Information extraction system, information extraction device, information extraction method, and program

Abstract

To provide an information extraction system, an information extraction device, an information extraction method, and a program that can improve convenience for a user to manage necessary information.SOLUTION: A terminal device 1 comprises: an anonymization target document creation unit 114 that creates a plurality of pieces of anonymization target document information indicating an anonymization target document including part of a division unit text of each of a plurality of pieces of document information; an anonymization unit 115 that transmits anonymization request information to LLM management servers 2A, 2B to acquire a plurality of pieces of anonymized document information transmitted from the LL management servers 2A, 2B; an extraction target document creation unit 116 that creates a plurality of pieces of extraction target document information from the acquired plurality of anonymized document information; an extraction information acquisition unit 117 that transmits extraction request information to an LLM management server 2C to acquire extraction information transmitted from the LLM management server 2C; and an extraction information conversion unit 118 that converts type identification information included in the extraction information into corresponding terms.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to an information extraction system, an information extraction device, an information extraction method, and a program. [Background technology]

[0002] A medical data management system has been proposed that includes a management server and an analysis server, in which the management server anonymizes medical data including personal information to generate anonymous medical data that does not include the personal information, assigns an anonymous ID to the anonymous medical data that corresponds to a personal ID in the personal information, transmits the anonymous medical data with the anonymous ID assigned to it to an analysis server, and the analysis server analyzes the transmitted anonymous medical data to obtain analysis results, assigns the anonymous ID assigned to the transmitted anonymous medical data to the analysis results, and returns the analysis results with the anonymous ID assigned to it to the management server (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2023-37972 A Summary of the Invention [Problem to be solved by the invention]

[0004] Meanwhile, there is a demand for improving the convenience of users when managing the information they need by using an external analysis server such as that described in Patent Document 1 to extract information contained in electronic medical record information that users need from the electronic medical record information.

[0005] The present invention has been made in consideration of the above-mentioned reasons, and aims to provide an information extraction system, an information extraction device, an information extraction method, and a program that can improve the convenience for users when managing the information they need. [Means for solving the problem]

[0006] In order to achieve the above object, the information extraction system according to the present invention comprises: A terminal device; A large-scale language model management unit for managing a large-scale language model, The terminal device an extraction information acquisition unit that transmits, to the large-scale language model management unit, first prompt information requesting the execution of an extraction process for extracting type identification information designated by a user from an extraction target document that has been anonymized by converting at least one type of term set in advance into type identification information that identifies a type of term, and thereby acquires extraction information indicating at least one type of type identification information extracted by the extraction process transmitted from the large-scale language model management unit; and an extracted information conversion unit that converts the type identification information included in the extracted information into a corresponding term.

[0007] In addition, the information extraction system according to the present invention comprises: The terminal device, a text analysis unit that performs text analysis on each of the plurality of pieces of document information to identify each of the pieces of text (hereinafter referred to as "divided unit texts") obtained by dividing a document indicated by the document information into two or more pieces; an anonymization target document generation unit that generates a plurality of pieces of the anonymization target document information indicating the anonymization target document including a portion of a division unit text identified for each of the documents indicated by the plurality of pieces of document information; an anonymization unit that transmits second prompt information, including the plurality of pieces of document information to be anonymized and the anonymization command information, to the first large-scale language model management server, thereby acquiring the plurality of pieces of anonymized document information transmitted from the first large-scale language model management server; The information processing device may further include an extraction target document generating unit that generates a plurality of pieces of extraction target document information from the plurality of pieces of anonymized document information obtained. Effect of the Invention

[0008] According to the present invention, the first prompt information is transmitted to the large-scale language model management unit, and extraction information indicating at least one type identification information extracted by the extraction process is obtained from the large-scale language model management unit. Then, the extraction information conversion unit converts the type identification information included in the extraction information into a corresponding term. This makes it possible to extract terms of a type identified by the type identification information specified by the user from the document indicated by the document information, thereby improving the convenience for the user when managing necessary information included in the document information. [Brief description of the drawings]

[0009] [Figure 1] 1 is a schematic configuration diagram of an information extraction system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a block diagram showing a hardware configuration of a terminal device according to an embodiment. [Diagram 3] FIG. 2 is a block diagram showing a functional configuration of a terminal device according to an embodiment. [Figure 4] FIG. 2A is a diagram showing an example of a document before an anonymization process according to the embodiment, and FIG. 2B is a diagram illustrating the operation of an anonymization target document generation unit according to the embodiment. [Diagram 5] 1A is a diagram illustrating the operation of an anonymization unit according to an embodiment, and FIG. 1B is a diagram illustrating an example of a document after an anonymization process according to an embodiment. [Figure 6] FIG. 2A is a diagram showing an example of extracted information before conversion processing according to an embodiment, and FIG. 2B is a diagram showing an example of extracted information after conversion processing according to an embodiment. [Figure 7] FIG. 4 is a sequence diagram for explaining an operation of the information extraction system according to the embodiment. [Figure 8] 11 is a flowchart showing an example of the flow of an information extraction process executed by the terminal device according to the embodiment. [Figure 9] FIG. 13A is a diagram illustrating the operation of an anonymization target document generation unit according to the modified example, and FIG. 13B is a diagram illustrating the operation of an anonymization unit according to the modified example. [Figure 10]FIG. 13 is a block diagram showing a functional configuration of a terminal device according to a modified example. [Figure 11] FIG. 13A is a diagram showing an example of a document before an anonymization process according to the modified example, and FIG. 13B is a diagram showing an example of a document after an abstraction process and an anonymization process according to the modified example. [Figure 12] 13 is a flowchart showing an example of the flow of an information extraction process executed by a terminal device according to a modified example. [Figure 13] FIG. 1A is a diagram showing an example of an operation screen image displayed on a display unit of a terminal device relating to a modified example, and FIG. 1B is a diagram showing an example of an extracted information notification image displayed on a display unit of a terminal device relating to a modified example. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] An information extraction system according to an embodiment of the present invention will be described in detail below with reference to the drawings. The information extraction system according to the present embodiment includes a terminal device, a first large-scale language model management server that, upon receiving anonymization target document information transmitted from the terminal device and indicating an anonymization target document to be anonymized by converting at least one predetermined term into type identification information identifying the type of term, and anonymization command information indicating a sentence requesting an anonymization process for converting each of the at least one term included in the anonymization target document indicated by the anonymization target document information into the type identification information, performs the anonymization process using a large-scale language model set in advance and transmits anonymized document information indicating the anonymized document generated by the anonymization process to the terminal device, and a second large-scale language model management server that, upon receiving extraction target document information transmitted from the terminal device and indicating an extraction target document from which at least one type identification information of the at least one type identification information is to be extracted, and extraction command information indicating a sentence requesting an extraction process for extracting type identification information specified by a user included in the extraction target document indicated by the extraction target document information, performs the extraction process using a large-scale language model set in advance and transmits extraction information indicating at least one type identification information extracted by the extraction process to the terminal device.Here, the terminal device has a text analysis unit that identifies division unit texts included in a document indicated by the document information by performing text analysis on each of the multiple document information; an anonymization target document generation unit that generates multiple anonymization target document information indicating documents to be anonymized including a portion of the division unit texts identified for each of the multiple document information; an anonymization unit that acquires multiple anonymization document information transmitted from a first large-scale language model management server by transmitting the multiple anonymization target document information and anonymization command information to a first large-scale language model management server; an extraction target document generation unit that generates multiple extraction target document information from the acquired multiple anonymization document information; an extraction information acquisition unit that acquires extraction information transmitted from the second large-scale language model management server by transmitting the extraction target document information and extraction command information to a second large-scale language model management server; and an extraction information conversion unit that converts type identification information included in the extraction information into a corresponding term.

