of divided document data notified by the division unit 503 to the worker terminals 130, 140, etc. The destinations to which the distribution unit 504 transmits the multiple pieces of divided document data may be determined according to a predetermined rule, or may be determined based on an operational instruction from the applicant 122.

[0046] The collection unit 506 collects the plurality of pieces of input divided document data transmitted from the worker terminals 130, 140, etc. in response to the transmission of the plurality of pieces of divided document data by the distribution unit 504. The collection unit 506 also notifies the document data generation unit 507 and the storage unit 509 of the collected plurality of pieces of input divided document data.

[0047] Document data generation unit 507 generates input document data by combining multiple pieces of input divided document data notified by collection unit 506 into a single file format. Specifically, document data generation unit 507 generates input document data by combining multiple pieces of input divided document data in text format, for example, into a single piece of document data and converting it into a predetermined file format such as PDF. Document data generation unit 507 also notifies document data transmission unit 508 and storage unit 509 of the generated input document data.

[0048] Document data transmission unit 508 transmits the input document data notified by document data generation unit 507 to authority terminal 110 (or transmits it by facsimile to the authority).

[0049] The attribute information extraction unit 505 extracts attribute information contained in the authority inquiry document data acquired by the document data acquisition unit 501. The attribute information extraction unit 505 also extracts attribute information contained in the input document data generated by the document data generation unit 507. The attribute information extraction unit 505 notifies the storage unit 509 of the extracted attribute information. The attribute information extracted by the attribute information extraction unit 505 will be described in detail later.

[0050] The user interface unit 502 displays the input divided document data notified to the storage unit 509 by the collection unit 506 to the applicant 122. The user interface unit 502 also acquires keywords input by the applicant 122 in response to displaying the input divided document data. The user interface unit 502 also notifies the storage unit 509 of the acquired keywords. The keywords input by the applicant 122 are used as part of the search conditions in the search phase.

[0051] The storage unit 509 stores the registration data in the database 151 of the server device 150. The registration data includes: Input divided document data notified by the collection unit 506, Input document data notified by the document data generation unit 507, Attribute information notified by the attribute information extraction unit 505, Keywords notified by the user interface unit 502, etc. are included.

[0052] <Specific examples of processing by each part of the data management device> Next, a specific example of the processing of each unit (here, division unit 503, distribution unit 504, collection unit 506, document data generation unit 507, and attribute information extraction unit 505) included in management unit 121 of data management device 120 will be described.

[0053] (1) Specific example of processing by the division unit included in the management unit of the data management device First, a specific example of the processing of the dividing unit 503 included in the management unit 121 of the data management device 120 will be described. Fig. 6 is a diagram showing a specific example of the processing of the dividing unit included in the management unit of the data management device.

[0054] As shown in Fig. 6, in the case of authority inquiry document data 400 containing three pairs of inquiry items and inquiry contents (three pairs of inquiry items 1 to 3 and inquiry contents indicated by reference numerals 410, 420, and 430), the data is divided into three pieces of divided document data 610, 620, ... by dividing unit 503. Note that in the example of Fig. 6, due to space limitations, only two pieces of divided document data 610 and 620 are shown.

[0055] 6, divided document data 610 includes one inquiry, its corresponding inquiry content, and one reply content (blank). Similarly, divided document data 620 includes one inquiry, its corresponding inquiry content, and one reply content (blank).

[0056] (2) Specific examples of the processing of the distribution unit and collection unit included in the management unit of the data management device Next, a specific example of the processing of the distribution unit 504 and collection unit 506 included in the management unit 121 of the data management device 120 will be described. Fig. 7 is a diagram showing a specific example of the processing of the distribution unit and collection unit included in the management unit of the data management device.

[0057] As shown in FIG. 7, the divided document data 610, 620 sent by the distribution unit 504 to the worker terminals 130, 140, etc. each includes one inquiry item, its corresponding inquiry content, and one answer content (blank).

[0058] 7, each of the input divided document data 710, 720 collected by the collection unit 506 from the worker terminals 130, 140, etc., includes one inquiry item, its corresponding inquiry content, and one answer content. However, the one answer content here refers to the answer content that has already been input by the worker 131, 141, etc.

[0059] (3) Specific example of processing by the document data generation unit included in the management unit of the data management device Next, a specific example of the processing of document data generation unit 507 included in management unit 121 of data management device 120 will be described. Fig. 8 is a diagram showing a specific example of the processing of document data generation unit included in the management unit of the data management device.

[0060] As shown in FIG. 8, input divided document data 710 and 720 collected by collection unit 506 are put into one file format by document data generation unit 507, and input document data 800 is generated.

[0061] In addition, the example in Figure 8 is Input divided document data 710 (including one inquiry item, the corresponding inquiry content, and one reply content), The input divided document data 720 (including one inquiry item, the corresponding inquiry content, and one reply content), are arranged consecutively on one page. However, the arrangement method of the input divided document data in the input document data 800 is arbitrary, and multiple input divided document data may be arranged on one page, or only one input divided document data may be arranged on one page.

[0062] (4) Specific example of processing by the attribute information extraction unit included in the management unit of the data management device Next, a specific example of the processing of the attribute information extraction unit 505 included in the management unit 121 of the data management device 120 will be described. Fig. 9 is a diagram showing a specific example of the processing of the attribute information extraction unit included in the management unit of the data management device.

[0063] As shown in FIG. 9, the attribute information extraction unit 505 extracts attribute information 900 from the authority inquiry document data 400 acquired by the document data acquisition unit 501 and the input document data 800 generated by the document data generation unit 507.

[0064] The example in Figure 9 shows that the time of inquiry, the disease area of ​​the inquiry, the inquired event, the category of the inquiry, the item of the inquiry, the time of response, etc. are extracted as attribute information 900. The items extracted as attribute information 900 are arbitrary, and the attribute information may include at least one piece of information selected from the time of inquiry, the disease area of ​​the inquiry, the inquired event, the category of the inquiry, the item of the inquiry, and the time of response. The disease area may also include the area of ​​responsibility of the regulatory review unit and the intended use.

[0065] The inquiry time refers to the date on which the authority inquiry document data 400 was received from the authority terminal 110. The disease area of ​​the inquiry content refers to the disease area to which the inquiry content belongs. The disease area is assumed to be predefined, and in this embodiment, the disease area includes, for example, the digestive system, circulatory system, respiratory organs, central nervous system, vaccines, vitamins, antibodies, etc. The queried event refers to which event of the various events included in the drug manufacturing and sales approval application (face-to-face consultation, clinical trial notification, approval review, reliability survey, etc.) the inquiry is about. The item of the inquiry content refers, for example, to the type of drug. The response time refers to the date on which the input document data 800 was sent to the authority terminal 110.

[0066] <Description of registered data> Next, a specific example of the registration data stored in the database 151 of the server device 150 by the data management device 120 will be described. Fig. 10 is a diagram showing an example of the registration data.

[0067] As shown in Figure 10, the registered data 1000 includes the following information items: "Number," "Authority Inquiry ID," "Authority Inquiry Document Data," "Attribute Information," "Inputted Document Data," "Inputted Divided Document Data," and "Keywords."

[0068] In the "number" field, a number indicating the order in which each piece of registration data included in the registration data 1000 was stored in the database 151 is stored.

[0069] An identifier for identifying each piece of authority inquiry document data sent from the authority terminal 110 is stored in the "authority inquiry ID."

[0070] The "authority inquiry document data" stores the authority inquiry document data transmitted from the authority terminal 110. In FIG.

[0071] Attribute information extracted from the corresponding authority inquiry document data or input document data is stored in "attribute information." In Fig. 10, "AAA" indicates attribute information 900, for example.

[0072] The input document data generated based on the corresponding authority inquiry document data is stored in the "input document data." In Fig. 10, "PDF001" indicates the input document data 800, for example.

[0073] "Input divided document data" stores input divided document data that was used to generate the corresponding input divided document data. In FIG. 10, "text 1001" refers to, for example, input divided document data 710. Furthermore, "text 1002" refers to, for example, input divided document data 720. Note that "text 1001" may be input divided document data 710 itself, or may be text data in which a text data portion is extracted from input divided document data 710. Similarly, "text 1002" may be input divided document data 720 itself, or may be text data in which a text data portion is extracted from input divided document data 720.

[0074] "Keywords" stores keywords assigned by applicant 122 based on the corresponding input divided document data. In FIG. 10, "K1001" indicates, for example, a keyword assigned by applicant 122 based on input divided document data 710. Also, "K1002" indicates, for example, a keyword assigned by applicant 122 based on input divided document data 720.

[0075] <Functional configuration of the search unit of the server device> Next, the functional configuration of the search unit of the server device 150 will be described. FIG. 11 is a first diagram showing an example of the functional configuration of the search unit of the server device. As described above, in the search phase, the server device 150 functions as the search unit 201. As shown in FIG. 11, the search unit 201 further A search screen providing unit 1101, A search condition acquisition unit 1102, Search control unit 1103, Includes.

[0076] The search screen providing unit 1101 provides a search screen when accessed by the worker terminal 130, 140, etc. Furthermore, the search screen providing unit 1101 receives a search request from the worker terminal 130, 140, etc. in response to providing the search screen. Furthermore, when the search control unit 1103 notifies the search results, the search screen providing unit 1101 includes the search results in a search screen and provides the search results to the worker terminal 130, 140, etc. Furthermore, when the search screen providing unit 1101 provides a search screen including the search results, the search screen providing unit 1101 receives an instruction to display details from the worker terminal 130, 140, etc. Furthermore, when the search control unit 1103 notifies the search screen providing unit 1101 of detailed information, the search screen providing unit 1101 includes the detailed information in a search screen and provides the search screen to the worker terminal 130, 140, etc.

[0077] The search condition acquisition unit 1102 acquires the search conditions included in the search request received by the search screen provision unit 1101. The search condition acquisition unit 1102 also notifies the search control unit 1103 of the acquired search conditions.

[0078] The search control unit 1103 searches the database 151 and acquires input divided document data that matches the search conditions notified by the search condition acquisition unit 1102. The search control unit 1103 also notifies the search screen providing unit 1101 of the acquired input divided document data as the search results.

[0079] Furthermore, when the search control unit 1103 acquires an instruction for detailed display received by the search screen providing unit 1101, it acquires registered data (attribute information, input document data, keywords, etc.) other than the input divided document data notified as the search result from the database 151. Furthermore, the search control unit 1103 notifies the search screen providing unit 1101 of the registered data (attribute information, input document data, keywords, etc.) other than the input divided document data notified as the search result as detailed information.

[0080] <Explanation of the search screen> Next, a description will be given of a search screen provided to worker terminals 130, 140, etc. by server device 150. Fig. 12 is a first diagram showing an example of a search screen displayed on a worker terminal.

[0081] Of these, the search screen 1210 is a screen for inputting search conditions and sending a search request. For example, as shown in Fig. 12, the search screen 1210 includes search items such as "character string," "time of inquiry to authorities," "disease area," "event," "item," and "time of response."

[0082] "String" is an item that is input when searching for input divided document data based on a string contained in the query content in the input divided document data or a string contained in a keyword associated with the input divided document data.

[0083] "Time of inquiry by authority" to "Time of response" are items that are input when searching for input divided document data based on the attribute information associated with the input divided document data. The first and last days of the period are input into "Time of inquiry by authority" and "Time of response". This allows the search control unit 1103 to search for input divided document data whose time of inquiry by authority or time of response falls within the input period.

[0084] Further, for "disease area" to "item", an option selected by the operator 131, 141, etc. from predetermined options is input. This enables the search control unit 1103 to search for input divided document data associated with attribute information that matches the input option.

[0085] The search screen 1210 also includes a search button 1211. When the input into the search items is completed and the search button 1211 is pressed, a search request is sent to the server device 150 using the information input into the search items as search conditions.

[0086] On the other hand, the search screen 1220 is a screen that is displayed on the worker terminals 130, 140, etc. when a search request is sent, the server device 150 searches for input divided document data that matches the search conditions, and provides a search screen containing the search results.

[0087] 12, the number of pieces of input divided document data that match the search criteria is displayed on search screen 1220. Also displayed on search screen 1220 are input divided document data that match the search criteria and registered data other than the input divided document data (however, for registered data other than the input divided document data, only the file name is displayed here).

[0088] The search screen 1220 also includes a details display button 1221, which is pressed when the operator wishes to display details of any of the registered data other than the input divided document data that matches the search criteria. The operator 131, 141, etc. selects the registered data (file name) other than the input divided document data for which the operator wishes to display details, and then presses the details display button 1221. As a result, an instruction to display details of the registered data (file name) other than the selected input divided document data is sent to the server device 150.

[0089] <Business support system processing (accumulation phase)> Next, a description will be given of the flow of processing in the accumulation phase by the business support system 100. Fig. 13 is a first sequence diagram showing the flow of processing in the accumulation phase by the business support system.

[0090] In step S1301, the authority terminal 110 transmits authority inquiry document data including a plurality of inquiry items, and the data management device 120 receives the authority inquiry document data transmitted from the authority terminal 110.

[0091] In step S1302, the data management device 120 divides the received authority inquiry document data into multiple pieces of divided document data for each inquiry item.

[0092] In step S1303, the data management device 120 transmits the generated multiple pieces of divided document data to the corresponding worker terminals 130, 140, etc., and the worker terminals 130, 140, etc. receive the divided document data transmitted from the data management device 120.

[0093] In step S1304, the worker terminals 130, 140, etc. each display the received divided document data, thereby accepting the response to the inquiry input by the workers 131, 141, etc. As a result, the worker terminals 130, 140, etc. generate the input divided document data.

[0094] In step S1305, the worker terminals 130, 140, etc. transmit the generated input divided document data to the data management device 120.

[0095] In step S1306, the data management device 120 collects input divided document data transmitted from the worker terminals 130, 140, and the like.

[0096] In step S1307, the data management device 120 generates input document data by converting the collected input divided document data into one file format.

[0097] In step S1308, the data management device 120 extracts attribute information from the authority inquiry document data and the input document data.

[0098] In step S1309, the data management device 120 transmits the generated input document data to the authority terminal 110.

[0099] In step S1310, data management device 120 accepts keywords input by applicant 122 based on the input divided document data.

[0100] In step S1311, the data management device 120 transmits the registration data to the server device 150.

[0101] In step S 1312 , server device 150 stores the registration data transmitted from server device 150 in database 151 .

[0102] <Business support system processing (search phase)> Next, a description will be given of the flow of processing in the search phase by the business support system 200. Fig. 14 is a first sequence diagram showing the flow of processing in the search phase by the business support system.

[0103] In step S1401, the worker terminal 130, 140, etc. accesses the search unit 201 of the server device 150.

[0104] In step S1402, the server device 150 starts providing a search screen in response to access from the worker terminals 130, 140, etc.

