Information providing system and information providing method

Abstract

To provide an information providing system and an information providing method capable of reducing a response time.SOLUTION: The terminal 1 includes a patient information acquirer 111 that acquires patient information, an extraction query generator 112 that generates extraction query information indicating an extraction query for extracting extraction information in a preset format including medical information of preset attributes from the patient information, an extraction information acquirer 113 that acquires the extraction information from a first 1LLM management server 2A by transmitting the patient information and the extraction query information to the first 1LLM management server 2A, and a guide information acquirer 115 that acquires guide information from a second 2LLM management server 2B by transmitting the extraction information and guide generation query information indicating a guide generation query for requesting guide information obtained by analyzing contents of the patient information to the second 2LLM management server 2B. 105.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present invention relates to an information providing system and an information providing method. [Background technology]

[0002] A medical support device has been proposed in which a pre-trained large-scale language model is trained using a combination of text indicating a patient's condition and text indicating advice to be used in treating the patient as training data, and then the text is input into the large-scale language model, causing the large-scale language model to generate advice to be used in treating the patient (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7454090 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the case of a device using one relatively large language model as described in Patent Document 1, there is a risk that it may take a very long time to train the large language model, or that it may take a relatively long time to receive advice after inputting text indicating the patient's condition.

[0005] The present invention has been made in view of the above circumstances, and has an object to provide an information providing system and an information providing method that can reduce response time. [Means for solving the problem]

[0006] In order to achieve the above object, the information providing system according to the present invention comprises: a first language model management server that manages a first language model that, when event information indicating the contents of events that have occurred in chronological order and extraction query information indicating an extraction query for extracting extracted information in a predetermined format including parameter information indicating parameters of predetermined attributes from the event information, extracts and outputs the extracted information in the predetermined format from the event information; a second language model management server that, upon receiving the extraction information and analysis result generation query information indicating an analysis result generation query requesting the generation of an analysis result obtained by analyzing the content of the event, manages a second language model that generates and outputs the analysis result from the extraction information; a terminal device capable of communicating with the first language model management server and the second language model management server, The terminal device an input unit; an event information acquisition unit that acquires the event information; an extraction query generation unit that generates the extraction query information when a user performs an information provision request operation on the input unit to request provision of information related to the event; an extraction information acquisition unit that acquires the extraction information from the first language model management server by transmitting the event information and the extraction query information to the first language model management server; and an analysis result acquisition unit that acquires analysis result information indicating the analysis result from the second language model management server by transmitting the extraction information and the analysis result generation query information to the second language model management server. [Effects of the Invention]

[0007] According to the present invention, when a user performs the above-mentioned information request operation on the input unit of the terminal device, the extraction query generation unit of the terminal device generates extraction query information in response. The extraction information acquisition unit then transmits the above-mentioned event information and extraction query information to the first language model management server, thereby acquiring extraction information from the first language model management server. The analysis result acquisition unit then transmits the acquired extraction information and the above-mentioned analysis result generation query information to the second language model management server, thereby acquiring the above-mentioned analysis result information from the second language model management server. This allows the second language model management server to omit information that is not directly necessary for outputting the analysis result from the information input to the second language model, thereby shortening the time required for the second language model to output the analysis result. Therefore, the response time after requesting the provision of event-related information can be reduced. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a schematic configuration diagram of an information providing system according to an embodiment of the present invention; [Figure 2] 1 is a block diagram showing a hardware configuration of an information providing system according to an embodiment; [Figure 3] FIG. 2 is a block diagram showing the functional configuration of a first language model management server, a second language model management server, and a third language model management server according to the embodiment. [Figure 4] FIG. 2 is a diagram illustrating the operation of the first language model management server according to the embodiment. [Figure 5] FIG. 2 is a block diagram showing a functional configuration of a terminal device according to an embodiment. [Figure 6] FIG. 3 is a sequence diagram illustrating the operation of the information extraction system according to the embodiment. [Figure 7] FIG. 3 is a sequence diagram illustrating the operation of the information extraction system according to the embodiment. [Figure 8] FIG. 3 is a sequence diagram illustrating the operation of the information extraction system according to the embodiment. [Figure 9]FIG. 3 is a sequence diagram illustrating the operation of the information extraction system according to the embodiment. [Figure 10] 10 is a flowchart illustrating an example of the flow of a guide display process executed by the terminal device according to the embodiment. [Figure 11] 10 is a flowchart showing an example of the flow of a guide information generation process executed by a second LLM management server according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] An information providing system according to an embodiment of the present invention will be described in detail below with reference to the drawings. The information providing system according to this embodiment includes a first language model management server, a second language model management server, and a terminal device capable of communicating with the first and second language model management servers. The first language model management server receives event information indicating the contents of events occurring in chronological order and extraction query information indicating an extraction query for extracting extracted information in a predetermined format from the event information, including parameter information indicating parameters of predetermined attributes. The first language model management server manages a first language model that extracts and outputs the extracted information from the event information. The second language model management server receives the extraction information and analysis result generation query information indicating an analysis result generation query requesting the generation of analysis results obtained by analyzing the contents of the event. The second language model management server manages a second language model that generates and outputs analysis results from the extracted information. The terminal device has an input unit, an event information acquisition unit that acquires event information, an extraction query generation unit that generates the above-mentioned extraction query information when a user performs an information provision request operation on the input unit to request the provision of information related to an event, an extraction information acquisition unit that acquires extraction information from the first language model management server by transmitting the event information and the extraction query information to the first language model management server, and an analysis result acquisition unit that acquires analysis result information indicating the above-mentioned analysis result from the second language model management server by transmitting the extraction information and the above-mentioned analysis result generation query information to the second language model management server.