[0011] As shown in FIG. 1, the information extraction system according to the present embodiment includes a terminal device 1 and large language model (LLM) management servers (hereinafter referred to as "LLM management servers") 2A, 2B, and 2C that can communicate with the terminal device 1 via a network NW1. The LLM management servers 2A, 2B, and 2C each manage an LLM, and are realized as a virtual server in a cloud server, for example. Here, the LLM is composed of a trained neural network trained using a large corpus as training data. The LLM management servers 2A, 2B, and 2C are each managed by a business operator that provides a different IaaS (Infrastructure as a Service). A text processing mechanism that executes text processing based on information transmitted from the terminal device 1 is configured from the LLM management servers 2A, 2B, and 2C.

[0012] When the LLM management servers 2A and 2B receive anonymization request information requesting anonymization of a part of a document indicated by document information transmitted from the terminal device 1, the servers 2A and 2B perform an anonymization process to anonymize the part of the document using a preset LLM, and transmit anonymized document information indicating the anonymized document generated by the anonymization process to the terminal device 1. Here, the anonymization request information includes prompt information including anonymization target document information indicating a document to be anonymized by converting at least one term set in advance into type identification information identifying the type of the term, and anonymization command information indicating a sentence requesting anonymization process to be performed to convert at least one term included in the document to be anonymized indicated by the anonymization target document information into type identification information. When the LLM management servers 2A and 2B receive the anonymization request information, the servers 2A and 2B use a preset LLM to decrypt the contents of the anonymization process based on the sentence indicated by the anonymization command information constituting the prompt information included in the anonymization request information. Then, the LLM management servers 2A, 2B perform an anonymization process on the document to be anonymized indicated by the document information to be anonymized in accordance with the decrypted content, and transmit the anonymized document information indicating the anonymized document generated by the anonymization process to the terminal device 1. Here, when the document information indicates a medical document, the at least one type of term includes at least one of a term indicating a patient's attribute, a term indicating a medicine name, a term indicating a disease name, a term indicating a patient's symptom, a term indicating a patient's condition, and a term indicating the content of treatment for the patient.

[0013] When the LLM management server 2C receives extraction request information requesting the extraction of type identification information designated by a user from an extraction target document, which will be described later, transmitted from the terminal device 1, the LLM management server 2C performs an extraction process to extract type identification information using a preset LLM, and transmits extraction information indicating at least one type identification information extracted by the extraction process to the terminal device 1. Here, the extraction request information includes prompt information consisting of extraction target document information indicating an extraction target document from which at least one type identification information designated by a user among at least one type identification information is to be extracted, and extraction command information indicating a sentence requesting the execution of an extraction process to extract at least one type identification information designated by a user that is included in the extraction target document indicated by the extraction target document information. When the LLM management server 2C receives the above-mentioned extraction request information, the LLM management server 2C uses a preset LLM to decipher the contents of the extraction process based on the sentence indicated by the extraction command information constituting the above-mentioned prompt information included in the extraction request information. Then, the LLM management server 2C performs extraction processing on the extraction target document indicated by the extraction target document information according to the deciphered contents, and transmits extraction information indicating at least one type identification information extracted by the extraction process to the terminal device 1.

[0014] The terminal device 1 is installed in a medical institution, a welfare facility, etc., and is used by users such as doctors and other medical professionals. The terminal device 1 is, for example, a personal computer, and as shown in FIG. 2, includes a CPU (Central Processing Unit) 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 each unit, and functions as an information extraction device used when extracting desired information from document information such as electronic medical record information. The main memory unit 102 is composed of a volatile memory such as a RAM (Random Access Memory), and is used as a working area for the CPU 101. The auxiliary memory unit 103 is composed of a non-volatile memory such as a ROM (Read Only Memory), a semiconductor memory, an SSD (Solid State Drive), and an HDD (Hard Disk Drive), 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, an organic EL display, etc. The input unit 105 is an input device such as a keyboard, a touch pad, etc. The communication unit 106 includes a modem and a gateway, and communicates with the LLM management servers 2A, 2B, and 2C via the network NW1.

[0015] 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 document acquisition unit 111, a command generation unit 112, a text analysis unit 113, an anonymization target document generation unit 114, an anonymization unit 115, an extraction target document generation unit 116, an extraction information acquisition unit 117, an extraction information conversion unit 118, and a display control unit 119, as shown in Fig. 3. Also, the auxiliary storage unit 103 shown in Fig. 2 has a document information storage unit 131, an extraction command storage unit 132, an analysis result storage unit 133, an anonymization target document storage unit 134, an anonymized document storage unit 135, an extraction target document storage unit 136, an extraction information storage unit 137, and a command generation information storage unit 138, as shown in Fig. 3. The document information storage unit 131 stores a plurality of document information such as electronic medical record information. The document information storage unit 131 stores document information indicating a document including terms indicating disease name, age, sex, drug name, and treatment name (see the part surrounded by the dashed line), for example, as shown in FIG. 4(A).

[0016] Returning to FIG. 3, the command generation information storage unit 138 stores information indicating sentences corresponding to various command information, including sentences requesting the above-mentioned anonymization process and extraction process, in association with operation content identification information identifying the operation content performed by the user on the input unit 105. The command storage unit 132 stores anonymization command information indicating sentences requesting the above-mentioned anonymization process and extraction command information indicating sentences requesting the above-mentioned extraction process. As the anonymization command information, for example, a sentence such as "Create a sentence in which "Disease name" is replaced with "DISEASE*", "Age" is replaced with "AGE*", "Gender" is replaced with "GENDER*", "Drug name" is replaced with "MEDICINE*", and "Treatment name" is replaced with "TREATMENT*" can be adopted. In this case, "DISEASE*", "AGE*", "GENDER*", "MEDICINE*" and "TREATMENT*" correspond to the above-mentioned type identification information. As the extraction command information, "Extract a combination of "MEDICINE*" and "DIEASE*" when "DISEASE*" occurs after "MEDICINE*" is administered. " document may be adopted.