[0105] In step S1403, the worker terminal 130, 140, etc. displays a search screen provided by the server device 150.

[0106] In step S1404, the worker terminals 130, 140, etc. accept input of search conditions by the workers 131, 141, etc.

[0107] In step S1405, the worker terminals 130, 140, etc. accept a search instruction from the workers 131, 141, etc. In response, the worker terminals 130, 140, etc. transmit a search request including the input search conditions to the server device 150.

[0108] In step S1406, the server device 150 receives the search request transmitted from the worker terminal 130, 140, etc., and performs search processing based on the search conditions included in the received search request.

[0109] In step S1407, the server device 150 acquires the search results and provides a search screen including the acquired search results to the worker terminals 130, 140, etc.

[0110] In step S1408, the worker terminal 130, 140, etc. displays a search screen including the search results.

[0111] In step S1409, the worker terminals 130, 140, etc. accept an instruction to display details from the workers 131, 141, etc. As a result, the worker terminals 130, 140, etc. transmit an instruction to display details to the server device 150, and the server device 150 receives the instruction to display details transmitted from the worker terminals 130, 140, etc.

[0112] In step S1410, server device 150 acquires detailed information based on the received instruction to display details, and provides a search screen including the acquired detailed information to worker terminals 130, 140, etc.

[0113] In step S1411, the worker terminal 130, 140, etc. displays a search screen including detailed information.

[0114] <Summary> As is clear from the above description, the business support system 100 according to the first embodiment performs the following operations in the accumulation phase: -Multiple inquiry items contained in the document data of an inquiry from an authority are divided into multiple pieces of divided document data by each inquiry item. Acquire a plurality of input divided document data in which the response content to each of a plurality of inquiries is input. Acquire input document data that is document data in a single file format generated from the acquired multiple pieces of input divided document data. Each of the acquired multiple pieces of input divided document data is stored in a database in association with the input document data.

[0115] In this way, in the accumulation phase, the business support system 100 according to the first embodiment builds a database that enables easy searching of similar past cases of inquiries from authorities. As a result, the business support system 100 according to the first embodiment can improve the efficiency of searches by workers and support the workers' response work. In other words, the first embodiment can provide a mechanism for supporting the response work to inquiries from authorities.

[0116] [Second embodiment] In the first embodiment, the input divided document data is displayed to the applicant 122, and the applicant 122 assigns keywords based on the displayed input divided document data. However, the method of assigning keywords is not limited to this, and for example, keywords may be assigned using a trained keyword model generated by training a model on past cases. The second embodiment will be described below, focusing on the differences from the first embodiment.

[0117] <System configuration of business support system> First, a system configuration of a business support system according to the second embodiment will be described. In the second embodiment, the business support system executes processing in a learning phase and processing in a storage phase, and therefore, the system configuration will be described here for each phase.

[0118] (1) Work support system during the learning phase FIG. 15 is a third diagram showing an example of the system configuration of a business support system, illustrating the system configuration in the learning phase. The learning phase is a phase executed separately from the task of responding to inquiries from authorities, and is a phase in which a trained keyword model for automatically assigning keywords is generated. The difference from the business support system 100 in the accumulation phase shown in FIG. 1 is that the business support system 1500 in the learning phase has a learning device 1510.

[0119] A learning program is installed in the learning device 1510, and the learning device 1510 functions as a learning unit 1511 by executing the program in the learning phase.

[0120] The learning unit 1511 reads out registered data stored in the database 151 of the server device 150 and generates a learning dataset. The learning dataset generated by the learning unit 1511 is a learning dataset that uses the input divided document data as input data and the corresponding keywords as answer data.

[0121] Furthermore, the learning unit 1511 performs a learning process on a keyword model (a model for predicting keywords) using the generated learning data set, and generates a learned keyword model.

[0122] (2) Business support system for the accumulation phase FIG. 16 is a fourth diagram showing an example of the system configuration of the business support system, and shows the system configuration in the accumulation phase.

[0123] The differences from the system configuration in the learning phase are that the management unit 1601 of the data management device 120 in the accumulation phase has a different function, and that a prediction device 1610 is newly provided.

[0124] The management unit 1601 of the data management device 120 acquires keywords suitable for the input divided document data from the prediction device 1610 .

[0125] A prediction program is installed in the prediction device 1610, and the prediction device 1610 functions as a prediction unit 1611 by executing the program in the accumulation phase.

[0126] The prediction unit 1611 includes a trained keyword model generated by the learning unit 1511. Each time authority inquiry document data is transmitted from the authority terminal 110 and input divided document data is collected by the data management device 120, the prediction unit 1611 predicts keywords and notifies the data management device 120.

[0127] Specifically, each time the prediction unit 1611 receives collected input divided document data from the management unit 1601 of the data management device 120, it predicts keywords suitable for the input divided document data and notifies the management unit 1601 of the data management device 120.

[0128] The keywords notified by prediction unit 1611 are associated with the input divided document data and transmitted to server device 150. As a result, server device 150 can store the registered data to which keywords have been assigned by prediction unit 1611 in database 151.

[0129] <Functional configuration of the learning unit of the learning device> Next, the functional configuration of the learning device 1510 will be described. FIG. 17 is a diagram showing an example of the functional configuration of the learning unit of the learning device. As shown in FIG. 17, the learning unit 1511 reads the registered data 1000 from the database 151 and generates a learning dataset. The example in FIG. 17 is as follows: Based on the read-out registration data 1000, the inquiry contents and answer contents included in the input divided document data are input as input data, Keywords that are recalled from the contents of each inquiry and answer contained in the input divided document data are used as correct answer data. This shows how a training dataset was generated.

[0130] As shown in FIG. 17, the learning unit 1511 includes a morphological analysis unit 1711, a vectorization model 1712, a keyword model 1713, and a comparison / modification unit 1714.

[0131] The morphological analysis unit 1711 reads out the inputted divided document data included in the learning dataset and performs morphological analysis on the sentences included in the inquiry content and the response content of the inputted divided document data. The morphological analysis unit 1711 also notifies the vectorization model 1712 of the analysis result of the morphological analysis.

[0132] The vectorization model 1712 processes the analysis results notified by the morphological analysis unit 1711 using vectorization techniques such as TF-IDF (Term Frequency-Inverse Document Frequency), LSI (Latent Semantic Indexing), LDA (Latent Dirichlet Allocation), Doc2Vec, Sent2Vec, RNN (Recurrent Neural Network), LSTM (Long Short-Term Memory), GRU (Gated-Recurrent Unit), USE (Universal Sentence Encoder), BERT (Bidirectional Encoder Representations from Transformers), Sentence-Transformers, etc., or a combination of these, to generate vector data. The vectorization model 1712 also notifies the keyword model 1713 of the vector data.

[0133] The keyword model 1713 processes the vector data notified by the vectorization model 1712 using a category classification technique such as a support vector machine, a decision tree, k-means, nearest neighbor, logistic regression, random forest, gradient boosting decision tree, neural network, deep learning (e.g., RNN (Recurrent Neural Network), LSTM (Long Short-Term Memory), GRU (Gated-Recurrent Unit), Seq2seq, Sentence-Transformers, etc.), or a combination of these, and outputs output data.

[0134] The comparison / change unit 1714 compares the output data output from the keyword model 1713 with the keywords included in the training data set, and updates the model parameters of the keyword model 1713 so as to reduce the error between them. This enables the training unit 1511 to generate a trained keyword model.

[0135] <Functional configuration of the management unit of the data management device and functional configuration of the prediction device> Next, a description will be given of the functional configuration of the management unit 1601 of the data management device 120 and the functional configuration of the prediction unit 1611 of the prediction device 1610. Fig. 18 is a diagram showing an example of the functional configuration of the management unit of the data management device and the functional configuration of the prediction unit of the prediction device.

[0136] The management unit 1601 of the data management device 120, like the management unit 121 shown in FIG. Document data acquisition unit 501, ·Divided part 503, ·Distribution section 504, ·Attribute information extraction unit 505, Collection unit 506, Document data generation unit 507, Document data transmission unit 508, Storage unit 1801, Of these, document data acquisition unit 501 to document data transmission unit 508 have already been explained using FIG.

[0137] When the storage unit 1801 is notified of the input divided document data by the collection unit 506, it notifies the prediction device 1610 of the notified input divided document data. In addition, the storage unit 1801 acquires the keywords notified by the prediction device 1610 in response to being notified of the input divided document data.

[0138] Similarly to the storage unit 509, the storage unit 1801 stores the registration data in the database 151 of the server device 150. The registration data includes the following: Input divided document data notified by the collection unit 506, Input document data notified by the document data generation unit 507, Attribute information notified by the attribute information extraction unit 505, Keywords notified by the prediction device 1610, etc. are included.

[0139] The prediction unit 1611 of the prediction device 1610 ·Morphological analysis section 1711, Vectorized Model 1712, · Pre-trained keyword model 1802, Among these, the morphological analysis unit 1711 and the vectorization model 1712 have already been explained using FIG.

[0140] The learned keyword model 1802 is a model generated by the learning unit 1511, and predicts keywords when vector data is notified from the vectorization model 1712. The keywords predicted by the learned keyword model 1802 are notified to the storage unit 1801.

[0141] <Business support system processing (learning phase)> Next, a description will be given of the flow of processing in the learning phase by the business support system 1500. Fig. 19 is a sequence diagram showing the flow of processing in the learning phase by the business support system.

[0142] In step S 1901 , server device 150 reads out registration data stored in database 151 and transmits it to learning device 1510 .

[0143] In step S1902, the learning device 1510 acquires the transmitted enrollment data.

[0144] In step S1903, the learning device 1510 generates a learning dataset based on the acquired enrollment data.

[0145] In step S1904, the learning device 1510 performs a learning process on the keyword model 1713 using the generated learning data set, and generates a trained keyword model.

[0146] In step S1905, the learning device 1510 stores the generated trained keyword model in the storage unit of the prediction device 1610 together with the morphological analysis unit 1711 and the vectorization model 1712.

[0147] <Business support system processing (accumulation phase)> Next, a description will be given of the flow of processing in the accumulation phase by the business support system 1600. Fig. 20 is a second sequence diagram showing the flow of processing in the accumulation phase by the business support system.

[0148] Note that steps S1301 to S1309, S1311, and S1312 in FIG. 20 are the same as steps S1301 to S1309, S1311, and S1312 in FIG. 13, and therefore will not be described here.

[0149] In step S2001, the prediction device 1610 predicts keywords based on the input divided document data notified by the data management device 120, and notifies the data management device 120 of the predicted keywords.

[0150] <Summary> As is clear from the above description, the business support system 1500 according to the second embodiment performs the following in the learning phase: A learning dataset is generated that includes each inquiry and response contained in the input divided document data read from the database, and keywords that are associated with each inquiry and response contained in the input divided document data. A trained keyword model is generated by performing a training process using a training dataset.

[0151] Furthermore, the business support system 1600 according to the second embodiment performs the following in the accumulation phase: -By using a trained keyword model, it predicts keywords that are evoked from each inquiry and response contained in the input document data.

[0152] In this way, the business support system 1600 according to the second embodiment is configured to automatically assign keywords in the accumulation phase when constructing a database that allows easy searching of similar past cases of inquiries from authorities.

[0153] As a result, according to the second embodiment, it is possible to reduce the workload imposed on the applicant when storing registration data in a database.

[0154] [Third embodiment] In the above first and second embodiments, for example, a case where a worker 131, 141, etc. inputs search conditions into the search items shown on the search screen 1210 in Figure 12 to search for past cases similar to the inquiry content included in the authority inquiry document data was described.

[0155] However, the search method for searching for past cases similar to the inquiry content included in the official inquiry document data is not limited to this. For example, the search may be configured to search for similar past cases by inputting the inquiry content itself included in the official inquiry document data. The following describes the third embodiment, focusing on the differences from the first and second embodiments.

[0156] <System configuration of business support system> First, a system configuration of a business support system in a search phase will be described as a business support system according to the third embodiment. Fig. 21 is a fifth diagram showing an example of the system configuration of the business support system.

[0157] The difference between the business support system 200 in the search phase shown in Figure 2 and the business support system 2100 in the search phase shown in Figure 21 is that when making a search request, the inquiry content itself contained in the authority inquiry document data to be processed is input.

[0158] 2 is that the function of the search unit 2101 in the business support system 2100 in the search phase shown in FIG. 21 is different from the function of the search unit 201. Specifically, the search unit 2101 vectorizes the inquiry content included in the search request (the inquiry content of the official inquiry document data to be processed) to generate vector data. The search unit 2101 also vectorizes the inquiry content included in each input divided document data stored in the database 151 to generate vector data. Furthermore, the search unit 2101 calculates the similarity of the vector data to search for input divided document data that includes inquiry content similar to the inquiry content included in the search request.

[0159] <Functional configuration of the search unit of the server device> Next, the functional configuration of the search unit 2101 of the server device 150 will be described. FIG. 22 is a second diagram showing an example of the functional configuration of the search unit of the server device. As described above, in the search phase of the third embodiment, the server device 150 functions as the search unit 2101. As shown in FIG. 22, the search unit 2101 further ·Search screen providing unit 2201, ·Extraction unit 2202, a first vectorization unit 2203; ·Separation part 2204, A second vectorization unit 2205, ·Similarity calculation unit 2206, Output unit 2207, Includes.

[0160] The search screen providing unit 2201 provides a search screen when accessed by the worker terminal 130, 140, etc. Furthermore, the search screen providing unit 2201 receives a search request from the worker terminal 130, 140, etc. in response to providing the search screen. Furthermore, when the search screen providing unit 2201 is notified of search results from the output unit 2207, the search screen providing unit 2201 includes the search results in a search screen and provides it to the worker terminal 130, 140, etc. Furthermore, when the search screen providing unit 2201 provides a search screen including the search results, it receives an instruction to display details from the worker terminal 130, 140, etc. Furthermore, when detailed information is notified from the output unit 2207, the search screen providing unit 2201 includes the detailed information in a search screen and provides it to the worker terminal 130, 140, etc.

[0161] The extraction unit 2202 extracts the query content included in the search request received by the search screen providing unit 2201. The extraction unit 2202 also notifies the first vectorization unit 2203 of the extracted query content.

[0162] The first vectorization unit 2203 vectorizes the query content notified by the extraction unit 2202, and generates vector data. The first vectorization unit 2203 also notifies the similarity calculation unit 2206 of the generated vector data.

[0163] The separating unit 2204 reads out the input divided document data from the database 151 and separates the read input divided document data into the inquiry content and the response content. Furthermore, the separating unit 2204 notifies the second vectorizing unit 2205 of the inquiry content out of the separated inquiry content and response content.

