Information processing device, information processing method, and recording medium

Abstract

In this invention, a selection support device includes an acquisition unit, a generation unit, and a conversion unit. The acquisition unit acquires one or more phrases to be structured. On the basis of a relationship between terms included in the phrases, the generation unit generates an intermediate expression used for structuring the phrases. The conversion unit converts the phrases into a structured text on the basis of the intermediate expression. Accuracy of converting target phrases into a structured text is improved by using such an intermediate expression when converting the target phrases into the structured text. Thus, decision making can be appropriately performed on the basis of information that is extracted using such a structured text.

Description

Information processing device, information processing method, and recording medium

[0001] The present disclosure relates to an information processing device and the like.

[0002] In the medical field, for example, information extraction is sometimes performed from information related to treatment in order to confirm patient information. For example, in a clinical trial of a pharmaceutical product, a person in charge of extracting patients who meet the selection criteria extracts patients who meet the selection criteria for clinical trial patients by checking the descriptions in the medical records. The person in charge then determines, for example, whether the selection criteria match the contents of the medical records. The person in charge then extracts patients whose selection criteria match the contents of the electronic medical records as candidate clinical trial patients. In addition, a system that supports information extraction may be used to extract patient information.

[0003] The information processing system of Patent Document 1 structures text information entered on a predetermined screen, and then identifies patients who are candidates for clinical trial subjects using the structured text information and reference information for the clinical trial.

[0004] JP 2022-180080 A

[0005] In the information processing system described in Patent Document 1, it may be difficult to convert information into text suitable for extraction.

[0006] In order to solve the above-mentioned problems, an object of the present disclosure is to provide an information processing device and the like that can improve the accuracy of conversion into structured text.

[0007] In order to solve the above problems, the information processing device of the present disclosure includes an acquisition means for acquiring a sentence to be structured, a generation means for generating an intermediate representation to be used for structuring the sentence based on the relationship between terms contained in the sentence, and a conversion means for converting the sentence into structured text based on the intermediate representation.

[0008] The information processing method disclosed herein acquires a sentence to be structured, generates an intermediate representation to be used for structuring the sentence based on the relationships between terms contained in the sentence, and converts the sentence into structured text based on the intermediate representation.

[0009] The recording medium of the present disclosure non-temporarily records a program that causes a computer to execute the following processes: a process of acquiring a sentence to be structured; a process of generating an intermediate representation to be used for structuring the sentence based on the relationship between terms contained in the sentence; and a process of converting the sentence into structured text based on the intermediate representation.

[0010] According to the present disclosure, the accuracy of conversion to structured text can be improved.

[0011] FIG. 1 is a diagram illustrating an example of the configuration of an extraction support system according to the present disclosure. FIG. 2 is a diagram illustrating an example of the configuration of an extraction support system used in selecting clinical trial patients according to the present disclosure. FIG. 3 is a diagram illustrating an example of the configuration of an information processing device according to the present disclosure. FIG. 4 is a diagram illustrating an example of a sentence to be structured according to the present disclosure. FIG. 5 is a diagram illustrating an example of an intermediate representation generated from a sentence to be structured according to the present disclosure. FIG. 6 is a diagram illustrating an example of a sentence indicating inclusion criteria according to the present disclosure. FIG. 7 is a diagram illustrating an example of structured inclusion criteria according to the present disclosure. FIG. 8 is a diagram illustrating an example of a sentence indicating exclusion criteria according to the present disclosure. FIG. 9 is a diagram illustrating an example of structured exclusion criteria according to the present disclosure. FIG. 10 is a diagram illustrating an example of structured information on treatment according to the present disclosure. FIG. 11 is a diagram illustrating an example of a sentence of a summary of information on treatment according to the present disclosure. FIG. 12 is a diagram illustrating an example of an extraction result according to the present disclosure. FIG. 13 is a diagram illustrating an example of the operation flow of an information processing device according to the present disclosure. FIG. 14 is a diagram illustrating an example of the hardware configuration of an information processing device according to the present disclosure.

[0012] An embodiment of the present disclosure will be described in detail with reference to the drawings. FIG. 1 is a diagram illustrating an example of the configuration of an extraction support system. The extraction support system includes, for example, an information processing device 10, a terminal device 20, and a data management device 30. The information processing device 10 is connected to the terminal device 20, for example, via a network. The information processing device 10 is connected to the data management device 30, for example, via a network. Furthermore, there may be a plurality of terminal devices 20 and a plurality of data management devices 30. The number of each of the terminal devices 20 and the data management devices 30 may be set as appropriate.