[0017] The analysis result storage unit 133 stores analysis result information indicating the result of executing text analysis on the document indicated by the document information. Here, the analysis result storage unit 133 stores information indicating the delimiter positions of the division unit text included in the document indicated by the document information. The anonymization target document storage unit 134 stores anonymization target document information indicating the anonymization target document that is the target of the above-mentioned anonymization process to be requested from the LLM management servers 2A and 2B.

[0018] The anonymized document storage unit 135 stores anonymized document information indicating anonymized documents obtained by anonymization processing performed by the LLM management servers 2A, 2B on anonymization target documents indicated by the anonymization target document information. The anonymized document storage unit 135 also stores type identification information identifying the type of anonymized term in association with the term. The extraction target document storage unit 136 stores extraction target document information indicating extraction target documents that are the subject of a request to the LLM management server 2C for the above-mentioned extraction processing. The extraction information storage unit 137 stores term information indicating terms identified by type identification information extracted by the LLM management server 2C.

[0019] The document acquisition unit 111 acquires a plurality of pieces of document information designated by the user via the input unit 105, and stores the acquired document information in the document information storage unit 131. When the user performs an information extraction operation on the input unit 105 to extract information desired by the user from the document information, the command generation unit 112 generates the above-mentioned anonymization command information and extraction command information using sentence information associated with the information extraction operation stored in the command generation information storage unit 138, and stores the generated information in the extraction command storage unit 132. The text analysis unit 113 identifies the division unit text included in the document indicated by the document information by performing text analysis on each of the plurality of pieces of document information stored in the document information storage unit 131. Specifically, the text analysis unit 113 divides the document into at least one division unit text based on the positions of punctuation marks and the like included in the document indicated by the document information stored in the document information storage unit 131. Then, the text analysis unit 113 stores information indicating the delimiter positions of at least one division unit text in the analysis result storage unit 133.

[0020] The anonymization target document generating unit 114 identifies division unit texts included in each of the documents indicated by the plurality of document information stored in the document information storing unit 131, based on the information stored in the analysis result storing unit 133. Then, the anonymization target document generating unit 114 generates a plurality of anonymization target document information indicating the anonymization target document including a part of the division unit texts identified for the documents indicated by each of the plurality of document information. For example, as shown in FIG. 4(B), the anonymization target document generating unit 114 generates two anonymization target documents by mixing the division unit texts of document 1 composed of division unit texts La1[0], La1[1], . . ., La1[5] and document 2 composed of division unit texts La2[0], La2[1], . . ., La2[5]. Here, division unit text identification information La1[0], La1[1], ..., La1[5] reflecting their order is given to each division unit text of document 1, and division unit text identification information La2[0], La2[1], ..., La2[5] reflecting their order is given to each division unit text of document 2. Then, the anonymization target document generation unit 114 stores each of the two anonymization target documents in the anonymization target document storage unit 134 in a manner that each division unit text constituting the anonymization target document corresponds to the division unit text identification information. Here, the two anonymization target document information undergoes the above-mentioned anonymization process in each of the separate LLM management servers 2A and 2B.

[0021] Returning to FIG. 3, the anonymization unit 115 generates a plurality of the above-mentioned anonymization request information for each of a plurality of anonymization target document information, and transmits the generated plurality of anonymization request information to one of the LLM management servers 2A and 2B, thereby acquiring a plurality of anonymized document information transmitted from the LLM management servers 2A and 2B. Here, the order of the division unit texts of the anonymized document indicated by the anonymized document information and the anonymized document indicated by the corresponding anonymization target document information match. For example, for two anonymization target document information shown in FIG. 4(B), the anonymization unit 115 transmits anonymization request information including one of the anonymization target document information and the above-mentioned anonymization command information to the LLM management server 2A, and transmits anonymization request information including the other of the anonymization target document information and prompt information composed of the above-mentioned anonymization command information to the LLM management server 2B. In this way, the anonymization unit 115 acquires anonymized document information from each of the LLM management servers 2A and 2B. In this case, as shown in Fig. 5(A), the order of the division unit texts in the anonymized document matches the order of the division unit texts in the corresponding anonymization target document shown in Fig. 4(B). Returning to Fig. 3, the anonymization unit 115 stores each of the acquired multiple anonymized document information in the anonymized document storage unit 135 in a manner that associates each of the division unit texts constituting the information with the division unit text identification information, based on the delimiter positions of each division unit text in the anonymization target document indicated by the corresponding anonymization target document information. The anonymization unit 115 also stores type identification information for identifying the type of anonymized term in the anonymized document storage unit 135 in association with the term, by comparing the anonymization target document information stored in the anonymization target document storage unit 134 with the anonymized document information.

[0022] The extraction target document generating unit 116 generates extraction target document information indicating an extraction target document from an anonymized document indicated by each of the multiple anonymized document information stored in the anonymized document storage unit 135, based on the division unit text identification information described above. Here, as shown in FIG. 5(A), for example, the extraction target document generating unit 116 rearranges the division unit texts for each anonymized document indicated by the anonymized document information acquired from the LLM management servers 2A and 2B, based on the division unit text identification information, to generate extraction target document information indicating an anonymized extraction target document for document 1 and extraction target document information indicating an anonymized extraction target document for document 2. As shown in FIG. 5(B), for example, this extraction target document is a document in which "Disease name" is replaced with "__DISEASE*__", "Age" is replaced with "__AGE*__", "Gender" is replaced with "__GENDER*__", "Medicine name" is replaced with "__MEDICINE*__", and "Treatment name" is replaced with "__TREATMENT*__". 3, the extraction target document generating section 116 then stores the generated extraction target document information in the extraction target document storage section 136.

[0023] The extraction information acquisition unit 117 generates extraction request information including prompt information composed of the extraction target document information stored in the extraction target document storage unit 136 and the above-mentioned extraction command information, and transmits the generated extraction request information to the LLM management server 2C. As a result, the extraction information acquisition unit 117 acquires the extraction information transmitted from the LLM management server 2C. Here, the extraction information includes type identification information indicating "MEDICINE1" and "DISEASE2", for example, as shown in FIG. 6(A). Returning to FIG. 3, the extraction information acquisition unit 117 then notifies the acquired extraction information to the extraction information conversion unit 118. The extraction information conversion unit 118 converts the type identification information included in the notified extraction information into a corresponding term. Specifically, the extraction information conversion unit 118 identifies term information corresponding to the type identification information included in the notified extraction information from the term information stored in the anonymized document storage unit 135, and stores the identified term information in the extraction information storage unit 137. The display control unit 119 forms an extracted information notification image based on the term information stored in the extracted information storage unit 137, and causes the display unit 104 to display the extracted information notification image. Here, the display control unit 119 forms an extracted information notification image including the terms "CLO" corresponding to "MEDICINE1" and "Stent thrombosis" corresponding to "DISEASE2", for example, as shown in Fig. 6(B), and causes the display unit 104 to display the extracted information notification image.