[0164] The second vectorization unit 2205 vectorizes the query content notified by the separation unit 2204, and generates vector data. The second vectorization unit 2205 also notifies the similarity calculation unit 2206 of the generated vector data.

[0165] The similarity calculation unit 2206 calculates the similarity between the vector data notified by the first vectorization unit 2203 and the vector data notified by the second vectorization unit 2205. In addition, the similarity calculation unit 2206 notifies the output unit 2207 of the calculated similarity.

[0166] The output unit 2207 searches for a similarity that satisfies a predetermined similarity condition (for example, the highest similarity or the top n similarities (n is an arbitrary number)) from the similarities notified by the similarity calculation unit 2206, and identifies the corresponding input divided document data. The output unit 2207 also notifies the search screen providing unit 2201 of the identified input divided document data as the search result.

[0167] Furthermore, when the output unit 2207 acquires an instruction for detailed display received by the search screen providing unit 2201, it acquires registered data (attribute information, input document data, keywords, etc.) other than the input divided document data notified as the search result from the database 151. Furthermore, the output unit 2207 notifies the search screen providing unit 2201 of the registered data (attribute information, input document data, keywords, etc.) other than the input divided document data notified as the search result as detailed information. In other words, the output unit 2207 notifies the search screen providing unit 2201 of the data other than the input divided document data included in the registered data corresponding to the searched query content as detailed information.

[0168] <Explanation of the search screen> Next, a description will be given of a search screen provided to worker terminals 130, 140, etc. by server device 150. Fig. 23 is a second diagram showing an example of the search screen displayed on the worker terminal.

[0169] Of these, the search screen 2310 is a screen for inputting the contents of a query and sending a search request. As shown in Fig. 23, the search screen 2310 includes a field for inputting the contents of a query.

[0170] Workers 131, 141, etc., for example, input the inquiry content into the field for inputting the inquiry content by copying and pasting the inquiry content contained in the authority inquiry document data to be processed (actually, the inquiry content contained in the divided document data sent from the data management device 120).

[0171] The search screen 2310 also includes a search button 1211. When the user has completed inputting the query content into the field for inputting the query content and pressed the search button 1211, a search request including the query content is sent to the server device 150.

[0172] On the other hand, the search screen 1220 is a screen that is displayed on the worker terminals 130, 140, etc. when a search request is sent, the server device 150 searches for input divided document data that contains query content similar to the query content included in the search request, and provides a search screen containing the search results.

[0173] The search screen 1220 shown in FIG. 23 is the same as the search screen 1220 shown in FIG. 12, and therefore a description thereof will be omitted here.

[0174] <Specific examples of processing by each unit included in the search unit of the server device> Next, a specific example of the processing of each unit (here, first vectorization unit 2203, separation unit 2204, and second vectorization unit 2205) included in search unit 2101 of server device 150 will be described. Fig. 24 is a first diagram showing a specific example of the processing of the first vectorization unit, separation unit, and second vectorization unit included in the search unit of the server device.

[0175] As shown in FIG. 24( a ), the first vectorization unit 2203 includes a morphological analysis unit 2401 and a vectorization model 2402 .

[0176] When the morphological analysis unit 2401 is notified of the inquiry content by the extraction unit 2202, it performs morphological analysis on each sentence included in the notified inquiry content using a pharmaceutical development terminology dictionary stored in a pharmaceutical development terminology dictionary storage unit 2403. The pharmaceutical development terminology dictionary is a dictionary containing terms specific to pharmaceutical development. The morphological analysis unit 2401 also notifies the vectorization model 2402 of the analysis results of the morphological analysis.

[0177] The vectorization model 2402 is an example of a vectorization unit, and generates vector data by processing the analysis results of the morphological analysis notified by the morphological analysis unit 2401. Specifically, the vectorization model 2402 generates vector data by using, for example, TF-IDF or the like to vectorize each word extracted from the query content using a value corresponding to the frequency of appearance of each word. Specifically, the vector data may be generated using vectorization techniques such as TF-IDF, latent semantic indexing (LSI), latent Dirichlet allocation (LDA), Doc2Vec, Sent2Vec, Word2Vec, recurrent neural network (RNN), long short-term memory (LSTM), gated-recurrent unit (GRU), universal sentence encoder (USE), bidirectional encoder representations from transformers (BERT), sentence transformers, etc., or a combination of these.

[0178] As shown in Figure 24(b), the separation unit 2204 sequentially reads out the input divided document data from the database 151, separates the query content from the read input divided document data, and notifies the second vectorization unit 2205 of the separated query content.

[0179] The second vectorization unit 2205 has a morphological analysis unit 2411 and a vectorization model 2412. When the morphological analysis unit 2411 is notified of the inquiry content by the separation unit 2204, it performs morphological analysis on each sentence included in the notified inquiry content using a pharmaceutical development terminology dictionary stored in a pharmaceutical development terminology dictionary storage unit 2413. In addition, the morphological analysis unit 2411 notifies the vectorization model 2412 of the analysis results of the morphological analysis.

[0180] The vectorization model 2412 is an example of a vectorization unit, and generates vector data by processing the analysis results of the morphological analysis notified by the morphological analysis unit 2411. Specifically, the vectorization model 2412 uses, for example, TF-IDF or the like to vectorize each word extracted from the query content using a value corresponding to the frequency of appearance, thereby generating vector data. Specifically, vector data may be generated using vectorization techniques such as TF-IDF, latent semantic indexing (LSI), latent Dirichlet allocation (LDA), Doc2Vec, Sent2Vec, Word2Vec, recurrent neural network (RNN), long short-term memory (LSTM), gated-recurrent unit (GRU), universal sentence encoder (USE), bidirectional encoder representations from transformers (BERT), sentence transformers, and the like, as well as combinations of these.

[0181] For ease of explanation, the vectorization model and the morphological analysis unit are each shown as two functional blocks in Fig. 24. However, the vectorization model and the morphological analysis unit that process the query content included in the search request and the vectorization model and the morphological analysis unit that process the query content separated from the input segmented document data may be different, but are preferably the same.

[0182] Similarly, for ease of explanation, two pharmaceutical development dictionary storage units are shown in Figure 24. However, the pharmaceutical development dictionary used when processing the query content included in the search request and the pharmaceutical development dictionary used when processing the query content separated from the input segmented document data may be different, but it is preferable that they are the same.

[0183] <Business support system processing (search phase)> Next, a description will be given of the flow of processing in the search phase by the business support system 2100. Fig. 25 is a second sequence diagram showing the flow of processing in the search phase by the business support system. The differences from the processing in the search phase by the business support system 200 described using Fig. 14 are steps S2501 to S2503.

[0184] In step S2501, the worker terminal 130, 140, etc. accepts input of inquiry content by the worker 131, 141, etc.

[0185] In step S2502, the worker terminals 130, 140, etc. accept a search instruction from the workers 131, 141, etc. In response, the worker terminals 130, 140, etc. transmit a search request including the input inquiry content to the server device 150.

[0186] In step S2503, server device 150 receives the search request transmitted from worker terminal 130, 140, etc., and searches for input divided document data containing a query similar to the query contained in the received search request.

[0187] <Summary> As is clear from the above description, the business support system 2100 according to the third embodiment performs the following in the search phase: · Using a dictionary containing terms specific to pharmaceutical development, morphological analysis is performed on the inquiry content contained in the authority inquiry document data to be processed. -By using morphological analysis, the inquiry content contained in the authority inquiry document data to be processed is vectorized based on each word extracted from the inquiry content contained in the authority inquiry document data to be processed. Vectorization is used to calculate the similarity with the vector data generated from the inquiry content contained in the authority inquiry document data to be processed, and search for input divided document data that contains similar inquiry content.

[0188] In this way, in the business support system 2100 according to the third embodiment, in the search phase, in order to easily search for similar past cases of authority inquiries, instead of inputting search conditions, the inquiry content itself is input to perform the search.

[0189] As a result, the business support system 2100 according to the third embodiment can improve the efficiency of the work performed by the worker when searching, and can support the worker in responding to inquiries. In other words, the third embodiment can provide a mechanism for supporting the work of responding to inquiries from authorities.

[0190] [Fourth embodiment] In the third embodiment, the input divided document data containing similar inquiry content is searched for by inputting the inquiry content contained in the official inquiry document data. However, the method of searching for input divided document data containing similar inquiry content is not limited to this. For example, the input divided document data to be searched may be narrowed down by inputting predetermined search conditions, and then input divided document data containing inquiry content similar to the input inquiry content may be searched for. This can improve search accuracy. The fourth embodiment will be described below, focusing on the differences from the third embodiment.

[0191] <Functional configuration of the search unit of the server device> First, the functional configuration of the search unit 2101 of the server device 150 according to the fourth embodiment will be described. Fig. 26 is a third diagram showing an example of the functional configuration of the search unit of the server device. The functional configuration differs from that shown in Fig. 22 in that the search unit 2101 shown in Fig. 26 has a function of the extraction unit 2601 that is different from the function of the extraction unit 2202 shown in Fig. 22. Another difference from the functional configuration shown in Fig. 22 is that the search unit 2101 shown in Fig. 26 has a filter unit 2602.

[0192] The extraction unit 2601 extracts search conditions (in this embodiment, these are referred to as filter conditions because they are used to narrow down search targets) and query contents included in the search request received by the search screen provision unit 2201. The extraction unit 2601 also notifies the filter unit 2602 of the extracted filter conditions and notifies the first vectorization unit 2203 of the extracted query contents.

[0193] When the filtering condition is notified by the extraction unit 2601, the filtering unit 2602 searches the database 151 and acquires input divided document data that matches the filtering condition. The filtering unit 2602 also notifies the separation unit 2204 of the acquired input divided document data.

[0194] As a result, the separation unit 2204 notifies the second vectorization unit 2205 of the query content contained in the input divided document data acquired by the filter unit 2602, thereby narrowing down the targets for which similarity is to be calculated in the similarity calculation unit 2206.

[0195] <Explanation of the search screen> Next, a search screen provided to the worker terminals 130, 140, etc. by the server device 150 according to the fourth embodiment will be described. Fig. 27 is a third diagram showing an example of the search screen displayed on the worker terminal.

[0196] Of these, the search screen 2710 is a screen for inputting the inquiry content and filter conditions and sending a search request.

[0197] 27, the search screen 2710 includes a field for inputting the inquiry content and a field for inputting the filter conditions. The workers 131, 141, etc. input the inquiry content into the field for inputting the inquiry content, for example, by copying and pasting the inquiry content included in the authority inquiry document data.

[0198] 27, the search screen 2710 includes, as filter conditions, "character string," "time of inquiry to authorities," "disease area," "event," "item," and "time of response." Note that the "character string," "time of inquiry to authorities," "disease area," "event," "item," and "time of response" that are input as filter conditions have already been explained using FIG. 12, and therefore will not be explained here.

[0199] On the other hand, the search screen 1220 is a screen that is displayed on the worker terminals 130, 140, etc. when a search request is sent, the server device 150 searches for input divided document data that contains query content similar to the query content included in the search request, and provides a search screen containing the search results.

[0200] The search screen 1220 shown in FIG. 27 is the same as the search screen 1220 shown in FIG. 12, and therefore a description thereof will be omitted here.

[0201] <Summary> As is clear from the above description, the business support system 2100 according to the fourth embodiment performs the following operations in the search phase: A search is made for the inquiry content contained in the input divided document data that satisfies a predetermined filter condition among the plurality of input divided document data contained in the registration data stored in the database 151. Filter criteria include when the regulatory inquiry was made, when the regulatory inquiry was responded to, disease area, event referred, and item.

[0202] As described above, in the business support system 2100 according to the fourth embodiment, in the search phase, in order to easily search for past cases similar to the authority inquiry, the inquiry content itself is input, and a search is performed by inputting filter conditions. As a result, according to the fourth embodiment, the accuracy of searches performed by workers can be improved. In particular, whereas conventional searches required proficiency in search techniques such as selecting appropriate search terms and using Boolean operators, it is now possible to easily search for past similar cases simply by inputting the authority inquiry.

[0203] [Fifth embodiment] In the business support systems according to the third and fourth embodiments, input divided document data containing inquiry content similar to the inquiry content contained in the official inquiry document data is configured to be displayed as a search result on the worker terminal 130, 140, etc. However, the configuration of the business support system is not limited to this, and for example, when the search result is displayed on the worker terminal 130, 140, etc., an input of an evaluation of the search result may be accepted and the accepted evaluation may be fed back. The fifth embodiment will be described below, focusing on the differences from the third and fourth embodiments.

[0204] <System configuration of business support system> First, the system configuration of a business support system according to the fifth embodiment will be described. In the fifth embodiment, the business support system executes processing in a feedback phase. FIG. 28 is a sixth diagram showing an example of the system configuration of the business support system. The feedback phase is a phase executed separately from the task of responding to inquiries from authorities, and is a phase in which, when a certain amount of feedback information has been accumulated, the search unit 2101 is updated to improve search accuracy.

[0205] The difference between the business support system 2800 in the accumulation phase shown in Fig. 1 and the business support system 100 in the feedback phase is that the business support system 2800 in the feedback phase has an analysis device 2810. Also, the business support system 2800 in the feedback phase is configured so that the workers 131, 141, etc. can input evaluations of the search results.

[0206] An analysis program is installed in analysis device 2810, and by executing the program in the feedback phase, analysis device 2810 functions as analysis unit 2811. Analysis unit 2811 may execute various processes under operational instructions from analyst 2822.

[0207] Specifically, the analysis unit 2811 acquires feedback information from the operator terminals 130, 140, etc. Furthermore, the analysis unit 2811 updates the pharmaceutical development terminology dictionaries stored in the pharmaceutical development terminology dictionary storage units 2403, 2413 of the search unit 2101 based on the acquired feedback information.

[0208] Furthermore, the analysis unit 2811 updates the hyperparameters of the vectorization models 2402 and 2412 of the search unit 2101 based on the obtained feedback information.

[0209] Furthermore, the analysis unit 2811 updates the vectorization models 2402 and 2412 of the search unit 2101 to new trained vectorization models based on the acquired feedback information.

[0210] The analysis unit 2811 transmits update information to the server device 150 to realize the update.

[0211] When the worker terminal 130, 140, etc. receives an input of an evaluation of the search results, it acquires the input evaluation result and corresponding information. The information corresponding to the evaluation result includes the query content included in the search request, the query content of the search results, and keywords, categories, time periods, etc. corresponding to the query content and response content of the search results. The worker terminal 130, 140, etc. also transmits the input evaluation result and corresponding information (keywords, categories, time periods, etc.) to the analysis device 2810 as feedback information.

[0212] When the update information is transmitted from the analysis device 2810, the server device 150 updates the search unit 2101 in accordance with the transmitted update information.