[0013] The extraction support system, for example, converts a target sentence into structured text. Furthermore, the extraction support system, for example, uses the structured text to extract information from data related to patient treatment. The extraction support system, for example, is used to extract candidate clinical trial patients when selecting clinical trial patients. Furthermore, the extraction support system may be used, for example, to extract information that needs to be confirmed when determining whether a patient meets the selection criteria for clinical trial patients when selecting clinical trial patients.

[0014] Structuring, for example, refers to converting mutually related information contained in a sentence into a state in which the relationships can be understood. A state in which the relationships can be understood is, for example, a state in which mutually related information contained in a sentence is extracted and the extracted information is associated. For example, if there is a sentence that says, "A is B. C is D," structuring refers to extracting "A," "B," "C," and "D" from the sentence and converting it into data in which "A" and "B" and "C" and "D" are associated with each other.

[0015] The extraction support system is used, for example, in selecting clinical trial patients, to extract information from treatment information that needs to be confirmed in determining whether a patient meets the selection criteria for clinical trial patients. A clinical trial is, for example, a clinical trial conducted to obtain legal approval for the manufacture and sale of pharmaceuticals or medical devices. A clinical trial patient is, for example, a patient who is the subject of a clinical trial. The extraction support system is used, for example, to extract patients to be selected as clinical trial patients when a clinical trial is planned. The extraction support system may also be used to estimate the number of clinical trial patients at the stage of creating a clinical trial protocol. The uses of the extraction support system are not limited to those described above.

[0016] The target text is, for example, a text that indicates the content of the information to be extracted. For example, in selecting clinical trial patients, the target text is a text that indicates the selection criteria for clinical trial patients. The selection criteria are, for example, the criteria for selecting clinical trial patients. When extracting candidate clinical trial patients, the target text is, for example, information about the patient's treatment. The information about the patient's clinical trial is, for example, information recorded in the patient's medical record. The information about the patient's clinical trial is not limited to information recorded in the medical record. The extraction support system, for example, converts the target text into an intermediate representation for generating structured text. Then, the extraction support system uses the intermediate representation to convert the target text into structured text. In other words, the target text is also the target for generating structured text.

[0017] The intermediate representation is, for example, text that contains information necessary to convert a target sentence into structured text. In other words, the intermediate representation is text in an intermediate state between the target sentence and the text obtained by structuring the target sentence. Therefore, the intermediate representation is text that has a different expression form from, for example, both the target sentence and the text obtained by structuring the target sentence. Furthermore, the target sentence is, for example, text to which no information for structuring has been added or converted. The intermediate representation text and the structured text are text to which some processing has been performed on the target sentence.

[0018] The intermediate representation is, for example, text in which information indicating the relationship between terms contained in the sentence is added to the sentence. The relationship between terms is, for example, an inclusion relationship between the terms. The relationship between terms may also be a chronological causal relationship between the terms. The relationship between terms may also be, for example, both an inclusion relationship between the terms and a chronological causal relationship between the terms. The relationship between terms is not limited to the above.

[0019] The term may be, for example, a term related to the information to be extracted. For example, the term may be a phrase to be labeled. When extracting candidate clinical trial patients, the term related to the extraction of candidate clinical trial patients is a term related to information that needs to be confirmed when extracting candidate clinical trial patients. The information that needs to be confirmed when extracting candidate clinical trial patients is, for example, information indicating the patient's condition that may affect the conduct and results of the clinical trial. The term related to the extraction of candidate clinical trial patients is, for example, a term indicating one or more of the patient's condition, such as disease name, treatment details, drug name, test, test results, medical condition, and pre-existing condition. The term related to the extraction of candidate clinical trial patients may include patient attributes. The patient attributes are information indicating the patient's condition that does not change with treatment. The patient attributes are, for example, one or more of the patient's age, gender, weight, height, race, and family medical history. The patient attributes are not limited to the above. Furthermore, the term related to the extraction of candidate clinical trial patients is not limited to the above.

[0020] The intermediate representation is, for example, text in which each term included in the target sentence is labeled. The label is, for example, information indicating the relationship between the terms included in the target sentence. For example, the label may indicate an inclusion relationship between the terms. The label may also indicate a chronological relationship between the terms. The label is, for example, information indicating the role of each term in the sentence. For example, the label is information indicating at least one of the part of speech, classification, and position in the sentence of the term. The information indicated by the label is not limited to the above.

[0021] Furthermore, when extracting candidate clinical trial patients, the information regarding treatment is, for example, records related to the patient's treatment. The information regarding treatment is, for example, information recorded in a medical record and test data. The information recorded in a medical record is, for example, information recorded in an electronic medical record. The information regarding treatment may be, for example, either the information recorded in the medical record or test data. The information regarding treatment may also be information contained in medical accounting data or medical receipts. A medical receipt is, for example, a statement of medical fees that a medical institution bills an insurer. A medical institution is, for example, a hospital or a clinic. A medical institution may include a pharmacy that prescribes medications. Medical institutions are not limited to the above. The information regarding treatment may also be information contained in medical literature. Medical literature may include guidelines regarding treatment. The information regarding treatment is not limited to the above.