[0024] Next, the operation of the information extraction system according to the present embodiment will be described with reference to FIG. 7. First, it is assumed that the user of the terminal device 1 performs the above-mentioned information extraction operation on the input unit 105 of the terminal device 1. In this case, the terminal device 1 stores a plurality of document information items to be extracted, which are designated by the information extraction operation, in the document information storage unit 131 (step S1). Next, the terminal device 1 generates the above-mentioned anonymization command information and extraction command information based on the operation content of the information extraction operation, and stores them in the extraction command storage unit 132 (step S2). Next, the terminal device 1 executes the above-mentioned text analysis for each of the plurality of document information items stored in the document information storage unit 131 (step S3), and identifies the division unit text included in the document indicated by each of the document information items. After that, the terminal device 1 generates a plurality of anonymization target document information items indicating documents to be anonymized that include a part of the division unit text identified for each of the documents indicated by the plurality of document information items stored in the document information storage unit 131 (step S4).

[0025] Next, the terminal device 1 generates a plurality of pieces of anonymization request information including prompt information composed of the document information to be anonymized and the above-mentioned anonymization command information for each of the plurality of pieces of anonymization target information (step S5).Then, the generated plurality of pieces of anonymization request information are each transmitted to one of the LLM management servers 2A and 2B (step S6).

[0026] On the other hand, when the LLM management servers 2A and 2B acquire the anonymization request information, they use a preset LLM to decrypt the contents of the anonymization process based on the text indicated by the anonymization command information constituting the prompt information included in the anonymization request information.Then, the LLM management servers 2A and 2B perform anonymization process on the anonymization target document indicated by the anonymization target document information according to the decrypted contents, and generate anonymized document information indicating the anonymized document (step S7).Then, the generated multiple pieces of anonymized document information are transmitted from the LLM management servers 2A and 2B to the terminal device 1 (step S8).

[0027] On the other hand, when the terminal device 1 acquires a plurality of pieces of anonymized document information transmitted from the LLM management servers 2A and 2B, the terminal device 1 stores each of the acquired pieces of anonymized document information in the anonymized document storage unit 135 in a form in which each of the divided unit texts constituting the pieces of anonymized document information corresponds to the divided unit text identification information based on the delimiter positions of each divided unit text in the anonymized document indicated by the corresponding anonymized document information. Then, the terminal device 1 generates extraction target document information indicating the extraction target document from the anonymized document indicated by each of the plurality of anonymized document information stored in the anonymized document storage unit 135 based on the above-mentioned divided unit text identification information, and stores the extraction target document information in the extraction target document storage unit 136 (step S9). Next, the terminal device 1 generates extraction request information including prompt information composed of the extraction target document information and the above-mentioned extraction command information (step S10). Then, the generated extraction request information is transmitted from the terminal device 1 to the LLM management server 2C (step S11).

[0028] On the other hand, when the LLM management server 2C acquires the above-mentioned extraction request information, it uses a preset LLM to decipher the contents of the extraction process based on the sentence indicated by the above-mentioned extraction command information constituting the prompt information included in the extraction request information.Then, the LLM management server 2C executes the extraction process for the extraction target document indicated by the extraction target document information according to the deciphered contents, and generates extraction information indicating at least one type identification information extracted by the extraction process (step S12).Then, the generated extraction information is transmitted from the LLM management server 2C to the terminal device 1 (step S13).

[0029] On the other hand, when the terminal device 1 acquires the extraction information, it converts the type identification information included in the acquired extraction information into a corresponding term, and stores the term information obtained by the conversion in the extraction information storage unit 137 (step S14). Next, the terminal device 1 forms an extraction information notification image based on the term information stored in the extraction information storage unit 137, and displays it on the display unit 104 (step S15).

[0030] Next, the information extraction process executed by the terminal device 1 according to the present embodiment will be described with reference to FIG. 8. This information extraction process is started when a program for executing the information extraction process is started in the terminal device 1 after the terminal device 1 is powered on. First, the document acquisition unit 111 and the command generation unit 112 determine whether or not an information extraction operation performed by a user on the input unit 105 has been accepted (step S101). Here, as long as the document acquisition unit 111 and the command generation unit 112 determine that an information extraction operation has not been accepted (step S101: No), the process of step S101 is repeatedly executed. On the other hand, it is assumed that the document acquisition unit 111 and the command generation unit 112 determine that an information extraction operation has been accepted (step S101: Yes). In this case, the document acquisition unit 111 acquires a plurality of document information designated in the information extraction operation performed by the user via the input unit 105, and stores the acquired document information in the document information storage unit 131 (step S102). Next, the command generation unit 112 generates the aforementioned anonymized command information and extracted command information based on the information associated with the information extraction operation stored in the command generation information storage unit 138, and stores them in the extracted command storage unit 132 (step S103).

[0031] Next, the text analysis unit 113 performs text analysis on each of the multiple document information stored in the document information storage unit 131 to identify division unit texts included in the document indicated by the document information (step S104). Here, the text analysis unit 113 stores information indicating the delimiter position of each of the identified at least one division unit text in the analysis result storage unit 133. Thereafter, the anonymization target document generation unit 114 identifies division unit texts included in each of the documents indicated by the multiple document information stored in the document information storage unit 131 based on the information stored in the analysis result storage unit 133. Then, the anonymization target document generation unit 114 generates multiple anonymization target document information indicating anonymization target documents including a part of the division unit text identified for each document indicated by the multiple document information, and stores the information in the anonymization target document storage unit 134 (step S105).

[0032] Next, the anonymization unit 115 generates a plurality of pieces of the above-mentioned anonymization request information for each of the plurality of pieces of anonymization target document information, and transmits the generated plurality of pieces of anonymization request information to either one of the LLM management servers 2A and 2B (step S106). As a result, the anonymization unit 115 acquires a plurality of pieces of anonymization document information transmitted from the LLM management servers 2A and 2B (step S107). Here, the anonymization unit 115 stores each of the acquired plurality of pieces of anonymization document information in the anonymization document storage unit 135 in a form in which each of the segmentation unit texts constituting the pieces of anonymization document information corresponds to the segmentation unit text identification information, based on the delimiter positions of each segmentation unit text in the anonymization target document indicated by the corresponding anonymization target document information. In addition, the anonymization unit 115 compares the anonymization target document information stored in the anonymization target document storage unit 134 with the anonymization document information, and stores type identification information for identifying the type of anonymized term in the anonymization document storage unit 135 in association with the term.