[0213] <Outline of Business Support System Processing (Feedback Phase)> Next, an outline of the process in the feedback phase by the business support system 2800 will be explained. Fig. 29 is a diagram for explaining an outline of the process in the feedback phase by the business support system.

[0214] 29, time axis 2901 indicates the content and timing of operations performed by workers 131, 141, etc. on worker terminals 130, 140, etc. in the accumulation phase. As shown in Fig. 29, in the accumulation phase, workers 131, 141, etc. repeat search operations and input operations of evaluation results for search results. Note that worker terminals 130, 140, etc. transmit feedback information to analysis device 2810 every time an input operation of an evaluation result is performed.

[0215] In FIG. 29, a time axis 2902 indicates the content and timing of the processing performed by the analysis device 2810 in the feedback phase.

[0216] As shown in FIG. 29, in the feedback phase, each time feedback information is sent from the operator terminals 130, 140, etc., the analysis device 2810 analyzes the pharmaceutical development terminology dictionary, adds new terms, and deletes unnecessary terms.

[0217] Furthermore, in the feedback phase, after the pharmaceutical development terminology dictionary has been analyzed for a certain period of time, analysis device 2810 uses a certain amount of accumulated feedback information to analyze vectorized models 2402 and 2412. Specifically, analysis device 2810 adjusts the hyperparameters of vectorized models 2402 and 2412.

[0218] In the feedback phase, the vectorization models 2402 and 2412 are analyzed for a certain period of time, and after the hyperparameters are adjusted, the analysis device 2810 generates a training dataset using a certain amount of accumulated feedback information. Furthermore, the analysis device 2810 performs a learning process using the generated training dataset to generate a new trained vectorization model for generating vector data from the query content.

[0219] In FIG. 29, a time axis 2903 indicates the timing of updates to the pharmaceutical development terminology dictionary storage units 2403 and 2413 of the server device 150.

[0220] 29, in the feedback phase, instructions to add new terms and delete unnecessary terms are transmitted as update information from the analysis device 2810. As a result, the search unit 2101 of the server device 150 updates the pharmaceutical development terminology dictionary.

[0221] In FIG. 29, a time axis 2904 indicates the update contents and update timing of the vectorized models 2402 and 2412 of the server device 150.

[0222] 29, in the feedback phase, the adjusted hyperparameters are transmitted as update information from the analysis device 2810. As a result, the search unit 2101 of the server device 150 updates the hyperparameters of the vectorized models 2402 and 2412 to the adjusted hyperparameters.

[0223] Furthermore, in the feedback phase, the analysis device 2810 transmits the trained vectorized model as update information. As a result, the search unit 2101 of the server device 150 updates the vectorized models 2402 and 2412 to the trained vectorized model.

[0224] <Functional configuration of the analysis unit of the analysis device> Next, the functional configuration of the analysis unit 2811 of the analysis device 2810 will be described. FIG. 30 is a diagram showing an example of the functional configuration of the analysis unit of the analysis device. As described above, in the feedback phase, the analysis device 2810 functions as the analysis unit 2811. Furthermore, as shown in FIG. 30, the analysis unit 2811 further Feedback information acquisition unit 3010, ·Additional term extraction unit 3021, A deletion term extraction unit 3022, a dictionary update unit 3023 (an example of a first update unit); a hyperparameter update unit 3031 (an example of a second update unit); A learning data generation unit 3041, ·Learning Department 3042, a model update unit 3043 (an example of a third update unit); Includes.

[0225] The feedback information acquisition unit 3010 acquires feedback information notified from the worker terminals 130, 140, etc. The feedback information notified from the worker terminals 130, 140, etc. includes: the query contained in the search request; - The query content contained in the entered divided document data of the search results, Evaluation results, Keywords associated with the entered divided document data in the search results, - Categories associated with the entered divided document data in the search results, - The inquiry time and response time associated with the input divided document data of the search results, Includes:

[0226] The additional term extraction unit 3021 extracts terms to be added to the pharmaceutical development terminology dictionary based on the feedback information acquired by the feedback information acquisition unit 3010. The additional term extraction unit 3021 also notifies the dictionary update unit 3023 of the extracted terms.

[0227] The deleted term extraction unit 3022 extracts terms to be deleted from the pharmaceutical development terminology dictionary based on the feedback information acquired by the feedback information acquisition unit 3010. The deleted term extraction unit 3022 also notifies the dictionary update unit 3023 of the extracted terms.

[0228] The dictionary update unit 3023 instructs the server device 150 to add the terms notified by the added term extraction unit 3021 to the pharmaceutical development terminology dictionary. The dictionary update unit 3023 also adds the terms notified by the added term extraction unit 3021 to the pharmaceutical development terminology dictionary storage unit 3024. Note that the analysis unit 2811 assumes that the latest pharmaceutical development terminology dictionaries stored in the pharmaceutical development terminology dictionary storage units 2403 and 2413 of the search unit 2101 of the server device 150 are also stored in the pharmaceutical development terminology dictionary storage unit 3024.

[0229] The hyperparameter update unit 3031 adjusts the hyperparameters of the vectorized model 2402 and the vectorized model 2412 based on the feedback information stored in the feedback information storage unit 3030 .

[0230] Furthermore, the hyperparameter update unit 3031 instructs the server device 150 to update the vectorization model 2402 of the first vectorization unit 2203 and the vectorization model 2412 of the second vectorization unit 2205 using the adjusted hyperparameters.

[0231] The learning data generation unit 3041 generates a learning dataset based on the feedback information acquired by the feedback information acquisition unit 3010 and stored in the feedback information storage unit 3030. Specifically, the learning data generation unit 3041 generates a learning dataset using the query content included in the search request and the query content included in the input divided document data of the search result as input data and the evaluation result as correct answer data.

[0232] The learning unit 3042 performs a learning process on the vectorized model using the training data set generated by the training data generation unit 3041. The learning unit 3042 performs morphological analysis when inputting the input data of the training data set into the vectorized model. At this time, the learning unit 3042 performs the morphological analysis using the latest pharmaceutical development terminology dictionary stored in the pharmaceutical development terminology dictionary storage unit 3024.

[0233] The model update unit 3043 instructs the server device 150 to update the vectorized model 2402 and the vectorized model 2412 using the trained vectorized model generated by the training unit 3042 performing the training process.

[0234] In this way, the update information transmitted from the analysis unit 2811 to the server device 150 includes the following: - Addition and deletion of terms from the pharmaceutical development dictionary; Instructions for updating hyperparameters of the vectorization model 2402 of the first vectorization unit 2203 and the vectorization model 2412 of the second vectorization unit 2205; An instruction to update the vectorization model 2402 of the first vectorization unit 2203 and the vectorization model 2412 of the second vectorization unit 2205 to new trained vectorization models; Includes:

[0235] <Specific example of processing by the learning unit included in the analysis unit of the analysis device> Next, a specific example of the processing of learning unit 3042 included in analysis unit 2811 of analysis device 2810 will be described. Fig. 31 is a diagram showing a specific example of the processing of the learning unit included in the analysis unit of the analysis device.

[0236] As shown in FIG. 31, the training data set 3100 includes information items such as "input data" and "correct answer data."

[0237] "Input data" further includes "search source" and "search results." "Search source" stores the query content included in the search request, among the feedback information. "Search results" stores the query content included in the input divided document data of the search results, among the feedback information.

[0238] The "correct answer data" stores the evaluation result from the feedback information. In the example of Fig. 31, the evaluation result is stored as "good" or "bad", but the evaluation result may also be stored as a numerical value between 0 and 100, for example.

[0239] As shown in FIG. 31, the learning unit 3042 further includes a morphological analysis unit 3101, a vectorization model 3102, a similarity calculation unit 3103, and a comparison / modification unit 3104.

[0240] Input data of the learning dataset 3100 is input to the morphological analysis unit 3101, and the morphological analysis unit 3101 performs morphological analysis on the input data (the query content included in the search request, and the query content included in the input divided document data of the search results). When performing the morphological analysis, the morphological analysis unit 3101 uses the latest pharmaceutical development terminology dictionary stored in the pharmaceutical development terminology dictionary storage unit 3024.

[0241] The vectorization model 3102 receives the analysis results of morphological analysis of input data and outputs vector data. Specifically, when the vectorization model 3102 receives the analysis results of morphological analysis of the query content included in the search request, it outputs vector data of the query content included in the search request. Furthermore, when the vectorization model 3102 receives the analysis results of morphological analysis of the query content included in the input divided document data as a search result, it outputs vector data of the query content included in the input divided document data as a search result.

[0242] Note that the type of vectorization model 3102 is arbitrary, and for example, a transformer such as Sentence-Transformers may be used for the vectorization model 3102. Alternatively, the vectorization model 3102 may be TF-IDF, LSI (Latent Semantic Indexing), LDA (Latent Dirichlet Allocation), Doc2Vec, Sent2Vec, Word2Vec, RNN (Recurrent Neural Network), LSTM (Long short-Term Memory), GRU (Gated-Recurrent Unit), USE (Universal Sentence Encoder), BERT (Bidirectional Encoder Representations from Transformers), Sentence-Transformers, or a combination thereof.

[0243] The similarity calculation unit 3103 calculates the similarity between the vector data of the query content included in the search request output from the vectorized model 3102 and the vector data of the query content included in the input divided document data of the search result. The similarity calculation unit 3103 also notifies the comparison / change unit 3104 of the calculated similarity.

[0244] The comparison / change unit 3104 reads out the similarity notified by the similarity calculation unit 3103 and the evaluation results stored in the "correct answer data" of the learning dataset 3100 that correspond to the input data, and compares them with the notified similarity.

[0245] Furthermore, if the correct data is "good," the comparison / change unit 3104 updates the model parameters of the vectorized model 3102 so that the similarity notified by the similarity calculation unit 3103 becomes higher.

[0246] If the correct data is "not correct", the comparison / change unit 3104 updates the model parameters of the vectorized model 3102 so that the similarity notified by the similarity calculation unit 3103 becomes lower.

[0247] <Specific examples of processing by each unit included in the search unit of the server device> Next, a specific example of the processing of each unit (here, first vectorization unit 2203, separation unit 2204, and second vectorization unit 2205) included in search unit 2101 after update in server device 150, which was updated in the feedback phase, will be described. Fig. 32 is a second diagram showing a specific example of the processing of the first vectorization unit, separation unit, and second vectorization unit included in the search unit of the server device.

[0248] As shown in FIG. 32, update information is sent from analysis device 2810 in the feedback phase, and thereby, pharmaceutical development terminology dictionary storage unit 2403 of first vectorization unit 2203 is updated to pharmaceutical development terminology dictionary storage unit 3203.

[0249] 32 , update information is transmitted from analysis device 2810 in the feedback phase, and the adjusted hyperparameters are applied to the hyperparameters of vectorized model 2402. As a result, vectorized model 2402 is updated to vectorized model 3201. Similarly, update information is transmitted from analysis device 2810 in the feedback phase, and the adjusted hyperparameters are applied to the hyperparameters of vectorized model 2412. As a result, vectorized model 2412 is updated to vectorized model 3211.

[0250] 32, update information is transmitted from the analysis device 2810 in the feedback phase, whereby a new trained vectorization model is applied to the vectorization model 3201. As a result, the vectorization model 3201 is updated to the vectorization model 3202 (trained vectorization model). Similarly, update information is transmitted from the analysis device 2810 in the feedback phase, whereby a new trained vectorization model is applied to the vectorization model 3211. As a result, the vectorization model 3211 is updated to the vectorization model 3212 (trained vectorization model).

[0251] <Business support system processing (feedback phase)> Next, a description will be given of the flow of processing in the feedback phase by the task support system 2800. Fig. 33 is a sequence diagram showing the flow of processing in the feedback phase by the task support system.

[0252] In step S3301, the worker terminal 130, 140, etc. accepts the evaluation result for the search result input by the worker 131, 141, etc., and transmits feedback information to the analysis device 2810.

[0253] In step S3302, analysis device 2810 analyzes the pharmaceutical development terminology dictionary every time feedback information is received, and transmits update information including the addition of new terms and the deletion of unnecessary terms to server device 150.

[0254] In step S3303, the server device 150 updates the pharmaceutical development terminology dictionary based on the update information from the analysis device 2810.

[0255] In step S3304, analysis device 2810 adjusts the hyperparameters of vectorized models 2402 and 2412 based on the accumulated feedback information, and transmits update information including the adjusted hyperparameters to server device 150.

[0256] In step S3305, the server device 150 updates the hyperparameters of the vectorization model 2402 of the first vectorization unit 2203 and the vectorization model 2412 of the second vectorization unit 2205 based on the update information from the analysis device 2810.

[0257] In step S3306, analysis device 2810 generates a training data set based on the accumulated feedback information.

[0258] In step S3307, the analysis device 2810 performs a learning process on the vectorization model using the generated training dataset, thereby generating a new trained vectorization model for generating vector data from the query content. The analysis device 2810 also transmits update information including the generated trained vectorization model to the server device 150.

[0259] In step S3308, the server device 150 updates the vectorization model 2402 of the first vectorization unit 2203 and the vectorization model 2412 of the second vectorization unit 2205 to new trained vectorization models based on the update information from the analysis device 2810.

[0260] <Summary> As is clear from the above description, the business support system 2800 according to the fifth embodiment: The pharmaceutical development terminology dictionary is updated based on the evaluation results for the query content contained in the searched input divided document data. Based on the evaluation result for the query content included in the searched input divided document data, the hyperparameters used by the vectorization model of the first vectorization unit and the vectorization model of the second vectorization unit when generating vector data are updated. A learning dataset is generated using the query content included in the search request and the query content included in the input divided document data of the search results as input data and the evaluation results as correct answer data. A new trained vectorization model is generated by performing a training process using the training dataset, and the vectorization model of the first vectorization unit and the vectorization model of the second vectorization unit are updated using the generated trained vectorization model.

[0261] As described above, the task support system 2800 according to the fifth embodiment is configured to update the search unit 2101 of the server device 150 by feeding back the evaluation results of the search results in the feedback phase. As a result, according to the fifth embodiment, it is possible to improve the search accuracy when the worker performs a search.

[0262] [Sixth embodiment] In the above-described embodiments, the system configuration of each phase is illustrated using Figures 1, 2, 15, 16, 21, 28, etc., but the system configuration of each phase is not limited to those shown in the above-described embodiments. For example, in the accumulation phase of the second embodiment, in Figure 16, the data management device 120 and the prediction device 1610 are configured as separate entities, but the data management device 120 and the prediction device 1610 may be configured as an integrated entity.

[0263] Furthermore, in the second and fifth embodiments, the hardware configurations of the learning device 1510, the prediction device 1610, and the analysis device 2810 were not mentioned, but these hardware configurations may be the same as, for example, the data management device 120 shown in Fig. 3. Therefore, for example, in the prediction device 1610, the storage unit that stores the morphological analysis unit 1711, the vectorization model 1712, and the trained keyword model 1802 may be realized in the auxiliary storage device 303 shown in Fig. 3.