[0010] The information providing system according to this embodiment is employed, for example, in medical settings, to provide patients with lifestyle guidance or advice regarding medication. For example, it is an interactive system that accepts questions from medical professionals and responds to them regarding the patient's condition or behavior. As shown in FIG. 1 , this information providing system includes a terminal device 1 held by a medical professional such as a doctor or pharmacist, a first large-scale language model (LLM) management server (hereinafter referred to as the "LLM management server") 2A that can communicate with the terminal device 1 via a local area network NW2, and a second LLM management server 2B and a third LLM management server 2C that can communicate with the terminal device 1 via a wide area network NW1. Furthermore, the terminal device 1 can communicate, via the wide area network NW1, with a terminal device 3 held by a patient, a related information management server 4 that manages related information indicating the patient's blood test results, electrocardiogram, and ultrasound diagnosis results, and a content management server 5 that manages multiple types of content information indicating multiple types of medical content.

[0011] The related information management server 4 manages related information in association with patient identification information, and upon receiving related information request information (described later) transmitted from the terminal device 1, selects related information corresponding to the patient identification information included in the related information request information. The related information management server 4 then generates related information notification information including the selected related information and the patient identification information and transmits it to the terminal device 1.

[0012] When the content management server 5 receives related content request information (described below) transmitted from the second LLM management server 2B, it selects at least one piece of content information from among the multiple types of content information it manages, based on location identification information indicating the location of the content information (described below) contained in the received related content request information. Here, the content information may be, for example, content containing specialized medical information related to the heart and heart diseases. The content management server 5 then transmits the selected at least one piece of content information to the second LLM management server 2B.

[0013] The first LLM management server 2A receives extraction query information including patient information and instruction information for extracting extracted information in a predetermined format from the patient information, including parameter information indicating parameters of predetermined attributes. The first LLM management server 2A manages a first LLM that extracts and outputs the extracted information from the patient information. The patient information includes an electronic health record (EHR) including a patient diary, physiological data, etc. The EHR includes information representing time series of conditions such as exercise, diet, preferences, and sleep. The physiological data also includes time series information such as the patient's blood pressure, weight, electrocardiogram, pulse rate, respiratory rate, and body temperature measured by a terminal device 3. As shown in FIG. 2, the first LLM management server 2A includes a central processing unit (CPU) 201A, a main memory unit 202A, an auxiliary memory unit 203A, a communication unit 206A, and a bus 209A interconnecting these units. The CPU 201A is, for example, a multi-core processor. The main memory unit 202A has a volatile memory and is used as a work area for the CPU 201A. The auxiliary memory unit 203A is composed of a large-capacity nonvolatile memory and stores programs for realizing various functions of the LLM management server 2A. The communication unit 106 is connected to a local network NW2 constructed in, for example, a facility where a medical professional who owns the terminal device 1 is located.

[0014] The CPU 201A loads the programs stored in the auxiliary storage unit 203A into the main storage unit 202A and executes them, thereby functioning as an extraction query acquisition unit 211, an extraction information generation unit 212, and an extraction information notification unit 213, as shown in FIG. 3. The auxiliary storage unit 203A shown in FIG. 2 also includes a first LLM storage unit 231, as shown in FIG. 3, that stores first LLM information indicating a first LLM. When the patient information and extraction query information described above are input, the first LLM extracts and outputs the extraction information described above from the patient information. This first LLM is an LLM adjusted based on a combination of multiple types of prompts and responses created in advance by a medical professional, for example, and can be configured using, for example, Mistral7b or Zephyr7b.

[0015] Upon receiving extraction query information from the terminal device 1, the extraction query acquisition unit 211 notifies the extraction information generation unit 212 of the patient information and instruction information included in the received extraction query information. The extraction information generation unit 212 then executes an extraction process using the first LLM indicated by the first LLM information stored in the first LLM storage unit 231 to extract the extraction information from the instruction information and patient information received from the extraction query acquisition unit 211. Here, as shown in FIG. 4, for example, the first LLM is configured to output the patient's daily alcohol intake, calorie consumption, sodium intake, and potassium intake as time-series information in JSON format upon receiving the instruction information and patient information. Returning to FIG. 3, the extraction information generation unit 212 then notifies the extraction information notification unit 213 of the extracted extraction information. When the extraction information notification unit 213 is notified of the extraction information from the extraction information generation unit 212, the extraction information notification unit 213 generates extraction information notification information including the notified extraction information and patient identification information that identifies the corresponding patient, and transmits the extraction information notification information to the terminal device 1.

[0016] The second LLM management server 2B manages a second LLM that generates and outputs guide information when it receives the extracted information and analysis result generation query information indicating an analysis result generation query that requests the generation of guide information indicating guidelines, such as lifestyle advice and medication advice for patients, obtained by analyzing the extracted information. This second LLM management server 2B optimizes the guide information using so-called Retrieval Augmented Generation (RAG) technology. The second LLM management server 2B has a hardware configuration similar to that of the first LLM management server 2A, and includes, as shown in FIG. 2, a CPU 201B, a main memory unit 202B, an auxiliary memory unit 203B, a communication unit 206B, and a bus 209B interconnecting these units. Here, the communication unit 206B is connected to a wide area network NW1.