[0033] Next, the extraction target document generation unit 116 generates extraction target document information indicating the extraction target document from the anonymized documents indicated by each of the multiple anonymized document information stored in the anonymized document storage unit 135, based on the aforementioned division unit text identification information, and stores the generated extraction target document information in the extraction target document storage unit 136 (step S108).

[0034] Thereafter, the extraction information acquisition unit 117 generates extraction request information including prompt information composed of the extraction target document information stored in the extraction target document storage unit 136 and the above-mentioned extraction command information, and transmits the generated extraction request information to the LLM management server 2C (step S109). As a result, the extraction information acquisition unit 117 acquires the extraction information transmitted from the LLM management server 2C (step S110). Next, the extraction information conversion unit 118 converts the type identification information included in the acquired extraction information into a corresponding term, and stores term information indicating the converted term in the extraction information storage unit 137 (step S111). Next, the display control unit 119 forms an extraction information notification image based on the term information stored in the extraction information storage unit 137, and displays it on the display unit 104 (step S112). Thereafter, the process of step S101 is executed again.

[0035] As described above, according to the information extraction system of the present embodiment, the anonymization target document generation unit 114 generates a plurality of anonymization target document information indicating documents to be anonymized including a part of the division unit text specified for each of the documents indicated by the plurality of document information. The anonymization unit 115 generates the above-mentioned anonymization request information for each of the plurality of anonymization target document information and transmits it to the LLM management servers 2A and 2B, thereby acquiring a plurality of anonymized document information transmitted from the LLM management servers 2A and 2B. Furthermore, the extraction target document generation unit 116 generates a plurality of extraction target document information from the acquired plurality of anonymized document information, and the extraction information acquisition unit 117 generates the above-mentioned extraction request information and transmits it to the LLM management server 2C, thereby acquiring the extraction information transmitted from the LLM management server 2C. Then, the extraction information conversion unit 118 converts the type identification information included in the extraction information into a corresponding term. This makes it possible to extract a term of a type identified by the type identification information specified by the user from the document indicated by the document information, thereby improving the convenience for the user when managing necessary information included in the document information.

[0036] In addition, in the information extraction system according to the present embodiment, there are two LLM management servers 2A and 2B. The anonymization unit 115 transmits two pieces of anonymization target document information indicating documents to be anonymized that are generated by mixing division unit texts included in documents indicated by two pieces of document information to the different LLM management servers 2A and 2B, thereby acquiring anonymized document information transmitted from each of the LLM management servers 2A and 2B. This makes it possible to prevent the contents of the document information before anonymization from being leaked to the business operators who manage the LLM management servers 2A and 2B, thereby improving the security of the document information.

[0037] Although the embodiments of the present invention have been described above, the present invention is not limited to the configuration of each of the above-mentioned embodiments. For example, the anonymization target document generation unit 114 may specify division unit texts included in each of the document indicated by the document information stored in the document information storage unit 131 and a dummy document set in advance, and generate a plurality of anonymization target document information indicating an anonymization target document including a part of the division unit texts specified for each of the document indicated by each document information and the dummy document. In this case, as shown in FIG. 9(A), for example, the anonymization target document generation unit 114 generates two anonymization target documents by mixing the division unit texts of document 1 consisting of division unit texts La1[0], La1[1], . . . , La1[5] and a dummy document consisting of division unit texts La9[0], La9[1], . . . , La9[5]. Here, division unit text identification information La9[0], La9[1], . . . , La9[5] that reflects the order of the division unit texts is assigned to each division unit text of the dummy document. Then, the anonymization target document generating unit 114 stores each of the two anonymization target documents in the anonymization target document storage unit 134 in a manner that associates each divided unit text constituting the anonymization target document with divided unit text identification information. Then, the anonymization unit 115 generates a plurality of the above-mentioned anonymization request information for each of the above-mentioned plurality of anonymization target document information, and transmits the generated plurality of anonymization request information to the common LLM management server 2002, thereby acquiring a plurality of anonymized document information transmitted from the LLM management server 2002. In this case, as shown in FIG. 9(B), the order of each divided unit text in the anonymized document matches the order of each divided unit text in the corresponding anonymization target document shown in FIG. 9(A). Then, the anonymization unit 115 stores each of the acquired plurality of anonymized document information in the anonymization target document storage unit 135 in a manner that associates each divided unit text constituting the document with divided unit text identification information, based on the delimiter positions of each divided unit text in the anonymization target document indicated by the corresponding anonymization target document information.

[0038] Furthermore, the extraction target document generating unit 116 generates extraction target document information indicating the extraction target document based on the division unit text identification information described above from the anonymized document indicated by each of the multiple anonymized document information stored in the anonymized document storage unit 135. Here, the extraction target document generating unit 116 generates extraction target document information indicating the extraction target document obtained by anonymizing document 1 by rearranging the division unit texts based on the division unit text identification information for each anonymized document indicated by the anonymized document information acquired from the LLM management server 2002, as shown in Fig. 9(B) for example.

[0039] According to this configuration, since extraction target document information can be generated using one LLM management server 2002, the information extraction system can be operated even in an environment where only one LLM management server 2002 can be used, for example.

[0040] In an embodiment, the terminal device may further include an abstraction unit that acquires multiple anonymization target document information, some of which have terms abstracted, sent from the large scale language model management unit by transmitting prompt information to the large scale language model management unit, the multiple document information to be anonymized and the abstraction command information requesting abstraction by converting at least one type of terminology that has been preset for the multiple anonymized documents into abstract terminology information indicating abstracted terms.

[0041] 10, the LLM management servers 3002A and 3002B according to this modification have the same functions as the LLM management servers 2A and 2B described in the embodiment, and when they acquire abstraction request information requesting abstraction of at least one term specified by a user included in a document indicated by document information transmitted from the terminal device 1, they perform an abstraction process to abstract at least one term included in the document using a preset LLM, and transmit abstraction document information indicating the abstract document generated by the abstraction process to the terminal device 1. Here, the abstraction request information includes prompt information consisting of anonymization target document information to be abstracted by converting at least one term specified by a user into an abstract term obtained by abstracting the at least one term specified by a user included in the anonymization target document indicated by the anonymization target document information, and abstraction command information indicating a sentence requesting the execution of an abstraction process to convert each of at least one term specified by a user included in the anonymization target document indicated by the anonymization target document information into an abstract term. When the LLM management servers 3002A and 3002B acquire the abstraction request information, they use a preset LLM to decipher the contents of the abstraction process based on the text indicated by the abstraction command information constituting the prompt information included in the abstraction request information. The LLM management servers 3002A and 3002B then perform abstraction processing on the document to be anonymized indicated by the document information to be anonymized according to the deciphered contents, and transmit anonymized document information indicating the anonymized document generated by the abstraction process to the terminal device 1. Here, when the document information indicates a medical document, the at least one term includes at least one of a term indicating the patient's place of residence and a term indicating the patient's age.