[0264] Furthermore, in the above fifth embodiment, the feedback information is described as being used to update the search unit 2101, but the feedback information may also be used, for example, to update the prediction unit 1611 described in the second embodiment.

[0265] [Seventh embodiment] In the above embodiments, the workers 131, 141 have been described as generating the response content of the input divided document data. However, instead of the workers 131, 141 generating the response content of the input divided document data, the response content may be generated by a language model (LM: Language Model) such as a large language model (LLM: Large Language Model). In this case, the workers 131, 141 only need to input the query content, check the response content generated by the language model, and correct it as necessary, which makes it possible to significantly reduce the workload for generating the input divided document data. The seventh embodiment will now be described, focusing on the differences from the above embodiments.

[0266] <System configuration of the business support system (LM learning phase 1)> First, a system configuration of a business support system according to the seventh embodiment will be described. In the seventh embodiment, the business support system executes processing in the LM learning phase and processing in the generation phase, so here, the system configuration of the business support system in the LM learning phase will be described.

[0267] FIG. 34 is a seventh diagram showing an example of the system configuration of a business support system, and shows the system configuration in the LM learning phase. The LM learning phase is a phase in which learning processing is executed for a language model. The business support system 3400 in FIG. 34 differs from the business support system 2100 described in the third embodiment with reference to FIG. 21 in that: The functions of the data management device 3410 differ from those of the data management device 120, The server device 150 does not have a search unit 2101. The external server device 3420 is connected to the external network 160. is.

[0268] Specifically, the data management device 3410 has an LM learning data generation unit 3411 (an example of a first generation unit). The LM learning data generation unit 3411 reads input divided document data stored in the database 151 of the server device 150, generates an LM learning dataset in which the query content is input data and the response content is correct answer data, and stores the generated dataset in the LM learning data storage unit 3412.

[0269] Furthermore, the LM learning data generation unit 3411 transmits the stored LM learning data set to the external server device 3420, thereby causing the external server device 3420 to execute learning processing for the language model it has.

[0270] The external server device 3420 functions as an LM service providing unit 3421. The LM service providing unit 3421 provides services corresponding to various language processing tasks by executing a language model. In the seventh embodiment, a learning process is performed on the language model held by the external server device 3420 using an LM training data set transmitted from the LM training data generating unit 3411. As a result, the language model is fine-tuned for generating answer content.

[0271] In the business support system 3400 shown in FIG. 34, the external server device 3420 is shown as being arranged on the external network 160. However, the external server device 3420 may be arranged, for example, on the internal network 170.

[0272] <Explanation of the LM learning dataset> Next, an LM learning dataset used when fine-tuning the language model possessed by the external server device 3420 for generating response content will be described. FIG. 35 is a first diagram showing an example of the LM learning dataset.

[0273] As shown in FIG. 35, the LM learning dataset 3500 includes, as information items, "number" and "input segmented document data". The "number" stores the number of the registered data 1000 stored in the database 151 of the server device 150. The "input segmented document data" further includes, as information items, "query content" and "response content".

[0274] The "query content" stores the query content included in the input segmented document data corresponding to the "number" among the input segmented document data included in the registered data 1000. The "response content" stores the response content corresponding to the query content included in the input segmented document data corresponding to the "number" among the input segmented document data included in the registered data 1000.

[0275] <Processing of the business support system (LM learning phase part 1)> Next, the processing flow in the LM learning phase by the business support system 3400 will be described. FIG. 36 is a first sequence diagram showing the processing flow in the LM learning phase by the business support system.

[0276] In step S3601, server device 150 transmits input divided document data including the inquiry content and the response content from registered data 1000 stored in database 151 to data management device 3410. Note that server device 150 may transmit input divided document data to data management device 3410 at predetermined intervals, or may transmit input divided document data each time new input divided document data is stored in registered data 1000.

[0277] In step S3602, data management device 3410 generates LM training data set 3500 based on the input divided document data transmitted from server device 150, and stores it in LM training data storage unit 3412.

[0278] In step S3603, the data management device 3410 transmits the LM training dataset 3500 stored in the LM training data storage unit 3412 to the external server device 3420. As a result, the data management device 3410 instructs the external server device 3420 to execute a training process for the language model it has. The data management device 3410 may transmit the LM training dataset 3500 to the external server device 3420 at predetermined intervals. Alternatively, the data management device 3410 may transmit the LM training dataset 3500 to the external server device 3420 every time input divided document data is transmitted from the server device 150 and the LM training dataset 3500 is updated.

[0279] In step S3604, the external server device 3420 performs a learning process on the language model using the LM learning dataset 3500 transmitted from the data management device 3410, thereby fine-tuning the language model for generating answer content.

[0280] <System configuration of the business support system (LM learning phase 2)> Next, as a business support system according to the seventh embodiment, a system configuration of a business support system different from the business support system shown in FIG. 34 will be described. FIG. 37A is an eighth diagram showing an example of the system configuration of a business support system, and shows another system configuration in the LM learning phase. The difference from the business support system 3400 described using FIG. 34 is that in the case of the business support system 3700 in FIG. 37A, The data management device 3410 has a first LM learning data generation unit 3411 (another example of the first generation unit) instead of the LM learning data generation unit 3411. The external server device 3420 includes a second LM learning data generation unit 3710 (an example of a second generation unit), is.

[0281] The first LM learning data generation unit 3411 reads out the input divided document data stored in the database 151 of the server device 150. Furthermore, the first LM learning data generation unit 3411 generates a first LM learning dataset from the read input divided document data, in which the query content is used as input data and the answer content is used as correct answer data, and stores the first LM learning dataset in the LM learning data storage unit 3412. Note that the first LM learning dataset is an example of a first learning dataset.

[0282] In addition, the first LM learning data generation unit 3411 transmits the stored first LM learning data set to the external server device 3420, thereby causing the LM service providing unit 3421 of the external server device 3420 to execute learning processing on the language model it has.

[0283] The second LM learning data generation unit 3710 included in the external server device 3420 further includes a learning target information collection unit 3711 and a self-supervised learning data generation unit 3712.

[0284] The learning subject information collecting unit 3711 accesses the Web server device 3720 via the external network 160 and collects learning subject information. The learning subject information refers to any information that contributes to the LM service providing unit 3421 generating an answer content based on the latest information. The learning subject information collecting unit 3711 is configured to automatically collect learning subject information from a predetermined access destination at a predetermined timing. Alternatively, the learning subject information collecting unit 3711 may be configured to collect learning subject information from the specified access destination at a timing specified by the applicant 122. Note that the learning subject information collecting unit 3711 may be configured to accept input by the applicant 122 if login is required to access the specified access destination.

[0285] The self-supervised learning data generation unit 3712 generates a second LM training dataset based on the training target information collected by the training target information collection unit 3711. The self-supervised learning data generation unit 3712 generates a second LM training dataset including, for example, masked documents in which part of the collected training target information is masked, and documents (correct answers documents) consisting of the collected training target information (unmasked training target information). The second LM training dataset is an example of a second training dataset.

[0286] The self-supervised learning data generation unit 3712 notifies the LM service providing unit 3421 of the generated second LM learning data set, and causes the LM service providing unit 3421 to execute learning processing for the language model it has.

[0287] In the business support system 3700 shown in FIG. 37A, the external server device 3420 is shown as being located on the external network 160, but the external server device 3420 may also be located on, for example, the internal network 170.

[0288] 37A, the second LM learning data generation unit 3710 is illustrated as being implemented in the external server device 3420. However, the second LM learning data generation unit 3710 may be implemented, for example, in any device (e.g., in the data management device 3410) located on the internal network 170. The second LM learning data generation unit 3710 may be implemented both in the external server device 3420 and in any device (e.g., in the data management device 3410) located on the internal network 170. In other words, the second LM learning data generation unit 3710 may be implemented in multiple devices.

[0289] Furthermore, in the business support system 3700 shown in FIG. 37A, the LM service providing unit 3421 has been described assuming that it has one language model, but the LM service providing unit 3421 may have three types of language models.

[0290] Specifically, the LM service providing unit 3421: A language model that is fine-tuned using the first LM training dataset; A language model that is fine-tuned using the second LM training dataset; a language model that is fine-tuned using both the first LM training dataset and the second LM training dataset; may have

[0291] <System configuration of the business support system (LM learning phase 3)> Next, a modified example of the business support system shown in FIG. 37A will be described as a business support system according to the seventh embodiment. FIG. 37B is a ninth diagram showing an example of the system configuration of the business support system, and shows another system configuration in the LM learning phase. The difference from the business support system 3700 described using FIG. 37A is that in the case of the business support system 3700 in FIG. 37B, The data management device 3410 has a first generation control unit 3721, · The external server device 3420 has a second generation control unit 3722. That's it.

[0292] The first generation control unit 3721 appropriately selects and discards the first LM learning data set stored in the LM learning data storage unit 3412 by referring to the learning target information collected by the learning target information collection unit 3711 of the external server device 3420. This is because the first LM learning data set includes data that is not appropriate for fine-tuning the language model for answer content generation in light of the latest information.

[0293] The second generation control unit 3722 appropriately changes the source of collection when the learning target information collection unit 3711 collects learning target information by referring to the first LM learning data set generated by the first LM learning data generation unit 3411 of the data management device 3410. This is because the learning target information that the learning target information collection unit 3711 should collect is important information for replacing the first LM learning data set with the latest information.

[0294] <Explanation of the LM learning data set> Next, the first LM learning data set and the second LM learning data set used when the external server device 3420 fine-tunes the language model for answer content generation will be described. FIG. 38 is a second diagram showing an example of the LM learning data set.

[0295] In FIG. 38, since the first LM learning data set 3500 is the same as the LM learning data set 3500 shown in FIG. 35, the description is omitted here.

[0296] In FIG. 38, the second LM learning data set 3820 is a learning data set generated by the self-supervised learning data generation unit 3712 based on the learning target information 3810 collected by the learning target information collection unit 3711.

[0297] The learning target information 3810 includes information items such as "number", "information name", "file name", "collection source", and "collection date and time".

[0298] The "number" stores a number assigned to each document collected as learning information. The "information name" stores the name of the document collected as learning information. In the example of FIG. 38, documents collected as learning information include, for example, -documents indicating drug guidelines, Drug development documents, ·Documents on industry activity deliverables; Pharmaceutical-related magazines, Medical / chemical journals, Journals that publish medical / chemical articles, etc. are included.

[0299] "File name" stores the file name of the document stored in "Information name." "Collection source" stores the access destination when the document stored in "Information name" was collected. "Collection date and time" stores the collection date and time when the document stored in "Information name" was collected.

[0300] The second LM training dataset 3820 includes information items such as "number" and "self-supervised training data." The "number" stores the number of the training target information used when the self-supervised training data generation unit 3712 generates the second LM training dataset.

[0301] The "self-supervised learning data" further includes the information items of a "masked document" and a "correct answer document." The "masked document" stores a partially masked masked document that is included in the learning target information and is identified by a corresponding "number" (e.g., "N1"). The "correct answer document" stores the corresponding masked document before it is masked. As shown in FIG. 38, a combination of multiple masked documents and multiple correct answer documents may be generated from one document of the learning target information.

[0302] When the learning target information 3810 is collected, the self-supervised learning data generation unit 3712 automatically (that is, without the applicant 122's involvement) generates the second LM learning dataset 3820.

[0303] <Business Support System Processing (LM Learning Phase 2)> Next, we will explain the processing flow in the LM learning phase by business support system 3700. Figure 39A is a second sequence diagram showing the processing flow in the LM learning phase by business support system 3700, and shows the processing flow in the LM learning phase by business support system 3700 shown in Figure 37A.

[0304] In FIG. 39A, the processing in steps S3601 to S3604 is generally the same as the processing in steps S3601 to S3604 in FIG. 36, and therefore a description thereof will be omitted here.

[0305] In step S3901, the external server device 3420 accesses the Web server device 3720 via the external network 160 and collects learning information.

[0306] In step S3902, the external server device 3420 generates a second LM training dataset based on the collected training subject information.

[0307] In step S3903, the external server device 3420 performs a learning process on the language model using the generated second LM learning data set, thereby fine-tuning the language model for generating answer content.

[0308] 39A, the transmission of the first LM training dataset by the data management device 3410 to the external server device 3420 may be executed at predetermined intervals. Alternatively, the transmission of the first LM training dataset by the data management device 3410 to the external server device 3420 may be executed every time input divided document data is transmitted from the server device 150 and the first LM training dataset is updated.

[0309] 39A, the external server device 3420 may execute the learning process using the second LM training dataset (step S3903) at predetermined intervals. The external server device 3420 may execute the learning process using the second LM training dataset (step S3903) at the same intervals as the learning process using the first LM training dataset (step S3604), or may execute the learning process at different intervals.

[0310] Furthermore, the external server device 3420 repeatedly executes a learning process using the first LM learning dataset (step S3604) and a learning process using the second LM learning dataset (step S3903). This enables the LM service providing unit 3421 to output answer content that takes into account the latest information on previously input divided document data. Alternatively, when outputting answer content that takes into account the latest information, the LM service providing unit 3421 can output answer content that takes into account the expression method of recent input divided document data.

[0311] <Business Support System Processing (LM Learning Phase 3)> Next, we will explain the processing flow in the LM learning phase by the business support system 3700. Figure 39B is a third sequence diagram showing the processing flow in the LM learning phase by the business support system, and shows the processing flow in the LM learning phase by the business support system 3700 shown in Figure 37B.

[0312] In FIG. 39B, the processing in steps S3601 to S3604 is generally the same as the processing in steps S3601 to S3604 in FIG. 36, and therefore a description thereof will be omitted here.

[0313] In step S3911, the external server device 3420 refers to the first LM learning dataset generated by the data management device 3410 to change the collection source when the learning object information collection unit 3711 collects learning object information.

[0314] The processing in steps S3901 to S3902 and S3903 is the same as the processing in steps S3901 to S3902 and S3903 in FIG. 39A, and therefore description thereof will be omitted here.

[0315] In step S3912, the external server device 3420 transmits the second LM training dataset generated in step S3902 to the data management device 3410.

[0316] In step S3913, the data management device 3410 selects the first LM training dataset by referring to the training target information included in the second LM training dataset.

[0317] <System configuration of business support system (generation phase)> Next, the system configuration in the generation phase of the business support system according to the seventh embodiment will be described. Fig. 40 is a tenth diagram showing an example of the system configuration of the business support system, and shows the system configuration in the generation phase. The generation phase is a phase in which input document data is generated using a language model fine-tuned for generating answer content. The difference from the business support system 3400 in the LM learning phase described using Fig. 34 is that in the case of the business support system 4000 in Fig. 40, The functions of the data management device 4010 differ from those of the data management device 3410, The data management device 4010 has an LM utilization unit 4011. The data management device 4010 does not have the LM learning data storage unit 3412. is.