[0017] The CPU 201B loads and executes programs stored in the auxiliary storage unit 203B into the main storage unit 202B, thereby functioning as an analysis result generation query acquisition unit 214, a search query generation unit 215, a related content search unit 216, a related content acquisition unit 217, a guide generation unit 218, a check result acquisition unit 219, and a guide notification unit 220, as shown in FIG. 3. The auxiliary storage unit 203B shown in FIG. 2 also includes a search model storage unit 232 that stores search model information indicating a search model, and a second LLM storage unit 233 that stores second LLM information indicating a second LLM, as shown in FIG. 3. Here, when instruction information instructing a search for content information related to patient information from content information managed by the content management server 5, along with the aforementioned extraction information and related information, is input, the search model searches for and identifies at least one piece of content information from among multiple types of content information managed by the content management server 5. Here, the search query information includes the aforementioned extraction information. The search model also acquires, for example, location identification information indicating the location of content information managed by the content management server 5. This location identification information may be, for example, URL (Uniform Resource Locator) information.

[0018] The second LLM generates and outputs guide information when the extracted information and other related information related to the patient included in the analysis result generation query information and the searched content information are input. Here, the second LLM is configured using a multimodal LLM such as GPT4-V or LLaVA. Here, the related information may be image information showing the results of the blood test, an electrocardiogram, a chest X-ray, or an ultrasound diagnosis result.

[0019] When the analysis result generation query acquisition unit 214 acquires analysis result generation query information transmitted from the terminal device 1, it extracts the instruction information, extraction information, and related information included in the acquired analysis result generation query information. Then, the analysis result generation query acquisition unit 214 notifies the search query generation unit 215 of the extracted extraction information and related information, and notifies the guide generation unit 218 of the instruction information, extraction information, and related information.

[0020] When the search query generation unit 215 is notified of the extracted information and related information by the analysis result generation query acquisition unit 214, the search query generation unit 215 generates search query information including instruction information for instructing a search for content information related to the patient information, and the notified extracted information and related information. Then, the search query generation unit 215 notifies the generated search query information to the related content search unit 216. Furthermore, when the search query generation unit 215 is notified of check result information indicating that the check result (described below) is NG from the check result acquisition unit 219, the search query generation unit 215 generates new search query information and notifies the related content search unit 216.

[0021] When the related content search unit 216 receives the search query information from the search query generation unit 215, it identifies at least one piece of content information related to the patient information based on the received search query information and using the search model indicated by the search model information stored in the search model storage unit 232. The related content search unit 216 then notifies the related content acquisition unit 217 of location identification information that identifies the location of the identified at least one piece of content information. When the related content acquisition unit 217 receives the location identification information from the related content search unit 216, it generates related content request information including the notified location identification information. The related content acquisition unit 217 then transmits the generated related content request information to the content management server 5, thereby acquiring the identified at least one piece of content information transmitted from the content management server 5. The related content acquisition unit 217 also notifies the guide generation unit 218 of the acquired at least one piece of content information.

[0022] The guide generation unit 218 uses the second LLM indicated by the second LLM model information stored in the second LLM storage unit 233 to generate guide information from the instruction information, extraction information, and related information notified by the analysis result generation query acquisition unit 214, and at least one piece of content information notified by the related content acquisition unit 217. Then, the guide generation unit 218 notifies the check result acquisition unit 219 and the guide notification unit 220 of the generated guide information. Furthermore, when the guide generation unit 218 is notified by the check result acquisition unit 219 of check result information indicating that the check result, which will be described later, is NG, the guide generation unit 218 again uses the second LLM to generate guide information from the instruction information, extraction information, and related information, and at least one piece of content information newly notified by the related content acquisition unit 217.

[0023] When the check result acquisition unit 219 is notified of the guide information, it generates check query information including instruction information requesting the third LLM management server 2C to check the validity of the guide information and the notified guide information.The check result acquisition unit 219 then transmits the generated check query information to the third LLM management server 2C to acquire check result information transmitted from the third LLM management server 2C.Then, if the check result indicated by the check result information is OK, that is, indicates that the content of the guide information is valid, the check result acquisition unit 219 notifies the guide notification unit 220 of the check result information.On the other hand, if the check result indicated by the check result information is NG, that is, indicates that the content of the guide information is inappropriate, the check result acquisition unit 219 notifies the search query generation unit 215 and the guide generation unit 218 of the check result information.

[0024] When the guide notification unit 220 is notified by the check result acquisition unit 219 of check result information indicating that the check result is OK, the guide notification unit 220 generates guide notification information including the guide information notified by the guide generation unit 218 and transmits it to the terminal device 1.

[0025] The third LLM management server 2C manages a third LLM that outputs a check result when check query information including the guide information and instruction information requesting a check of the validity of the guide information is input. The third LLM management server 2C has a hardware configuration similar to that of the first LLM management server 2A, and as shown in FIG. 2, includes a CPU 201C, a main memory unit 202C, an auxiliary memory unit 203C, a communication unit 206C, and a bus 209C interconnecting these. Here, the communication unit 206C is connected to a wide area network NW1.