[0042] The terminal device 3001 has the same hardware configuration as the terminal device 1 described in the embodiment. Therefore, in the description of this modification, the hardware configuration of the terminal device 3001 will be described using the symbols shown in FIG. 2 as appropriate. The CPU 101 of the terminal device 3001 loads a program stored in the auxiliary storage unit 103 into the main storage unit 102 and executes it, thereby functioning as a document acquisition unit 111, a command generation unit 3112, a text analysis unit 113, an anonymization target document generation unit 114, an anonymization unit 3115, an extraction target document generation unit 116, an extraction information acquisition unit 3117, an extraction information conversion unit 118, a display control unit 119, and an abstraction unit 3120. In FIG. 10, the same components as those in the embodiment are denoted by the same symbols as those in FIG. 3. Moreover, the auxiliary storage unit 102 of the terminal device 3001 has a document information storage unit 131, a command storage unit 3132, an analysis result storage unit 133, an anonymization target document storage unit 134, an anonymized document storage unit 3135, an extraction target document storage unit 3136, an extraction information storage unit 137, and a command generation information storage unit 3138. The command storage unit 3132 stores the above-mentioned anonymization command information and the above-mentioned extraction command information, as well as abstraction command information indicating a sentence requesting the above-mentioned abstraction process to be performed.

[0043] The anonymized document storage unit 3135 stores anonymized document information indicating anonymized documents obtained by performing abstraction processing and anonymization processing on the anonymization target documents indicated by the anonymization target document information by the LLM management servers 3002A and 3002B. The extraction target document storage unit 3136 stores extraction target document information indicating extraction target documents that are the subject of a request for the above-mentioned extraction processing to the LLM management server 2C.

[0044] When a user performs the above-mentioned information extraction operation on the input unit 105, the command generation unit 3112 generates the above-mentioned abstraction command information, anonymization command information and extraction command information using text information corresponding to the information extraction operation stored in the command generation information storage unit 3138, and stores them in the command storage unit 3132.

[0045] The abstraction unit 3120 generates the above-mentioned abstraction request information for each of the multiple anonymization target document information, and transmits the generated abstraction request information to either one of the LLM management servers 3002A and 3002B, thereby acquiring the multiple anonymization target document information abstracted for at least one term specified by the user, which is transmitted from the LLM management servers 3002A and 3002B. Then, the abstraction unit 3120 notifies the anonymization unit 3115 of the multiple anonymization target document information abstracted for at least one term. In addition, the abstraction unit 3120 compares the anonymization target document information stored in the anonymization target document storage unit 134 with the anonymized document information abstracted for at least one term, thereby storing the abstracted term in the anonymized document storage unit 3135 in association with the term before abstraction.

[0046] The anonymization unit 3115 generates a plurality of the above-mentioned anonymization request information for each of the plurality of anonymization target document information notified from the abstraction unit 3120, and transmits the generated plurality of anonymization request information to one of the LLM management servers 3002A and 3002B, thereby acquiring a plurality of anonymized document information transmitted from the LLM management servers 3002A and 3002B. Here, the type of term corresponding to the anonymization command information included in the anonymization request information is different from the type of term corresponding to the abstraction command information included in the abstraction request information. Then, the anonymization unit 3115 stores each of the acquired plurality of anonymized document information in the anonymized document storage unit 3135 in a form in which each of the divided unit texts constituting the anonymization target document information corresponds to the divided unit text identification information, based on the delimiter position of each divided unit text in the anonymization target document indicated by the corresponding anonymization target document information. For example, it is assumed that the document information storage unit 131 stores text information indicating a text as shown in FIG. 11(A). In this case, the abstraction unit 3120 and the anonymization unit 3115 generate anonymized document information indicating an anonymized document in which, for example, as shown in Fig. 11(B), for the document shown in Fig. 11(A), the "area of ​​residence" "Kyoto City" is abstracted to "Kinki region", the "age" "66 years old" is abstracted to "60s", and the "disease name" is replaced with "__DISEASE*__", "age" with "__AGE*__", "gender" with "__GENDER*__", "medicine name" with "__MEDICINE*__", and "treatment name" with "__TREATMENT*__". Here, "*" represents any positive integer.

[0047] Returning to FIG. 10, the extraction target document generating unit 3116 generates a plurality of pieces of extraction target document information from a plurality of pieces of anonymized document information based on the above-mentioned division unit text identification information, depending on whether or not the anonymized document indicated by each of the plurality of pieces of anonymized document information stored in the anonymized document storage unit 3135 includes the abstract term specified by the user. Specifically, if any of the anonymized documents indicated by each of the plurality of pieces of anonymized document information includes the abstract term specified by the user, the extraction target document generating unit 3116 generates a plurality of pieces of extraction target document information from the plurality of pieces of anonymized document information. On the other hand, if the anonymized document indicated by each of the plurality of pieces of anonymized document information does not include the abstract term specified by the user, the extraction target document generating unit 3116 does not generate extraction target document information. Then, the extraction target document generating unit 3116 stores the generated extraction target document information in the extraction target document storage unit 136.

[0048] Next, the information extraction process executed by the terminal device 3001 according to this modification will be described with reference to FIG. 12. Note that in FIG. 12, the same processes as those in the embodiment are denoted by the same reference numerals as those in FIG. 8. First, when the document acquisition unit 111 and the command generation unit 3112 determine that the above-mentioned information extraction operation has been received (step S101: Yes), the process of step 102 is executed. This information extraction operation is an operation to specify the patient's location (place of residence), the patient's age group, the diagnosis name, the drug name, and the like, while the display control unit 119 of the terminal device 3001 is displaying, for example, an operation screen image GA31 as shown in FIG. 13(A) on the display unit 104. Returning to FIG. 12, next, the command generation unit 3112 generates the above-mentioned abstraction command information, anonymization command information, and extraction command information based on information associated with the information extraction operation stored in the command generation information storage unit 138, and stores them in the extraction command storage unit 132 (step S3101). Next, after the processes of steps S104 and S105 are executed, the abstraction unit 3120 generates the above-mentioned abstraction request information for each of the multiple anonymization target document information, and transmits the generated abstraction request information to either one of the LLM management servers 3002A and 3002B (step S3102). As a result, the abstraction unit 3120 acquires the multiple anonymization target document information abstracted for at least one term specified by the user, which is transmitted from the LLM management servers 3002A and 3002B (step S3103). At this time, the abstraction unit 3120 notifies the anonymization unit 3115 of the multiple anonymization target document information abstracted for at least one term.