[0318] 40, when authority inquiry document data is transmitted from the authority terminal 110 ((1)), the management unit 121 of the data management device 4010 generates a plurality of divided document data by dividing the received authority inquiry document data for each inquiry item. The management unit 121 of the data management device 4010 also transmits the generated plurality of divided document data to the corresponding worker terminals 130, 140, etc. ((2)).

[0319] When the divided document data is sent from the data management device 4010, the worker 131 extracts the inquiry content from the divided document data and generates a prompt including the extracted inquiry content. The worker 131 also sends the generated prompt to the LM utilization unit 4011 of the data management device 4010 via the worker terminal 130, thereby instructing the generation of an answer content to the extracted inquiry content ((3)).

[0320] The LM utilization unit 4011 transmits the inquiry to the external server device 3420 based on the prompt received from the operator terminal 130, and instructs the external server device 3420 to generate a response to the transmitted inquiry.

[0321] The LM service providing unit 3421 of the external server device 3420 has a language model that is fine-tuned for generating answer content. When the LM service providing unit 3421 of the external server device 3420 receives an inquiry and an instruction to generate an answer content for the inquiry content from the data management device 4010, the LM service providing unit 3421 generates the answer content in accordance with the instruction to generate the answer content.

[0322] The answer content generated by the external server device 3420 is acquired by the LM utilization unit 4011 of the data management device 4010 ((4)) and sent to the instructing worker terminal 130. As a result, the worker 131 inputs the answer content to the inquiry content while referring to the answer content generated by the language model fine-tuned for answer content generation, and generates input divided document data.

[0323] The input divided document data generated by the worker 131 is transmitted to the management unit 121 of the data management device 4010 via the worker terminal 130 ((5)).

[0324] The management unit 121 of the data management device 4010 generates input document data based on the input divided document data collected from the worker terminal 130, worker terminal 140, etc., and transmits the data to the authority terminal 110 via the external network 160 ((6)).

[0325] In this way, according to the task support system 4000, the workers 131, 141, etc. can generate input divided document data while referring to the answer content acquired using a language model fine-tuned for answer content generation. In other words, the task support system 4000 can support the workers 131, 141, etc. in generating input divided document data.

[0326] <Business support system processing (generation phase)> Next, a description will be given of the flow of processing in the generation phase by the task support system 4000. Fig. 41 is a first sequence diagram showing the flow of processing in the generation phase by the task support system.

[0327] Of these, the processes in steps S1301 to S1303 and S1305 to S1309 are the same as the processes in steps S1301 to S1303 and S1305 to S1309 in FIG. 13, and therefore will not be described here.

[0328] In step S4101, the worker terminals 130, 140, etc. each display the received divided document data.

[0329] In step S4102, when a prompt including an inquiry and an instruction to generate a response to the inquiry is input by the worker 131, 141, etc. in response to displaying the divided document data, the worker terminal 130, 140, etc. accepts the input. The worker terminal 130, 140, etc. also transmits the inputted prompt to the data management device 4010.

[0330] In step S4103, the data management device 4010 transmits to the external server device 3420 the inquiry content and an instruction to generate a response content to the inquiry content.

[0331] In step S4104, upon receiving the inquiry and an instruction to generate an answer to the inquiry from the data management device 4010, the external server device 3420 executes a language model fine-tuned for answer generation to generate an answer. The external server device 3420 also transmits the generated answer to the data management device 4010.

[0332] In step S4105, the data management device 4010 receives the response content transmitted from the external server device 3420 and transmits it to the worker terminals 130, 140, etc.

[0333] In step S4106, the worker terminals 130, 140, etc. display the response content transmitted from the data management device 4010. This allows the workers 131, 141, etc. to generate input divided document data while referring to the response content displayed on the worker terminals 130, 140, etc.

[0334] <Summary> As is clear from the above description, the business support system 3400 according to the seventh embodiment: A learning dataset is generated in which the inquiry content and the response content contained in the plurality of input divided document data are associated with each other, and the generated learning dataset is transmitted to execute a learning process for the language model.

[0335] Furthermore, the business support system 4000 according to the seventh embodiment includes: When multiple divided document data are generated for the authority inquiry document data to be processed, the inquiry content for each of the multiple inquiry items is sent to the language model that has undergone learning processing, and the response content for each of the inquiry content for each of the multiple inquiry items is obtained. A plurality of input divided document data items generated based on the acquired response contents are acquired.

[0336] As a result, the work support system according to the seventh embodiment can support the work of the worker who generates input divided document data.

[0337] [Eighth embodiment] In the seventh embodiment, the LM training dataset 3500 including the inquiry content and the response content is used when fine-tuning the language model. However, the LM training dataset used when fine-tuning the language model is not limited to the LM training dataset 3500. For example, an LM training dataset including an authority-submitted document (an example of submitted document data) that was submitted to the authority before the authority inquiry document data was transmitted from the authority terminal 110 and the inquiry content transmitted from the authority terminal 110 may be used. The eighth embodiment will be described below, focusing on the differences from the seventh embodiment.

[0338] <System configuration of business support system (LM learning phase)> First, the system configuration of a business support system according to the eighth embodiment will be described. In the eighth embodiment, the business support system executes processing in the LM learning phase, processing in the pharmaceutical application phase, and processing in the pharmaceutical inquiry phase. Here, the system configuration of the business support system in the LM learning phase will be described.

[0339] Fig. 42 is an eleventh diagram showing an example of the system configuration of the business support system, and shows the system configuration in the LM learning phase. The difference from the business support system 3400 explained in the seventh embodiment with reference to Fig. 34 is that the business support system 4200 in Fig. 42 The functions of the data management device 4210 differ from those of the data management device 3410, The server device 4220 is connected to the internal network 170. is.

[0340] Specifically, the data management device 4210 has an LM learning data generation unit 4211. The LM learning data generation unit 4211 reads out the official-submitted materials stored in the official-submitted material storage unit 4221 of the server device 4220.

[0341] The LM learning data generation unit 4211 also reads out the inquiry content included in the authority inquiry document data sent from the authority in response to the submission of the authority-submitted materials to the authority from the database 151 of the server device 150. Furthermore, the LM learning data generation unit 4211 generates an LM learning dataset using the read authority-submitted materials as input data and the inquiry content as correct answer data, and stores the dataset in the LM learning data storage unit 4212.

[0342] Furthermore, the LM learning data generation unit 4211 transmits the stored LM learning data set to the external server device 3420, thereby causing the external server device 3420 to execute learning processing for the language model it has.

[0343] In the eighth embodiment, a learning process is performed on the language model held by the external server device 3420 using an LM training data set transmitted from the LM training data generation unit 4211. As a result, the language model is fine-tuned for query content generation.

[0344] In the eighth embodiment, the "materials submitted to the authority" include, for example, in addition to the "main materials" such as the application documents for the approval of the manufacture and sale of pharmaceuticals, etc., the "supplementary materials" submitted to the authority in connection with the submission of the application documents for the approval of the manufacture and sale of pharmaceuticals, etc. The "supplementary materials" refer to, for example, consultation records, test result data, etc. However, even if the materials correspond to consultation records, test result data, etc., the internal materials that have not been submitted to the authority as "supplementary materials" are not included in the LM learning data set.

[0345] In addition, when the materials submitted to the authority include data other than character information, such as images, graphs, tables, etc., the LM learning data generation unit 4211 may attach annotations to the data other than the character information to generate the LM learning data set. Alternatively, assuming that the materials submitted to the authority include data other than character information, such as images, graphs, tables, etc., the LM service providing unit 3421 may have a multimodal model instead of the language model.

[0346] <Explanation of the LM learning data set> Next, an explanation will be given of the LM learning data set used when the language model possessed by the external server device 3420 is fine-tuned for query content generation. FIG. 43 is a third diagram showing an example of the LM learning data set.

[0347] As shown in FIG. 43, the LM learning data set 4300 includes, as information items, "number", "materials submitted to the authority", and "query content". The "number" stores the number assigned to each of the materials submitted to the authority. The "materials submitted to the authority" stores the main materials submitted to the authority and the supplementary materials submitted to the authority in connection with the submission of the main materials. The "query content" stores each query content included in the authority query document data transmitted from the authority in response to the submission of the main materials and the supplementary materials to the authority.

[0348] <Processing of the business support system (LM learning phase)> Next, a description will be given of the flow of processing in the LM learning phase by the task support system 4000. Fig. 44 is a fourth sequence diagram showing the flow of processing in the LM learning phase by the task support system.

[0349] In step S4401, the server device 4220 transmits the official-submitted materials stored in the official-submitted material storage unit 4221 to the data management device 4210. The transmission of the official-submitted materials by the server device 4220 to the data management device 4210 may be performed at predetermined intervals, or may be performed each time new official-submitted materials are stored in the official-submitted material storage unit 4221.

[0350] In step S4402, server device 150 transmits the authority inquiry document data from registration data 1000 stored in database 151 to data management device 4210. Note that server device 150 may transmit the authority inquiry document data to data management device 4210 at predetermined intervals, or may transmit the data every time input divided document data is newly stored in registration data 1000.

[0351] In step S4403, the data management device 4210 generates an LM training dataset 4300 based on the official submission materials transmitted from the server device 4220 and the official inquiry document data transmitted from the server device 150. The data management device 3410 also stores the generated LM training dataset 4300 in the LM training data storage unit 4212.

[0352] In step S4404, the data management device 4210 transmits the LM training dataset 4300 stored in the LM training data storage unit 4212 to the external server device 3420. As a result, the data management device 4210 executes a training process for the language model held by the external server device 3420. Note that the data management device 4210 may transmit the LM training dataset 4300 to the external server device 3420 at predetermined intervals, or may transmit the LM training dataset 4300 every time the LM training dataset 4300 is updated.

[0353] In step S4405, the external server device 3420 performs a learning process on the language model using the LM training dataset 4300 transmitted from the data management device 4210, thereby fine-tuning the language model for query content generation.

[0354] <System configuration of business support system (pharmaceutical application phase)> Next, the system configuration in the pharmaceutical application phase of the business support system according to the eighth embodiment will be described. Fig. 45 is a twelfth diagram showing an example of the system configuration of the business support system, and shows the system configuration in the pharmaceutical application phase. The pharmaceutical application phase is a phase in which documents to be submitted to the authorities when applying for approval to manufacture and sell pharmaceuticals, etc. are generated using a language model fine-tuned for generating inquiry content. The difference from the business support system 4200 in the LM learning phase described using Fig. 42 is that in the case of the business support system 4500 in Fig. 45, The functions of the data management device 4510 differ from those of the data management device 4210, The data management device 4510 has a material generation unit 4511 and an LM utilization unit 4512. The data management device 4510 does not have the LM learning data storage unit 4212. is.

[0355] 45, first, the applicant 122 creates a draft document to be submitted to the authorities when applying for approval to manufacture and sell a drug, etc. ((1)) via the document generation unit 4511. Next, the applicant 122 generates a prompt including the generated draft document to be submitted to the authorities, and inputs the generated prompt to the LM usage unit 4512 to instruct the generation of inquiry content regarding the draft document to be submitted to the authorities.

[0356] Based on the prompt input by the applicant 122, the LM utilization unit 4512 transmits the draft documents to be submitted to the authority to the external server device 3420 and instructs the external server device 3420 to generate inquiry content for the transmitted draft documents to be submitted to the authority.

[0357] The LM service providing unit 3421 of the external server device 3420 has a language model that is fine-tuned for generating query contents. When the LM service providing unit 3421 of the external server device 3420 receives from the data management device 4510 a draft document to be submitted to the authority and an instruction to generate query contents for the draft document to be submitted to the authority, the LM service providing unit 3421 generates query contents in accordance with the instruction to generate query contents.

[0358] The inquiry content generated by the external server device 3420 is acquired by the LM utilization unit 4512 of the data management device 4510 ((2)) and displayed to the applicant 122. This allows the applicant 122 to revise the draft documents to be submitted to the authorities while referring to the inquiry content for the draft documents to be submitted to the authorities. Specifically, the applicant 122 revise the main documents and delete or add supplementary documents so that the authorities do not send the authority inquiry document data for the displayed inquiry content ((3)).

[0359] Next, the applicant 122 generates a prompt including the revised draft documents to be submitted to the authority, and inputs the generated prompt into the LM usage unit 4512 to instruct the generation of inquiry content for the revised draft documents to be submitted to the authority.

[0360] Based on the prompt entered by the applicant 122, the LM utilization unit 4512 sends the revised draft of the documents to be submitted to the authority to the external server device 3420 and instructs the external server device 3420 to generate inquiry content for the revised draft of the documents to be submitted to the authority.

[0361] When the LM service providing unit 3421 of the external server device 3420 receives from the data management device 4510 the revised draft documents to be submitted to the authorities and an instruction to generate inquiry content for the revised draft documents to be submitted to the authorities, the LM service providing unit 3421 generates the inquiry content in accordance with the instruction to generate the inquiry content.

[0362] The inquiry content generated by the external server device 3420 is acquired by the LM utilization unit 4512 of the data management device 4510 ((4)) and displayed to the applicant 122. As a result, the applicant 122 revises the revised draft documents to be submitted to the authorities while referring to the inquiry content for the revised draft documents to be submitted to the authorities. Specifically, the applicant 122 revises the main documents and deletes or adds supplementary documents so that the authorities do not send the authority inquiry document data for the displayed inquiry content ((5)).

[0363] Next, the applicant 122 generates a prompt including the revised draft documents to be submitted to the authority, and inputs the generated prompt into the LM usage unit 4512 to instruct the generation of inquiry content regarding the revised draft documents to be submitted to the authority.

[0364] Based on the prompt entered by the applicant 122, the LM utilization unit 4512 sends the revised draft documents to the external server device 3420 and instructs the external server device 3420 to generate inquiry content for the revised draft documents to be submitted to the authority.

[0365] When the LM service providing unit 3421 of the external server device 3420 receives from the data management device 4510 the revised draft documents to be submitted to the authorities and an instruction to generate inquiry content for the revised draft documents to be submitted to the authorities, the LM service providing unit 3421 generates the inquiry content in accordance with the generation instruction.

[0366] The inquiry content generated by the external server device 3420 is acquired by the LM utilization unit 4512 of the data management device 4510 ((6)) and displayed to the applicant 122. As a result, the applicant 122 refers to the inquiry content for the revised draft documents to be submitted to the authorities and again revise the revised draft documents to be submitted to the authorities. Specifically, the applicant 122 revise the main documents and delete or add supplementary documents ((7)) so that the authorities do not send the authority inquiry document data for the displayed inquiry content, and complete the documents to be submitted to the authorities.