[0026] The CPU 201C reads out and executes a program stored in the auxiliary storage unit 203C into the main storage unit 202C, thereby functioning as a check query acquisition unit 221, a check unit 222, a related content acquisition unit 223, and a check result notification unit 224, as shown in FIG. 3. The auxiliary storage unit 203C shown in FIG. 2 also includes a third LLM storage unit 234 that stores third LLM information indicating a third LLM that checks the validity of the guide information and outputs a check result when guide information, instruction information requesting checking the validity of the guide information, and at least one piece of content information are input, as shown in FIG. 3. This third LLM can be configured using, for example, Mistral7b or Zephyr7b. The third LLM identifies a description to be checked in the guideline text indicated by the guide information, and outputs a check result based on the degree of consistency between the identified description and the content information.

[0027] When the check query acquisition unit 221 acquires check query information transmitted from the second LLM management server 2B, it extracts guide information and instruction information included in the acquired check query information and notifies the checking unit 222 of the extracted guide information and instruction information. When the related content acquisition unit 223 acquires at least one piece of content information transmitted from the second LLM management server 2B, it notifies the checking unit 222 of the acquired at least one piece of content information.

[0028] The check unit 222 generates check result information using the third LLM indicated by the third LLM information stored in the third LLM storage unit 234, based on the guide information and instruction information notified from the check query acquisition unit 221 and at least one piece of content information notified from the related content acquisition unit 223. Then, the check result notification unit 224 transmits the generated check result information to the second LLM management server 2B.

[0029] The terminal device 3 is, for example, a smartphone or wearable terminal device carried by the patient, and when the patient performs a patient information transmission operation on the input unit to transmit his / her own information to the terminal device 1, the terminal device 3 generates patient information notification information including the above-mentioned patient information and patient identification information that identifies the patient. Then, the terminal device 3 transmits the generated patient information notification information to the terminal device 1.

[0030] The terminal device 1 is installed in, for example, a medical institution, a welfare facility, or the like, and is used by medical professionals. The terminal device 1 is, for example, a personal computer, and as shown in FIG. 2, includes a CPU 101, a main memory unit 102, an auxiliary memory unit 103, a display unit 104, an input unit 105, a communication unit 106, and a bus 109 connecting the various units. The main memory unit 102 is composed of a volatile memory such as a random access memory (RAM) and is used as a work area for the CPU 101. The auxiliary memory unit 103 is composed of a non-volatile memory such as a read-only memory (ROM), a semiconductor memory, a solid-state drive (SSD), or a hard disk drive (HDD), and stores programs for realizing various functions of the terminal device 1. The display unit 104 is a display device such as a liquid crystal display or an organic electroluminescence (EL) display. The input unit 105 is an input device such as a keyboard or a touchpad. The communication unit 106 communicates with the first LLM management server 2A and the second LLM management server 2B via the local network NW2 and the wide area network NW1.

[0031] In the terminal device 1, the CPU 101 loads the program stored in the auxiliary storage unit 103 into the main storage unit 102 and executes it, thereby functioning as a patient information acquisition unit 111, an extraction query generation unit 112, an extraction information acquisition unit 113, a related information acquisition unit 114, a guide information acquisition unit 115, and a display control unit 116, as shown in Fig. 5. Furthermore, the auxiliary storage unit 103 shown in Fig. 2 has a patient information storage unit 131, an extraction query storage unit 132, an extraction information storage unit 133, a related information storage unit 134, an analysis result generation query storage unit 135, and a guide information storage unit 136, as shown in Fig. 5. The patient information storage unit 131 stores the above-mentioned patient information in association with patient identification information.

[0032] The extraction query storage unit 132 stores extraction query information including the patient information and instruction information instructing to extract the extraction information. The extraction information storage unit 133 stores the extraction information extracted from the patient information in association with patient identification information. The analysis result generation query storage unit 135 stores instruction information that requests to generate guide information, which is included in the analysis result generation query information. The guide information storage unit 136 stores the guide information generated by the second LLM management server 2B in association with patient identification information.

[0033] When the patient information acquisition unit 111 acquires the patient information notification information transmitted from the terminal device 3, it extracts the patient information and patient identification information contained in the acquired patient information notification information, and stores the extracted information in the patient information storage unit 131 in association with each other.

[0034] When the user performs a guide request operation on the input unit 105 to request the above-mentioned guide information, the extraction query generation unit 112 generates the above-mentioned extraction query information in response and stores it in the extraction query storage unit 132. The extraction information acquisition unit 113 acquires extraction information notification information transmitted from the first LLM management server 2A by transmitting the extraction query information stored in the extraction query storage unit 132 to the first LLM management server 2A. The extraction information acquisition unit 113 then extracts the extraction information and patient identification information contained in the acquired extraction information notification information, and stores the extracted information in the extraction information storage unit 133 in association with each other.

[0035] When the user performs a guide request operation on the input unit 105 to request the above-mentioned guide information, the related information acquisition unit 114 generates the above-mentioned related information request information in response to this and transmits it to the related information management server 4, thereby acquiring the related information notification information transmitted from the related information management server 4. Then, the related information acquisition unit 114 extracts the related information and patient identification information contained in the acquired related information notification information, and stores the extracted information in the related information storage unit 134 in association with each other.