[0049] Thereafter, the anonymization unit 3115 generates a plurality of the above-mentioned anonymization request information for each of the plurality of anonymization target document information notified from the abstraction unit 3120, transmits the generated plurality of anonymization request information to either of the LLM management servers 3002A and 3002B (step S106), and acquires a plurality of anonymized document information transmitted from the LLM management servers 3002A and 3002B (step S107). Here, the anonymization unit 115 stores each of the acquired plurality of anonymized document information in the anonymized document storage unit 135 in a form in which each of the division unit texts constituting the plurality of anonymized document information corresponds to the division unit text identification information based on the delimiter position of each division unit text in the anonymization target document indicated by the corresponding anonymization target document information. In addition, the anonymization unit 115 compares the anonymization target document information stored in the anonymization target document storage unit 134 with the anonymized document information, and stores type identification information for identifying the type of anonymized term in the anonymized document storage unit 135 in association with the term.

[0050] Next, the extraction target document generating unit 3116 generates extraction target document information indicating the extraction target document from the plurality of anonymized document information based on the above-mentioned division unit text identification information according to whether or not the anonymized document indicated by each of the plurality of anonymized document information stored in the anonymized document storage unit 3135 contains the abstract term specified by the user, and stores the generated extraction target document information in the extraction target document storage unit 136 (step S3104). Here, if any of the anonymized documents indicated by each of the plurality of anonymized document information contains the abstract term specified by the user, the extraction target document generating unit 3116 generates the plurality of extraction target document information from the plurality of anonymized document information. On the other hand, if the anonymized document indicated by each of the plurality of anonymized document information does not contain the abstract term specified by the user, the extraction target document generating unit 3116 does not generate the extraction target document information. After that, a series of processes from step S109 onwards are executed.

[0051] As a result, the display control unit 119 of the terminal device 3001 displays on the display unit 104 an extracted information notification image GA32 that includes, for example, as shown in Figure 13 (B), "generation" information that abstracts the patient's age and "location" information that abstracts the patient's place of residence.

[0052] According to this configuration, it becomes possible to provide a user with information based on document information selected based on abstracted terms.

[0053] In the embodiment, an example has been described in which the LLM management servers 2A and 2B perform an anonymization process using a large-scale language model when anonymization request information including anonymization target document information and anonymization command information transmitted from the terminal device 1 is acquired, and transmit anonymized document information indicating anonymized documents generated by the anonymization process to the terminal device 1. However, the present invention is not limited to this, and the information extraction system may not include the LLM management servers 2A and 2B. In this case, the information extraction system may include a type identification information management server having a type identification information storage unit that stores type identification information for identifying a term type for at least one type of term set in advance in association with term information indicating the corresponding term, a text analysis unit that performs text analysis on the anonymization target document indicated by the acquired anonymization target document information when the anonymization target document information is acquired, a term replacement unit that replaces a term included in the anonymization target document with type identification information based on a result of the text analysis by referring to the type identification information stored in the type identification information storage unit, and an anonymized document notification unit that transmits anonymized document information indicating an anonymized document in which some terms have been replaced with the type identification information to the terminal device. The terminal device may transmit anonymization request information including the document information to be anonymized to the type identification information management server, thereby acquiring the document information to be anonymized transmitted from the type identification information management server.

[0054] Alternatively, the terminal device may include a type identification information storage unit that stores type identification information that identifies a type of term for at least one type of term that is preset, in correspondence with term information indicating the corresponding term, and an anonymization unit that refers to the type identification information stored in the type identification information storage unit and replaces terms included in the document to be anonymized with the type identification information.

[0055] In the embodiment, the document information is not limited to medical information such as electronic medical record information, but may indicate documents in other fields.

[0056] The functions of the LLM management servers 2A, 2B, and 2C in the embodiments may be distributed and realized in a plurality of physical servers.

[0057] In the embodiment, an example has been described in which the prompt information included in the anonymization request information is composed of the anonymization target document information indicating the document to be anonymized and the anonymization command information indicating a sentence requesting to perform an anonymization process for converting at least one term included in the document to be anonymized into type identification information. However, the present invention is not limited to this, and the prompt information may be, instead of the document to be anonymized, storage location identification information such as a URL (Uniform Resource Locator) that identifies a location on a network where the document to be anonymized is stored. In this case, the anonymization command information may include instruction information to acquire the document to be anonymized from the storage location of the document to be anonymized identified by the storage location identification information. Furthermore, the prompt information included in the extraction request information is not limited to being composed of the extraction target document information indicating the document to be extracted and the extraction command information indicating a sentence requesting to perform an extraction process for extracting at least one type identification information specified by a user included in the extraction target document. The prompt information may be, for example, storage location identification information such as a URL that identifies a location on a network where the document to be extracted is stored. In this case, the extraction command information may include instruction information for acquiring the extraction target document information from the storage location of the extraction target document information identified by the storage location identification information.

[0058] Moreover, the various functions of the terminal device 1 according to the present invention can be realized by using a normal computer system, not by a dedicated system. For example, a program for executing the above operations may be stored in a non-transitory recording medium (such as a CD-ROM (Compact Disc Read Only Memory)) that can be read by a computer system and distributed to a computer connected to a network, and the terminal device 1 that executes the above-mentioned processes may be configured by installing the program in the computer system.

[0059] The method of providing the program to the computer is arbitrary. For example, the program may be uploaded to a bulletin board (BBS (Bulletin Board System)) on a communication line and distributed to the computer via the communication line. The computer then starts up the program and executes it under the control of an OS (Operating System) in the same way as other applications. In this way, the computer functions as a terminal device 1 that executes the above-mentioned processing.

[0060] 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 appropriate combinations of the embodiments and modifications, and appropriate modifications thereto. [Industrial Applicability]

[0061] The present invention is suitable as a system for managing user-specified types of information contained in document information. [Explanation of symbols]

[0062] 1, 3001: terminal device, 2A, 2B, 2C, 2002, 3002A, 3002B: LLM management server, 101: CPU, 102: main memory unit, 103: auxiliary memory unit, 104: display unit, 105: input unit, 106: communication unit, 109: bus, 111: document acquisition unit, 112, 3112: extraction command generation unit, 113: text analysis unit, 114: anonymization target document generation unit, 115: anonymization unit, 116, 3116: extraction document generation unit, 117: extracted information acquisition unit, 118: extracted information acquisition unit, 119: extracted information conversion unit, 120: display control unit, 131: document information storage unit, 132, 3132: extracted command storage unit, 133: analysis result storage unit, 134: document storage unit for anonymization, 135, 3135: anonymized document storage unit, 136: extracted information storage unit, 137: extracted command generation information storage unit, 3120: abstraction unit, NW1: network

Claims

1. A terminal device; A large-scale language model management unit for managing a large-scale language model, The terminal device an extraction information acquisition unit that transmits, to the large-scale language model management server, first prompt information requesting an extraction process to extract type identification information designated by a user from an extraction target document that has been anonymized by converting at least one type of term set in advance into type identification information that identifies a type of term, and thereby acquires extraction information indicating at least one type of type identification information extracted by the extraction process transmitted from the large-scale language model management server; an extraction information conversion unit that converts the type identification information included in the extraction information into a corresponding term, Information Extraction System.