[0367] The completed documents to be submitted to the authorities are sent to the authorities' terminal 110 via the document generation unit 4511, and the application for approval to manufacture and sell pharmaceuticals, etc. is completed ((8)).

[0368] In this way, the business support system 4500 allows the applicant 122 to generate documents to be submitted to authorities while referencing the inquiry content obtained using a language model fine-tuned for generating inquiry content. In other words, the business support system 4500 can support the applicant 122 in the task of generating documents to be submitted to authorities.

[0369] <Business support system processing (pharmaceutical application phase)> Next, a description will be given of the flow of processing in the pharmaceutical application phase by the business support system 4500. Fig. 46 is a sequence diagram showing the flow of processing in the pharmaceutical application phase by the business support system.

[0370] In step S4601, the data management device 4510 accepts a prompt including a draft of documents to be submitted to the authorities, if one is input, generated by the applicant 122. The data management device 4510 also transmits the draft of documents to the authorities and an instruction to generate inquiry content for the draft of documents to be submitted to the authorities to the external server device 3420. Note that the draft of documents to be submitted to the authorities generated by the applicant 122 may be generated based on the draft partial documents to be submitted to the authorities generated by the worker 131 or the worker 141 via the worker terminal 130 or the worker terminal 140.

[0371] In step S4602, upon receiving the draft documents to be submitted to the authority and an instruction to generate a query for the draft documents to be submitted to the authority from the data management device 4510, the external server device 3420 executes a language model fine-tuned for generating a query. As a result, the external server device 3420 generates a query. The external server device 3420 also transmits the generated query to the data management device 4510.

[0372] In step S4603, if a prompt including the draft authority submission documents revised by the applicant 122 is input, the data management device 4510 accepts this. The data management device 4510 also transmits the revised draft authority submission documents and an instruction to generate inquiry content for the revised draft authority submission documents to the external server device 3420. Note that the draft authority submission documents revised by the applicant 122 may be revised based on the draft partial authority submission documents revised by the operator 131 or operator 141 via the operator terminal 130 or operator terminal 140.

[0373] In step S4604, the external server device 3420 generates the inquiry content upon receiving the revised draft documents to be submitted to the authority and the instruction to generate the inquiry content for the revised draft documents to be submitted to the authority from the data management device 4510. In addition, the external server device 3420 transmits the generated inquiry content to the data management device 4510.

[0374] In step S4605, if a prompt including the draft authority submission documents revised by the applicant 122 is input, the data management device 4510 accepts this. The data management device 4510 also transmits the revised draft authority submission documents and an instruction to generate inquiry content for the revised draft authority submission documents to the external server device 3420. Note that the draft authority submission documents revised by the applicant 122 may be revised based on the draft partial authority submission documents revised by the operator 131 or the operator 141 via the operator terminal 130 or the operator terminal 140.

[0375] In step S4606, the external server device 3420 generates the inquiry content upon receiving the revised draft documents to be submitted to the authority and the instruction to generate the inquiry content for the revised draft documents to be submitted to the authority from the data management device 4510. The external server device 3420 also transmits the generated inquiry content to the data management device 4510.

[0376] In step S4607, the data management device 4510 accepts the completed documents to be submitted to the authorities through third revision by the applicant 122. The data management device 4510 also submits the completed documents to the authorities by transmitting them to the authority terminal 110. Note that the draft documents to be submitted to the authorities that are third revised by the applicant 122 may be third revised based on the draft partial documents to be submitted to the authorities that have been third revised by the operator 131 or 141 via the operator terminal 130 or 140.

[0377] <System configuration of the business support system (pharmaceutical affairs inquiry phase)> Next, the system configuration of the pharmaceutical affairs inquiry phase of the business support system according to the eighth embodiment will be described. Fig. 47 is the thirteenth diagram showing an example of the system configuration of the business system, and shows the system configuration in the pharmaceutical affairs inquiry phase. The pharmaceutical affairs inquiry phase is a phase in which input document data is generated using a language model fine-tuned for generating inquiry content. It is basically the same as the generation phase described in the seventh embodiment, but differs in that the language model used when generating input document data is fine-tuned for generating inquiry content. Therefore, the business support system 4000 in the generation phase described using Fig. 40 is The functions of the data management device 4710 are different from those of the data management device 4010. The difference is that a server device 4220 is connected.

[0378] 47, when authority inquiry document data is transmitted from the authority terminal 110 ((1)), the management unit 121 of the data management device 4710 generates a plurality of divided document data by dividing the received authority inquiry document data for each inquiry item. The management unit 121 of the data management device 4710 also transmits the generated plurality of divided document data to the corresponding worker terminals 130, 140, etc. ((2)).

[0379] When the divided document data is sent from the data management device 4710, the worker 131 extracts the inquiry content from the divided document data and generates a prompt including the extracted inquiry content. The worker 131 also instructs the generation of an answer content to the extracted inquiry content by sending the generated prompt to the LM utilization unit 4711 of the data management device 4710 via the worker terminal 130 ((3)).

[0380] The LM utilization unit 4711 transmits the inquiry to the external server device 3420 based on the prompt received from the operator terminal 130, and instructs the external server device 3420 to generate a response to the transmitted inquiry.

[0381] When the LM service providing unit 3421 of the external server device 3420 receives the inquiry content and an instruction to generate a response content to the inquiry content from the data management device 4710, the LM service providing unit 3421 generates the response content in accordance with the instruction to generate the response content.

[0382] The response content generated by the external server device 3420 is acquired by the LM utilization unit 4711 of the data management device 4710 ((4)) and sent to the instructing worker terminal 130. As a result, the worker 131 inputs the response content to the inquiry content while referring to the response content generated by the language model, and generates input divided document data.

[0383] Next, the worker 131 generates a prompt including the generated input divided document data. The worker 131 also sends the generated prompt to the LM utilization unit 4711 of the data management device 4710 via the worker terminal 130, thereby instructing the generation of query content for the generated input divided document data ((5)).

[0384] Based on the prompt received from the operator terminal 130, the LM utilization unit 4711 transmits the input divided document data to the external server device 3420 and instructs the external server device 3420 to generate query content for the input divided document data that has been transmitted.

[0385] When the LM service providing unit 3421 of the external server device 3420 receives the input divided document data and an instruction to generate a query for the input divided document data from the data management device 4710, the LM service providing unit 3421 generates a query in accordance with the instruction.

[0386] The inquiry content generated by the external server device 3420 is acquired by the LM utilization unit 4711 of the data management device 4710 ((6)), and is sent to the worker terminal 130 that issued the instruction.

[0387] Next, the worker 131 generates a prompt including the inquiry content, and sends the generated prompt to the LM usage unit 4711 of the data management device 4710 via the worker terminal 130, thereby instructing the generation of an answer content to the inquiry content ((7)).

[0388] The LM utilization unit 4711 transmits the inquiry to the external server device 3420 based on the prompt received from the operator terminal 130, and instructs the external server device 3420 to generate a response to the transmitted inquiry.

[0389] When the LM service providing unit 3421 of the external server device 3420 receives the inquiry content and an instruction to generate a response content to the inquiry content from the data management device 4710, the LM service providing unit 3421 generates the response content in accordance with the instruction to generate the response content.

[0390] The response content generated by the external server device 3420 is acquired by the LM utilization unit 4711 of the data management device 4710 ((8)) and sent to the instructing worker terminal 130. As a result, the worker 131 inputs the response content to the inquiry content while referring to the response content generated by the language model, and generates input divided document data.

[0391] Next, the worker 131 transmits the generated input divided document data to the management unit 121 of the data management device 4710 via the worker terminal 130 ((9)).

[0392] The management unit 121 of the data management device 4710 generates input document data based on the input divided document data collected from the worker terminal 130, worker terminal 140, etc., and transmits the data to the authority terminal 110 via the external network 160 ((10)).

[0393] In this way, according to the business support system 4700, the worker 131, worker 141, etc., when generating input divided document data, uses a language model fine-tuned for generating inquiry content to repeatedly acquire inquiry content and response content. This allows the worker 131, worker 141, etc. to generate input divided document data that is less likely to be subject to further inquiries from authorities. In other words, the business support system 4700 can support the worker 131, worker 141, etc. in the task of generating input divided document data.

[0394] <Business support system processing (pharmaceutical inquiry phase)> Next, a description will be given of the flow of processing in the pharmaceutical affairs inquiry phase by the business support system 4700. Figures 48A and 48B are first and second sequence diagrams showing the flow of processing in the pharmaceutical affairs inquiry phase by the business support system.

[0395] Of these, the processes in steps S1301 to S1303 and S1305 to S1309 are the same as the processes in steps S1301 to S1303 and S1305 to S1309 in FIG. 13, and therefore will not be described here.

[0396] In step S4801, the worker terminals 130, 140, etc. each display the received divided document data.

[0397] In step S4802, when a prompt including an instruction to generate a response to the inquiry is input by the worker 131, 141, etc. in response to displaying the divided document data, the worker terminal 130, 140, etc. accepts this. In addition, the worker terminal 130, 140, etc. transmits the inputted prompt to the data management device 4710.

[0398] In step S4803, the data management device 4710 transmits to the external server device 3420 the inquiry content and an instruction to generate a response content to the inquiry content.

[0399] In step S4804, the external server device 3420 generates the response content upon receiving the inquiry content and the instruction to generate the response content for the inquiry content from the data management device 4710. In addition, the external server device 3420 transmits the generated response content to the data management device 4710.

[0400] In step S4805, the data management device 4710 receives the reply sent from the external server device 3420 and sends it to the worker terminals 130, 140, etc.

[0401] In step S4806, the worker terminals 130, 140, etc. display the response content transmitted from the data management device 4710. This allows the workers 131, 141, etc. to generate input divided document data while referring to the response content displayed on the worker terminals 130, 140, etc.

[0402] In step S4807, when a prompt including an instruction to generate query content for the input divided document data generated by the workers 131, 141, etc. is input, the worker terminal 130, 140, etc. accepts the input. Furthermore, the worker terminal 130, 140, etc. transmits the inputted prompt to the data management device 4710.

[0403] In step S4808, data management device 4710 transmits to external server device 3420 the input divided document data and an instruction to generate query content for the input divided document data.

[0404] 48B, external server device 3420 generates the query content upon receiving the input divided document data and an instruction to generate the query content for the input divided document data from data management device 4710. In addition, external server device 3420 transmits the generated query content to data management device 4710.

[0405] In step S4810, the data management device 4710 receives the inquiry content transmitted from the external server device 3420 and transmits it to the worker terminals 130, 140, etc.

[0406] In step S4811, the worker terminal 130, 140, etc. displays the inquiry content transmitted from the data management device 4710.

[0407] In step S4812, when a prompt including an inquiry and an instruction to generate a response to the inquiry is input by the worker 131, 141, etc., the worker terminal 130, 140, etc. accepts the input. In addition, the worker terminal 130, 140, etc. transmits the inputted prompt to the data management device 4710.

[0408] In step S4813, the data management device 4710 transmits to the external server device 3420 the inquiry content and an instruction to generate a response content to the inquiry content.

[0409] In step S4814, the external server device 3420 generates the response content upon receiving the inquiry content and the instruction to generate the response content for the inquiry content from the data management device 4710. In addition, the external server device 3420 transmits the generated response content to the data management device 4710.

[0410] In step S4815, the data management device 4710 receives the reply sent from the external server device 3420 and sends it to the worker terminals 130, 140, etc.

[0411] In step S4816, the worker terminals 130, 140, etc. display the response content transmitted from the data management device 4710. This allows the workers 131, 141, etc. to generate input divided document data while referring to the response content displayed on the worker terminals 130, 140, etc.

[0412] <Summary> As is clear from the above description, the business support system 4200 according to the eighth embodiment: A learning dataset is generated in which the documents submitted by the authorities are associated with the inquiry content of each of the multiple inquiries contained in the authority inquiry document data, and the generated learning dataset is transmitted to execute a learning process for the language model.

[0413] Furthermore, the business support system 4500 according to the eighth embodiment includes: When a draft document to be submitted to the authorities is generated and the generated draft document to be submitted to the authorities is sent to a language model that has undergone training, A process of obtaining a query; sending the revised draft submission document to the language model that has undergone training; Repeat. · Repeatedly retrieving the inquiry content, repeatedly correcting it, and then sending the completed submission documents to the authorities.

[0414] Furthermore, the business support system 4700 according to the eighth embodiment includes: The inquiry content contained in the multiple divided document data generated by dividing the authority inquiry document data to be processed is sent to a language model that has undergone learning processing, and the response content to the inquiry content is obtained. -If multiple input divided document data are generated based on the obtained answers, A process of transmitting a plurality of pieces of input segmented document data to a language model that has undergone a learning process, and acquiring a query content for the plurality of pieces of input segmented document data; A process of sending the acquired query content to a language model that has undergone learning processing, and acquiring a response content to the acquired query content. Repeat. · Generate input document data based on the multiple input divided document data completed by repeatedly obtaining the response content, and send it to the authority.

[0415] As a result, the business support system according to the eighth embodiment can support the work of a worker who generates input divided document data.

[0416] [Ninth embodiment] In the seventh and eighth embodiments, a learning process is performed on the language model in advance and the language model is fine-tuned so that an appropriate answer can be obtained from the language model. In contrast, in the ninth embodiment, an appropriate answer can be obtained from the language model regardless of whether the language model is fine-tuned or not. Specifically, the business support system according to the ninth embodiment is configured to transmit auxiliary information useful for generating an appropriate answer when transmitting an instruction to generate an answer content, so that the language model can generate the answer content while referring to the auxiliary information. The ninth embodiment will be described below, focusing on the differences from the seventh or eighth embodiment.

[0417] <System configuration of business support system (search phase)> First, the system configuration of a business support system according to a ninth embodiment will be described. In the ninth embodiment, the business support system executes a search phase process for searching for auxiliary information useful for generating an answer, and a generation phase process for generating input document data based on the answer generated by referring to the auxiliary information. Here, the system configuration of the business support system in the search phase will be described.

[0418] 49A is a fourteenth diagram showing an example of the system configuration of the business support system, and shows the system configuration in the search phase. The difference from the business support system 4000 described in the seventh embodiment with reference to FIG. 40 is that the business support system 4900 in FIG. 49A The functions of the data management device 4910 differ from those of the data management device 4010, The data management device 4910 has an agent unit 4911. The server device 150 and the server device 4930 are connected to the internal network 170. is.

[0419] 49A, when authority inquiry document data is transmitted from authority terminal 110 ((1)), management unit 121 of data management device 4910 generates a plurality of divided document data by dividing the received authority inquiry document data by inquiry item. In addition, management unit 121 of data management device 4910 transmits the generated plurality of divided document data to corresponding worker terminals 130, 140, etc. ((2)).