[0036] When the user performs a guide request operation on the input unit 105 to request the above-mentioned guide information, the guide information acquisition unit 115 generates the above-mentioned analysis result generation query information including the instruction information stored in the analysis result generation query storage unit 135, the extraction information stored in the extraction information storage unit 133, and the related information stored in the related information storage unit 134. Furthermore, the guide information acquisition unit 115 transmits the generated analysis result generation query information to the second LLM management server 2B, thereby acquiring guide notification information transmitted from the second LLM management server 2B. Then, the guide information acquisition unit 115 extracts the guide information included in the acquired guide notification information and stores it in the guide information storage unit 136.

[0037] The display control unit 116 forms a guide notification image showing guidelines for giving advice to the patient based on the guide information stored in the guide information storage unit 136, and causes the display unit 104 to display the image.

[0038] Next, the operation of the information providing system according to this embodiment will be described with reference to Figures 6 to 9. First, it is assumed that a patient performs the above-mentioned patient information transmission operation on the input unit of terminal device 3. In this case, terminal device 3 generates patient information notification information including patient information and patient identification information (step S1), and the generated patient information notification information is transmitted from terminal device 3 to terminal device 1 (step S2). Meanwhile, upon acquiring the patient information notification information, terminal device 1 extracts the patient information and patient identification information included in the acquired patient information notification information, and stores the extracted information in the patient information storage unit 131 in association with each other (step S3).

[0039] Next, suppose that the medical professional performs the aforementioned guidance request operation on the input unit 105 of the terminal device 1. In this case, the terminal device 1 generates the aforementioned extraction query information (step S4), and the generated extraction query information is transmitted from the terminal device 1 to the first LLM management server 2A (step S5). Meanwhile, upon acquiring the extraction query information, the first LLM management server 2A extracts the patient information and instruction information contained in the acquired extraction query information. Then, the first LLM management server 2A executes an extraction process to extract the aforementioned extraction information from the extracted instruction information and patient information using the first LLM indicated by the first LLM information stored in the first LLM storage unit 231 (step S6). Thereafter, the first LLM management server 2A generates extraction information notification information including the extracted extraction information and patient identification information that identifies the corresponding patient (step S7), and the generated extraction information notification information is transmitted from the first LLM management server 2A to the terminal device 1 (step S8). On the other hand, when the terminal device 1 acquires the extraction information notification information, it extracts the extraction information and patient identification information contained in the acquired extraction information notification information, associates these pieces of information with each other, and stores them in the extraction information storage unit 133 (step S9).

[0040] Thereafter, the terminal device 1 generates the aforementioned related information request information (step S10), and transmits the generated related information request information from the terminal device 1 to the related information management server 4 (step S11). Meanwhile, upon acquiring the related information request information, the related information management server 4 selects, from the multiple types of content information it manages, related information corresponding to the patient identification information included in the acquired related information request information (step S12). Next, the related information management server 4 generates related information notification information including the selected related information and the patient identification information (step S13), and the generated related information notification information is transmitted from the related information management server 4 to the terminal device 1 (step S14). Meanwhile, upon acquiring the related information notification information, the terminal device 1 extracts the related information and the patient identification information included in the acquired related information notification information, and stores these pieces of information in the related information storage unit 134 in association with each other (step S15).

[0041] 7, the terminal device 1 generates the above-described analysis result generation query information, which includes the instruction information stored in the analysis result generation query storage unit 135, the extraction information stored in the extraction information storage unit 133, and the related information stored in the related information storage unit 134 (step S16). The generated analysis result generation query information is then transmitted from the terminal device 1 to the second LLM management server 2B (step S17). Meanwhile, upon receiving the analysis result generation query information, the second LLM management server 2B extracts the instruction information, extraction information, and related information contained in the received analysis result generation query information. The second LLM management server 2B then generates search query information, which includes the extracted extraction information and related information, and instruction information for instructing a search for content information related to the patient information (step S18). Next, the second LLM management server 2B searches for at least one piece of content information related to the patient information, based on the generated search query information, using a search model indicated by the search model information stored in the search model storage unit 232 (step S19). Next, the second LLM management server 2B generates related content request information including location identification information of at least one of the content information identified by the search (step S20), and the generated related content request information is sent from the second LLM management server 2B to the content management server 5 (step S21).

[0042] On the other hand, when the content management server 5 acquires the related content request information, it selects at least one piece of content information from the multiple types of content information it manages, based on the location identification information included in the acquired related content request information (step S22). Then, the selected at least one piece of content information is transmitted from the content management server 5 to the second LLM management server 2B (step S23). At this time, the selected at least one piece of content information is also transmitted from the content management server 5 to the third LLM management server 2C (step S24). On the other hand, when the second LLM management server 2B acquires the content information, it generates guide information from the extracted instruction information, extraction information, and related information, and the acquired at least one piece of content information, using the second LLM indicated by the second LLM model information stored in the second LLM storage unit 233 (step S25).