2. the first prompt information includes: extraction target document information indicating the extraction target document from which at least one type identification information specified by a user is to be extracted; and extraction command information indicating a sentence requesting to perform an extraction process for extracting the type identification information specified by the user that is included in the extraction target document indicated by the extraction target document information; when the large-scale language model management server receives the first prompt information transmitted from the terminal device, the large-scale language model management server performs the extraction process using a large-scale language model set in advance, based on the extraction target document information and the extraction command information included in the first prompt information, and transmits extraction information indicating at least one of the type identification information extracted by the extraction process to the terminal device. The information extraction system according to claim 1 .

3. the large-scale language model management server, upon receiving second prompt information transmitted from the terminal device, the large-scale language model management server performs the anonymization process based on the anonymization target document information and the anonymization command information included in the second prompt information, the anonymization target document information indicating an anonymization target document to be anonymized by converting at least one predetermined term into type identification information that identifies a term type, using a large-scale language model that is set in advance, based on the anonymization target document information and the anonymization command information included in the second prompt information, and transmits anonymized document information indicating the anonymized document generated by the anonymization process to the terminal device; The terminal device a text analysis unit that performs text analysis on each of a plurality of documents to identify division unit texts included in the document indicated by the document information to be anonymized; an anonymization target document generation unit that generates a plurality of pieces of the anonymization target document information indicating the anonymization target document including a portion of a division unit text identified for each of the documents indicated by the plurality of pieces of document information; an anonymization unit that transmits the second prompt information, which includes the plurality of pieces of document information to be anonymized and the anonymization command information, to the large-scale language model management server, thereby acquiring the plurality of pieces of anonymized document information transmitted from the large-scale language model management server; and an extraction target document generation unit that generates a plurality of pieces of extraction target document information from the acquired plurality of pieces of anonymized document information. The information extraction system according to claim 2.

4. The large scale language model management unit includes: a first large-scale language model management server that, upon receiving anonymization target document information transmitted from the terminal device, indicating an anonymization target document to be anonymized by converting at least one predetermined term into type identification information that identifies a term type, and anonymization command information indicating a sentence requesting to perform an anonymization process for converting each of the at least one term included in the anonymization target document indicated by the anonymization target document information into the type identification information, performs the anonymization process using a predetermined large-scale language model, and transmits anonymized document information indicating the anonymized document generated by the anonymization process to the terminal device; a second large-scale language model management server that, upon receiving the first prompt information transmitted from the terminal device, performs the extraction process based on the extraction target document information and the extraction command information included in the first prompt information by using a preset large-scale language model, and transmits extraction information indicating at least one of the type identification information extracted by the extraction process to the terminal device. The information extraction system according to claim 3.

5. the large-scale language model management unit, upon receiving third prompt information transmitted from the terminal device, the large-scale language model management unit performs the abstraction process based on the anonymization target document information and the abstraction command information included in the third prompt information, the anonymization target document being a target for abstraction by converting at least one predetermined term into abstract term information indicating an abstract term obtained by abstracting the term, using a large-scale language model; The terminal device further comprises: an abstraction unit that transmits a plurality of pieces of document information to be anonymized and the abstraction command information to the large-scale language model management unit, and acquires the plurality of pieces of anonymized document information transmitted from the large-scale language model management unit; the anonymization unit transmits the anonymization target document information generated by the plurality of abstraction processes and the anonymization command information to the large-scale language model management unit, thereby acquiring the plurality of pieces of anonymized document information transmitted from the large-scale language model management unit; the extraction target document generation unit generates a plurality of pieces of extraction target document information from the plurality of pieces of anonymized document information based on whether or not the plurality of pieces of anonymized document information obtained includes an abstract term designated by a user.

5. The information extraction system according to claim 3 or 4.

6. The document information to be anonymized indicates a medical-related document, the at least one term includes at least one of a term indicating an attribute of a patient, a term indicating a name of a medicine, a term indicating a name of a disease, a term indicating a symptom of a patient, a term indicating a condition of a patient, and a term indicating a content of a treatment for a patient; The information extraction system according to claim 4.

7. There are a plurality of first large-scale language model management servers; the anonymization unit transmits the plurality of pieces of document information to be anonymized and the anonymization command information corresponding to each of the plurality of pieces of document information to be anonymized to the first large-scale language model management servers different from each other, thereby acquiring the plurality of pieces of anonymized document information transmitted from each of the first large-scale language model management servers; The information extraction system according to claim 4 or 6.

8. The text analysis unit performs text analysis on each of the plurality of pieces of document information including the document information indicating a dummy document; the anonymization target document generation unit generates a plurality of pieces of the anonymization target document information indicating the anonymization target document including a portion of the division unit text identified for a plurality of documents including the dummy document; 5. The information extraction system according to claim 3 or 4.

9. The functions of the first large-scale language model management server and the second large-scale language model management server are realized by one large-scale language model management server that manages the large-scale language model. The information extraction system according to claim 4.

10. A terminal device used together with a large-scale language model management unit for managing a large-scale language model, an extraction information acquisition unit that transmits prompt information to the large-scale language model management unit, the prompt information requesting the large-scale language model management unit to perform an extraction process for extracting type identification information designated by a user from an extraction target document that has been anonymized by converting at least one type of term set in advance into type identification information that identifies a type of term, and thereby acquires extraction information indicating at least one type of type identification information extracted by the extraction process transmitted from the large-scale language model management unit; an extraction information conversion unit that converts the type identification information included in the extraction information into a corresponding term, Information extraction device.

11. An information extraction method using a large-scale language model management unit that manages a large-scale language model, comprising: a step of transmitting prompt information to the large-scale language model management unit, which prompts the large-scale language model management unit to perform an extraction process for extracting type identification information designated by a user from an extraction target document that has been anonymized by converting at least one type of term set in advance into type identification information that identifies a type of term, thereby obtaining extraction information indicating at least one type of type identification information extracted by the extraction process, which is transmitted from the large-scale language model management unit; and converting the type identification information included in the extracted information into a corresponding term. Information extraction method.

12. Computer, an extraction information acquisition unit that transmits prompt information to a large-scale language model management unit that manages a large-scale language model, requesting the execution of an extraction process for extracting type identification information designated by a user from an extraction target document that has been anonymized by converting at least one type of term set in advance into type identification information that identifies a type of term, and thereby acquires extraction information indicating at least one type of type identification information extracted by the extraction process transmitted from the large-scale language model management unit; an extraction information conversion unit that converts the type identification information included in the extraction information into a corresponding term; A program to function as a

Images