[0420] When the divided document data is sent from the data management device 4910, the worker 131 extracts the inquiry content from the divided document data and generates a prompt including the extracted inquiry content. The worker 131 also sends the generated prompt to the agent unit 4911 of the data management device 4910 via the worker terminal 130, thereby instructing a search for auxiliary information that is useful for generating a response content to the extracted inquiry content ((3)).

[0421] The agent unit 4911 determines a search destination based on the prompt received from the operator terminal 130, and transmits a search request for auxiliary information to the determined search destination. In the example of FIG. 49A, the following search destinations are used when searching for auxiliary information: A server device 150 in which registration data including divided document data, input divided document data, input document data, etc. is stored; Any server device 4930 other than the server device 150 (a server device 4930 having a database 4931 other than a specific database (other than the database 151)) connected to the internal network 170, Any web server device 3720 accessible via the external network 160; This indicates that the following are included:

[0422] Agent unit 4911 transmits a search request for auxiliary information to server device 150, server device 4930, and Web server device 3720 ((4a), (4b), (4c)).

[0423] The agent unit 4911 may be configured to send a search request to a preset search destination, or may be configured to narrow down the preset search destinations in accordance with the inquiry content included in the prompt and then send the search request.

[0424] Alternatively, the operator 131 may generate a prompt for guiding the search destination and transmit it to the agent unit 4911. The prompt for guiding the search destination refers to a prompt that specifies an object to be included in the search destination, a prompt that specifies an object to be excluded from the search destination, or the like.

[0425] Furthermore, the worker 131 may generate a prompt that guides the search content and transmit it to the agent unit 4911. The prompt that guides the search content refers to a prompt that specifies materials to be searched (for example, the latest guidelines) or a prompt that specifies materials to be excluded.

[0426] <System configuration of business support system (generation phase)> Next, the system configuration of the business support system in the generation phase will be described. Figure 49B is a fifteenth diagram showing an example of the system configuration of the business support system, and shows the system configuration in the generation phase.

[0427] 49B, in response to a search request from agent unit 4911, server device 150 transmits search results to agent unit 4911 ((5a)). The search results transmitted by server device 150 include, for example, answer contents associated with past inquiries similar to the inquiry content included in the search request. The answer contents transmitted by server device 150 as search results are answer contents that serve as reference when the language model generates answer contents (referred to as reference answer contents).

[0428] In response to a search request from the agent unit 4911, the server device 4930 transmits the search results to the agent unit 4911 ((5b)). The search results transmitted by the server device 4930 include, for example, enterprise search results. The enterprise search results refer to any search results that are obtained by searching the vast amount of digital data scattered throughout a company.

[0429] In response to a search request from the agent section 4911, the Web server device 3720 transmits the Web search results to the agent section 4911 ((5c)).

[0430] The agent unit 4911 transmits the search results sent from the server device 150, the enterprise search results sent from the server device 4930, and the web search results sent from the web server device 3720 as auxiliary information to the worker terminal 130 ((6)).

[0431] When the worker terminal 130 receives the auxiliary information from the agent unit 4911, the worker 131 selects the auxiliary information as appropriate. The worker 131 also generates a prompt including the inquiry content and the selected auxiliary information, and transmits the prompt to the agent unit 4911 of the data management device 4910 via the worker terminal 130 ((7)). Note that the agent unit 4911 may select the auxiliary information and generate a prompt including the inquiry content and the selected auxiliary information.

[0432] When the agent unit 4911 receives a prompt including the inquiry content and the selected auxiliary information from the worker terminal 130, it instructs the external server device 3420 to generate a response to the inquiry content while referring to the auxiliary information. Note that the agent unit 4911 may transmit a prompt including the inquiry content and the selected auxiliary information by the agent unit 4911, and instruct the external server device 3420 to generate a response to the inquiry content while referring to the auxiliary information.

[0433] When the LM service providing unit 3421 of the external server device 3420 receives from the data management device 4910 the inquiry content, auxiliary information, and a generation instruction to generate an answer content to the inquiry content by referring to the auxiliary information, the LM service providing unit 3421 generates the answer content in accordance with the generation instruction.

[0434] The response content generated by the external server device 3420 is acquired by the agent unit 4911 of the data management device 4910 ((8)) and sent to the instructing worker terminal 130. As a result, the worker 131 inputs the response content to the inquiry content while referring to the response content generated by the language model, and generates input divided document data.

[0435] The input divided document data generated by the worker 131 is transmitted to the management unit 121 of the data management device 4910 via the worker terminal 130 ((9)).

[0436] The management unit 121 of the data management device 4910 generates input document data based on the input divided document data collected from the worker terminal 130, worker terminal 140, etc., and transmits the data to the authority terminal 110 via the external network 160 ((10)).

[0437] In this way, according to the business support system 4900, the worker 131, worker 141, etc. can send auxiliary information that is useful for generating an appropriate answer when sending an instruction to generate an answer. Alternatively, according to the business support system 4900, the agent unit 4911 can select the auxiliary information to use, generate a prompt including the inquiry content and the selected auxiliary information, and instruct the external server device 3420 to generate an answer to the inquiry content while referring to the auxiliary information.

[0438] As a result, the language model can generate an answer content while referring to the auxiliary information according to the task support system 4900. As a result, the worker 131, worker 141, etc. can obtain an appropriate answer content regardless of whether the language model is fine-tuned or not.

[0439] In the description of FIGS. 49A and 49B, the worker 131 separately generates a prompt for searching for auxiliary information and a prompt for generating an answer. However, the worker 131 may generate a prompt for searching for auxiliary information and a prompt for generating an answer as a single integrated prompt. Alternatively, the agent unit 4911 may generate a prompt for searching for auxiliary information and a prompt for generating an answer as a single integrated prompt without involving the worker 131. In this case, upon completing collection of auxiliary information, the agent unit 4911 automatically transmits the collected auxiliary information together with the inquiry content to the external server device 3420. Furthermore, upon receiving the answer content from the external server device 3420, the agent unit 4911 transmits the answer content to the worker terminal 130 that issued the instruction.

[0440] <Business support system processing (search phase and generation phase)> Next, a description will be given of the processing flow in the search phase and the generation phase by the task support system 4900. Figures 50A and 50B are first and second sequence diagrams showing the processing flow in the search phase and the generation phase by the task support system.

[0441] Of these, the processes in steps S1301 to S1303 and S1305 to S1309 are the same as the processes in steps S1301 to S1303 and S1305 to S1309 in FIG. 13, and therefore will not be described here.

[0442] In step S5001, the worker terminals 130, 140, etc. each display the received divided document data.

[0443] In step S5002, when a prompt for instructing a search for auxiliary information useful for generating a response to the inquiry is input by the worker 131, 141, etc. in response to displaying the divided document data, the worker terminal 130, 140, etc. accepts this. In addition, the worker terminal 130, 140, etc. transmits the inputted prompt to the data management device 4910.

[0444] In step S5003, the data management device 4910 transmits a search request to the server device 150, the server device 4930, the Web server device 3720, and the like.

[0445] In step S5004, the server device 150 transmits the search results including the reference answer contents to the data management device 4910.

[0446] In step S5005, the server device 4930 transmits the enterprise search results to the data management device 4910.

[0447] In step S5006, the Web server device 3720 transmits the Web search results to the data management device 4910.

[0448] In step S5007, the data management device 4910 transmits the search results including the reference answer contents, the enterprise search results, and the web search results to the worker terminals 130, 140, etc. as auxiliary information.

[0449] In step S5010 of FIG. 50B, when the worker terminal 130, 140, etc. receives input of a prompt including the inquiry content and auxiliary information, it transmits the same to the data management device 4910.

[0450] In step S5011, the data management device 4910 instructs the external server device 3420 to generate a response to the inquiry while referring to the auxiliary information, based on a prompt received from the worker terminal 130, 140, etc. Note that the data management device 4910 may instruct the external server device 3420 to generate a response to the inquiry while referring to the auxiliary information, based on a prompt including the inquiry and the auxiliary information selected by the agent unit 4911, without going through steps S5007 and S5010. Furthermore, the prompt may include the inquiry, the auxiliary information selected by the worker terminal 130, etc., and the auxiliary information selected by the agent unit 4911.

[0451] In step S5012, the external server device 3420 generates a response in accordance with an instruction from the data management device 4910 and transmits the response to the data management device 4910.

[0452] In step S5013, the data management device 4910 receives the reply sent from the external server device 3420 and sends it to the worker terminals 130, 140, etc.

[0453] In step S5014, the worker terminals 130, 140, etc. display the response content transmitted from the data management device 4910. The workers 131, 141, etc. refer to the response content displayed on the worker terminals 130, 140, etc. to generate input divided document data.

[0454] <Summary> As is clear from the above description, the business support system 4900 according to the ninth embodiment: When multiple pieces of divided document data are generated for the document data of an authority inquiry to be processed and a search instruction based on the inquiry content contained in the multiple pieces of divided document data is received, auxiliary information is collected that is referenced by the language model when generating the response content to the inquiry content. As auxiliary information, answer contents contained in the input divided document data corresponding to the inquiry contents, enterprise search results searched based on the inquiry contents, and web search results searched based on the inquiry contents are collected. The collected auxiliary information is sent to the language model along with the query to obtain the answer to the query.

[0455] As a result, the business support system according to the ninth embodiment makes it possible to acquire appropriate answer content from the language model regardless of whether the language model has been fine-tuned. In this way, the business support system according to the ninth embodiment can support the work of a worker who generates input divided document data.

[0456] [Other embodiments] The present invention is not limited to the configurations described in the above embodiments, but may be combined with other elements, etc. These aspects can be changed without departing from the spirit of the present invention, and can be appropriately determined depending on the application form.

[0457] This application claims priority based on Japanese Patent Application No. 2022-209989, filed on December 27, 2022, the entire contents of which are incorporated herein by reference. [Explanation of symbols]

[0458] 100: Business support system 110: Authority terminal 120: Data management device 121: Management Department 130, 140: Worker terminal 150: Server device 151: Database 200: Business support system 201: Search Department 400: Authorities inquiry document data 501: Document data acquisition unit 502: User interface section 503 :Divided part 504:Distribution section 505: Attribute information extraction section 506: Collection Department 507: Document data generation unit 508: Document data transmission unit 610, 620: Divided document data 710, 720: Input divided document data 800: Input document data 900: Attribute information 1000: Registration data 1101: Search screen provider 1102: Search condition acquisition section 1103: Search control section 1210, 1220: Search screen 1500: Business support system 1510: Learning device 1511: Learning Department 1600: Business support system 1610: Prediction device 1611: Forecasting Department 1801: Storage area 2100: Business support system 2101: Search Department 2201: Search screen provider 2202 :Extraction part 2203: First Vectorization Unit 2204: Separation section 2205: Second Vectorization Unit 2206: Similarity calculation unit 2207: Output section 2310: Search screen 2401 :Morphological analysis section 2402: Vectorized model 2411 :Morphological analysis section 2412: Vectorized Model 2601:Extraction part 2602: Filter section 2710: Search screen 2800: Business support system 2810 :Analysis equipment 2811 :Analysis Department 3010: Feedback information acquisition unit 3021 :Additional term extraction part 3022: Deleted term extraction unit 3023: Dictionary Update Unit 3031: Hyperparameter update unit 3041: Learning data generation unit 3042: Learning Department 3043: Model Update Section 3201, 3202: Vectorized models 3211, 3212: Vectorized models 3400: Business support system 3410: Data management device 3411: LM learning data generation unit 3420: External server device 3421:LM Service Department 3500: LM training dataset 3700: Business support system 3710: Second LM learning data generation unit 3711: Learning Object Information Collection Department 3712: Self-supervised learning data generation unit 3721: First generation control unit 3722: Second generation control unit 3820: Second LM training dataset 4000: Business support system 4010: Data management device 4011 :LM usage department 4200: Business support system 4211: LM learning data generation unit 4220: Server device 4300: LM training dataset 4500: Business support system 4510: Data management equipment 4511: Material generation department 4512 :LM usage part 4520: Server equipment 4700: Business support system 4710 :Data management equipment 4711 :LM usage part 4900: Business support system 4910: Data management equipment 4911: Agent Department 4930: Server device

Claims

1. A business support system having a data management device that generates document data using a language model, a document data acquisition unit for acquiring document data of an inquiry from an authority; an agent unit that, when an inquiry content of an inquiry item is extracted from the document data, acquires auxiliary information to be used in generating an answer content to the inquiry content from at least one search source, and transmits a prompt including the inquiry content and the auxiliary information to the language model; the language model generates an answer to the query by referring to auxiliary information included in the prompt; Business support system.

2. The auxiliary information is a search result including a reference answer, an enterprise search result searched based on the inquiry, or a search result searched via an external network. The business support system according to claim 1 .

3. The search results searched via the external network include search results from a web search. The business support system according to claim 2 .

4. the search results including the reference answer content include answer content associated with past inquiry content similar to the inquiry content included in the search request from the agent unit; The business support system according to claim 2 .

5. a document data transmission unit that transmits document data including the generated response content to the authority; The business support system according to claim 1 .

6. The language model is configured as a multimodal model to which data other than character information is input in addition to character information. The business support system according to claim 1 .

7. the language model is fine-tuned using the inquiry content of the inquiry item included in the past document data of the inquiry to the authority and the content of past answers to the inquiry content of the inquiry item included in the past document data of the inquiry to the authority; The business support system according to claim 1 .

8. When generating a response to the inquiry, the language model refers to the inquiry of the inquiry item contained in the past document data of the inquiry to the authority and the past response to the inquiry of the inquiry item contained in the past document data of the inquiry to the authority. The business support system according to claim 1 .

9. a generating unit that generates document data to be transmitted to the authority based on the document data including the response to the inquiry; The business support system according to claim 5, further comprising:

10. the document data transmission unit converts the document data generated by the generation unit into a predetermined file format and transmits the converted document data to the authority; The business support system according to claim 9.

11. A business support system having a data management device that generates document data using a language model, acquiring document data of an authority inquiry; when a query content of an inquiry item is extracted from the document data, acquiring auxiliary information to be used in generating a response content to the query content from at least one search source, and sending a prompt including the query content and the auxiliary information to the language model; the language model generates an answer to the query by referring to auxiliary information included in the prompt; Business support methods.

12. A computer of a data management device that generates document data using a language model, acquiring document data of an authority inquiry; when a query content of an inquiry item is extracted from the document data, acquiring auxiliary information to be used in generating a response content to the query content from at least one search source, and transmitting a prompt including the query content and the auxiliary information to the language model; the language model generates an answer to the query by referring to auxiliary information included in the prompt; Data management program.

Images