[0043] Next, the second LLM management server 2B generates the above-mentioned check query information (step S26), and the generated check query information is transmitted from the second LLM management server 2B to the third LLM management server 2C (step S27). Meanwhile, upon receiving the check query information, the third LLM management server 2C extracts the guide information and instruction information contained in the received check query information. Then, based on the extracted guide information and instruction information and at least one piece of content information received from the content management server 5, the third LLM management server 2C executes a check process using the third LLM indicated by the third LLM information stored in the third LLM storage unit 234 (step S28). Next, the third LLM management server 2C generates check result information indicating the check result obtained by the check process (step S29), and the generated check result information is transmitted from the third LLM management server 2C to the second LLM management server 2B (step S30).

[0044] On the other hand, when the second LLM management server 2B determines that the check result indicated by the acquired check result information is OK, that is, that the content of the guide information is valid (step S31), it generates guide notification information including the generated guide information, as shown in FIG. 8 (step S32). Then, the generated guide notification information is transmitted from the second LLM management server 2B to the terminal device 1 (step S33). On the other hand, when the terminal device 1 acquires the guide notification information, it extracts the guide information included in the acquired guide notification information and stores it in the guide information storage unit 136 (step S34). Thereafter, the terminal device 1 forms the above-mentioned guide notification image based on the guide information stored in the guide information storage unit 136 and displays it on the display unit 104 (step S35).

[0045] Also, assume that the third LLM management server 2C executes the above-described check process (step S28) and generates check result information indicating that the check result obtained by the check process is NG (step S36). In this case, the generated check result information is transmitted from the third LLM management server 2C to the second LLM management server 2B (step S37), and the second LLM management server 2B determines that the check result indicated by the acquired check result information is NG, i.e., indicates that the content of the guide information is inappropriate (step S38). In this case, the series of processes from steps S18 to S25 are executed again. Thereafter, as shown in FIG. 9, the second LLM management server 2B generates check query information again (step S26), and the generated check query information is transmitted from the second LLM management server 2B to the third LLM management server 2C (step S27). On the other hand, when the third LLM management server 2C acquires the check query information, it executes the above-mentioned check process again based on the guide information and instruction information included in the acquired check query information and at least one piece of content information acquired from the content management server 5 (step S28). Here, it is assumed that the third LLM management server 2C generates check result information indicating that the check result obtained by the check process is OK (step S29). In this case, after the series of processes from the above-mentioned steps S30 to S34 are executed, the terminal device 1 forms the above-mentioned guide notification image and displays it on the display unit 104 (step S35).

[0046] Next, the guide display process executed by the terminal device 1 according to this embodiment will be described with reference to FIG. 10. This guide display process is started when a program for executing the guide display process is started in the terminal device 1 after the terminal device 1 is powered on. First, the patient information acquisition unit 111 determines whether or not it has acquired patient information notification information transmitted from the terminal device 3 (step S101). If the patient information acquisition unit 111 determines that it has not acquired patient information notification information (step S101: No), it executes the process of step S103, which will be described later. On the other hand, if the patient information acquisition unit 111 determines that it has acquired patient information notification information (step S101: Yes), it extracts the patient information and patient identification information included in the acquired patient information notification information, associates the extracted information with each other, and stores the extracted information in the patient information storage unit 131 (step S102).

[0047] Next, the extraction query generation unit 112 determines whether the medical professional has performed the above-mentioned guidance request operation on the input unit 105 (step S103). Here, if the extraction query generation unit 112 determines that the guidance request operation has not been performed on the input unit 105 (step S103: No), the processing of step S101 is executed again. On the other hand, if the extraction query generation unit 112 determines that the guidance request operation has been performed on the input unit 105 (step S103: Yes), it generates the above-mentioned extraction query information and stores it in the extraction query storage unit 132. Then, the extraction information acquisition unit 113 transmits the extraction query information stored in the extraction query storage unit 132 to the first LLM management server 2A (step S104). As a result, the extraction information acquisition unit 113 acquires the extraction information notification information transmitted from the first LLM management server 2A. Then, the extracted information acquiring unit 113 extracts the extracted information and the patient identification information included in the acquired extracted information notification information, and stores these pieces of extracted information in association with each other in the extracted information storage unit 133 (step S105).

[0048] Next, the related information acquisition unit 114 generates the above-mentioned related information request information and transmits it to the related information management server 4 (step S106). As a result, the related information acquisition unit 114 acquires the related information notification information transmitted from the related information management server 4. Then, the related information acquisition unit 114 extracts the related information and the patient identification information included in the acquired related information notification information, and stores the extracted information in the related information storage unit 134 in association with each other (step S107).

[0049] Thereafter, the guide information acquisition unit 115 generates the above-mentioned analysis result generation query information including the instruction information stored in the analysis result generation query storage unit 135, the extraction information stored in the extraction information storage unit 133, and the related information stored in the related information storage unit 134. Then, the guide information acquisition unit 115 transmits the generated analysis result generation query information to the second LLM management server 2B (step S108). As a result, the guide information acquisition unit 115 acquires guide notification information transmitted from the second LLM management server 2B (step S109). Next, the guide information acquisition unit 115 extracts guide information included in the acquired guide notification information and stores it in the guide information storage unit 136 (step S110). Subsequently, the display control unit 116 forms the above-mentioned guide notification image based on the guide information stored in the guide information storage unit 136 and displays it on the display unit 104 (step S111). Thereafter, the processing of step S101 is executed again.

[0050] Next, the guide information generation process executed by the second LLM management server 2B according to this embodiment will be described with reference to FIG. 11. This guide information generation process is started when a program for executing the guide information generation process is started in the second LLM management server 2B. First, the analysis result generation query acquisition unit 214 determines whether or not analysis result generation query information transmitted from the terminal device 1 has been acquired (step S201). Here, the process of step S201 is repeatedly executed unless the analysis result generation query acquisition unit 214 acquires analysis result generation query information (step S201: No). On the other hand, when the analysis result generation query acquisition unit 214 determines that analysis result generation query information has been acquired (step S201: Yes), it extracts instruction information, extraction information, and related information included in the acquired analysis result generation query information. Then, the search query generation unit 215 generates search query information including instruction information for instructing a search for the content information and the extracted extraction information and related information (step S202). Next, the related content search unit 216 searches for and identifies at least one piece of content information related to the patient information based on the generated search query information, using the search model indicated by the search model information stored in the search model storage unit 232 (step S203). Subsequently, the related content acquisition unit 217 generates related content request information including location identification information of the identified at least one piece of content information, and transmits the generated related content request information to the content management server 5 (step S204). As a result, the related content acquisition unit 217 acquires the identified at least one piece of content information transmitted from the content management server 5 (step S205).

[0051] Thereafter, the guide generation unit 218 generates guide information from the instruction information, extraction information, and related information included in the analysis result generation query information and at least one piece of content information acquired by the related content acquisition unit 217, using the second LLM indicated by the second LLM model information stored in the second LLM storage unit 233 (step S206). Next, the check result acquisition unit 219 generates the above-mentioned check query information and transmits the generated check query information to the third LLM management server 2C (step S207). As a result, the check result acquisition unit 219 acquires the check result information transmitted from the third LLM management server 2C (step S208). Next, the check result acquisition unit 219 determines whether the check result indicated by the check result information is OK (step S209). Here, if the check result acquisition unit 219 determines that the check result indicated by the check result information is NG (step S209: No), the process of step S202 is executed again. On the other hand, it is assumed that the check result acquisition unit 219 determines that the check result indicated by the check result information is OK (step S209: Yes). In this case, when the guide notification unit 220 is notified of the check result information indicating that the check result is OK from the check result acquisition unit 219, the guide notification unit 220 generates guide notification information including the guide information notified from the guide generation unit 218 and transmits it to the terminal device 1 (step S210). Subsequently, the process of step S201 is executed again.

[0052] As described above, in the information provision system according to this embodiment, when a medical professional performs the aforementioned guide request operation on the input unit 105 of the terminal device 1, the extraction query generation unit 112 of the terminal device 1 generates extraction query information in response to this. The extraction information acquisition unit 113 then transmits the extraction query information, including the aforementioned patient information, to the first LLM management server 2A, thereby acquiring the extraction information from the first LLM management server 2A. The guide information acquisition unit 115 then transmits the aforementioned analysis result generation query information, including the acquired extraction information, to the second LLM management server 2B, thereby acquiring guide information from the second LLM management server 2B. This allows the second LLM management server 2B to omit information that is not directly necessary for outputting the analysis results from the information input to the second LLM, thereby shortening the time required for the second LLM to output the analysis results. This reduces the response time after a request for guide information is made.

[0053] Conventional approaches to Q&A systems in the medical field have often used a single, large-scale LLM tuned using large-scale training data containing medical-related queries. However, using such a single, large-scale LLM requires training an LLM for each type of task, which can increase the burden of creating training data and result in longer response times. In contrast, the information provision system of this embodiment uses a first LLM for information extraction purposes to extract the extracted information from patient information such as PHR and physiological data, a second LLM for guide generation purposes to generate the guide information by referencing the extracted information and content information, and a third LLM for checking the validity of the guide information. In this way, rather than using only a single LLM, using multiple LLMs depending on the application can achieve faster responses and improved accuracy of the guide information.

[0054] Although the embodiments of the present invention have been described above, the present invention is not limited to the configurations of the above-described embodiments. For example, the information provision system may provide information in fields other than the medical field. For example, if the owner of terminal device 1 is a company that provides products and services unrelated to the medical field and terminal device 3 is a customer of that company, information useful to the customer may be displayed on display unit 104 of terminal device 1.

[0055] In the embodiment, the terminal device 1 may have a function of associating the extraction information, guide information, and check result information with one another and displaying them on the display unit 104. In this case, it is possible to provide medical personnel with information about a process for improving the reliability of the guide information.

[0056] The functions of the first LLM management server 2A, the second LLM management server 2B, and the third LLM management server 2C in the embodiment may be realized in one physical server.

[0057] Furthermore, the various functions of the terminal device 1, first LLM management server 2A, second LLM management server 2B, and third LLM management server 2C according to the present invention can be realized using a normal computer system, rather than a dedicated system. For example, a program for executing the above operations may be stored on a non-transitory recording medium (such as a CD-ROM (Compact Disc Read Only Memory)) that can be read by the computer system and distributed to a computer connected to a network, and the program may be installed in the computer system to configure the terminal device 1, first LLM management server 2A, second LLM management server 2B, and third LLM management server 2C that execute the above-described processes.

[0058] The method of providing the program to the computer is arbitrary. For example, the program may be uploaded to a bulletin board system (BBS) on a communication line and distributed to the computer via the communication line. The computer then launches the program and executes it under the control of an operating system (OS) in the same way as other applications. In this way, the computer functions as a terminal device 1 that executes the above-mentioned processes.

[0059] Although the embodiments and modifications of the present invention have been described above, the present invention is not limited to these. The present invention includes any combination of the embodiments and modifications, and any combination to which appropriate modifications have been made. [Industrial Applicability]

[0060] The present invention is suitable as a system that appropriately provides information desired by a user. [Explanation of symbols]

[0061] 1, 3: Terminal device, 2A: First LLM management server, 2B: Second LLM management server, 2C: Third LLM management server, 4: Related information management server, 5: Content management server, 8: BBR, 101, 201A, 201B, 201C: CPU, 102, 202A, 202B, 202C: Main memory unit, 103, 203A, 203B, 203C: Auxiliary memory unit, 104: Display unit, 105: Input unit, 106, 206A, 206B, 206C: Communication unit, 109, 209A, 209B, 209C: Bus, 111: Patient information acquisition unit, 112: Extraction query generation unit, 113: Extraction information acquisition unit, 114: Related information acquisition unit, 115: Guide information acquisition unit, 118: Display control unit, 119: Extraction query generation information acquisition unit, 131: Patient information storage unit, 132: Extraction query Area memory unit, 133: Extraction information memory unit, 134: Related information memory unit, 135: Analysis result generation query memory unit, 136: Guide information memory unit, 211: Extraction query acquisition unit, 212: Extraction information generation unit, 213: Extraction information notification unit, 214: Analysis result generation query acquisition unit, 215: Search query generation unit, 216: Related content search unit, 217: Related content acquisition unit, 218: Guide generation unit, 219: Check result acquisition unit, 220: Guide notification unit, 221: Check query acquisition unit, 222: Check unit, 223: Related content acquisition unit, 224: Check result notification unit, 231: First LLM memory unit, 232: Search model memory unit, 233: Second LLM memory unit, 234: Third LLM memory unit, NW1: Wide area network, NW2: Local network

Claims

1. a first language model management server that manages a first language model that, when event information indicating the contents of events that have occurred in chronological order and extraction query information indicating an extraction query for extracting extracted information in a predetermined format from the event information, including parameter information indicating parameters of predetermined attributes, is input, extracts and outputs the extracted information in the predetermined format from the event information; a second language model management server that, upon receiving the extraction information and analysis result generation query information indicating an analysis result generation query requesting generation of an analysis result obtained by analyzing the content of the event, manages a second language model that generates and outputs the analysis result from the extraction information; a terminal device capable of communicating with the first language model management server and the second language model management server, The terminal device an input unit; an event information acquisition unit that acquires the event information; an extraction query generation unit that generates the extraction query information when a user performs an information provision request operation on the input unit to request provision of information related to the event; an extraction information acquisition unit that acquires the extraction information from the first language model management server by transmitting the event information and the extraction query information to the first language model management server; an analysis result acquisition unit that acquires analysis result information indicating the analysis result from the second language model management server by transmitting the extraction information and the analysis result generation query information to the second language model management server; Information provision system.

2. further comprising a content management server that manages a plurality of types of content information indicating a plurality of types of content that have been set in advance; The second language model management server further a search model indicating a search model for searching and outputting at least one piece of content information from among the plurality of types of content information when search query information indicating a search query requesting a search for the content information related to the content of the event is input; upon acquiring the extracted information and the analysis result generation query information, the search model for searching the content information is generated; the generated search query information and the search model are used to identify the at least one piece of content information; the identified at least one piece of content information is acquired from the content management server; and the at least one piece of content information is input to the second language model together with the extracted information and the analysis result generation query information, thereby outputting the analysis result; The information providing system according to claim 1 .

3. a third language model management server that manages a third language model that, when analysis result information indicating the analysis result and check query information indicating a check query requesting checking the validity of the analysis result are input, checks the validity of the analysis result information and outputs a check result; The second language model management server further transmits analysis result information indicating the analysis result output by the second language model and the check query information to the third language model management server, thereby obtaining check result information indicating the check result from the third language model management server, and if the obtained check result information indicates that the analysis result information is valid, transmits the analysis result information to the terminal device, and if the obtained check result information indicates that the check result information is invalid, causes the second language model to output the analysis result again.

3. The information providing system according to claim 1 or 2.

4. A step in which the terminal device acquires event information indicating the contents of events that have occurred in chronological order; generating extraction query information indicating an extraction query for extracting, from the event information, extraction information in a preset format including parameter information indicating parameters of preset attributes, when a user performs an information provision request operation on an input unit to request provision of information related to the event; a step of transmitting, by the terminal device, the event information and the extraction query information to a first language model management server that manages a first language model; the first language model management server inputs the event information and the extraction query information into the first language model, causing the first language model to extract and output the extraction information from the event information, and transmitting the output extraction information to the terminal device; a step of the terminal device acquiring the extracted information from the first language model management server; the terminal device generating analysis result generation query information indicating an analysis result generation query requesting generation of an analysis result obtained by analyzing the content of the event, and transmitting the extracted information and the generated analysis result generation query information to a second language model management server that manages a second language model; the second language model management server inputs the extraction information and the analysis result generation query information into the second language model, causing the second language model to generate and output the analysis result, and transmits analysis result information indicating the output analysis result to the terminal device; and a step of acquiring the analysis result information from the second language model management server by the terminal device. Information provision method.

Images