[0022] The selection criteria are, for example, criteria for including patients in a clinical trial. The selection criteria may also be criteria for excluding patients from a clinical trial. The selection criteria may include both inclusion criteria and exclusion criteria. The inclusion criteria are, for example, information indicating the conditions for patients eligible for the clinical trial. The conditions for patients eligible for the clinical trial are indicated using the attributes and treatment history of patients suitable for the clinical trial. The exclusion criteria are information indicating the conditions for excluding patients from the clinical trial. That is, the exclusion criteria are, for example, information indicating the conditions for patients not to be selected as clinical trial patients. The exclusion criteria are indicated using the attributes and treatment history of patients not suitable for the clinical trial. The patient attributes are, for example, patient conditions that do not change with treatment. The patient attributes are, for example, information on one or more of the patient's age, gender, weight, height, and race. The patient attributes are not limited to the above. The patient's treatment history is the treatment performed on the patient for treatment and the patient's condition that changes as a result of the treatment. The patient's treatment history is, for example, information on one or more of surgery, examinations, medication, test results, and follow-up results. The patient's medical history is not limited to the above.

[0023] FIG. 2 is a diagram schematically illustrating an example of the configuration of an extraction support system that extracts candidate clinical trial patients. In the example of FIG. 2, a data management device 30 stores, for example, information related to treatment. In the example of FIG. 2, a medical institution staff member, such as a doctor, inputs, for example, a sentence indicating selection criteria as a sentence to be structured. The information processing device 10 converts the sentence indicating the selection criteria into structured text, for example, using an intermediate representation generated from the input sentence indicating the selection criteria. Then, the information processing device 10 extracts patients who meet the selection criteria, for example, using the structured selection criteria.

[0024] The person in charge at a medical institution is, for example, one or both of a person belonging to the medical institution and a person in charge of an institution that has been entrusted with work by the medical institution. The person in charge at a medical institution is, for example, a doctor, nurse, or clinical laboratory technician. The person in charge at a medical institution is not limited to the above. Furthermore, an institution that a medical institution entrusts with handling clinical trials is, for example, an SMO (Site Management Organization). Furthermore, a person in charge at an institution that a hospital entrusts with handling clinical trials is, for example, a CRC (Clinical Research Coordinator). The person in charge at a medical institution is not limited to the above.

[0025] A medical institution staff member may, for example, receive an inquiry from a pharmaceutical company staff member about information on patients who meet the selection criteria for patients to be included in the clinical trial. A staff member at an institution entrusted with a clinical trial by a pharmaceutical company conducting a clinical trial of a new drug may, for example, send the selection criteria to the medical institution staff member and inquire about information on patients who meet the selection criteria. The medical institution staff member may, for example, input text indicating the selection criteria corresponding to the inquiry into the information processing device 10 via the terminal device 20. The medical institution staff member may then, for example, obtain information from the information processing device 10 via the terminal device 20 indicating whether or not information regarding treatment meets the selection criteria. The medical institution staff member may, for example, refer to the obtained information to determine whether each patient is suitable as a candidate for a clinical trial patient.

[0026] The person in charge at a pharmaceutical company may be, for example, a person in charge at an institution that has been entrusted with a clinical trial by a pharmaceutical company that is conducting a clinical trial of a new drug. The institution that has been entrusted with a clinical trial by a pharmaceutical company may be, for example, a CRO (Contract Research Organization). The person in charge at the institution that has been entrusted with a clinical trial by a pharmaceutical company may be, for example, a CRA (Clinical Research Associate). The CRA may be an employee of the pharmaceutical company.

[0027] An inquiry about information on patients who meet the selection criteria for patients to be included in a clinical trial is made, for example, when a CRA prepares a clinical trial protocol. The above-mentioned inquiry may also be made when the clinical trial is being carried out in accordance with the clinical trial protocol. The clinical trial protocol is used, for example, to explain the contents of the clinical trial to and negotiate with medical institutions, and to submit notifications to related institutions. The uses of the clinical trial protocol are not limited to the above. Furthermore, the timing of an inquiry about information on patients who meet the selection criteria is not limited to the stage of preparing the clinical trial protocol.

[0028] The information to be extracted is not limited to information regarding treatments used in the extraction of candidate clinical trial patients. Furthermore, the information to be extracted is not limited to information in the medical field. For example, the information to be extracted may be information contained in legislative, administrative, or judicial documents. Furthermore, the information to be extracted may be information contained in documents in the educational or industrial fields. The fields in which the information to be extracted is used are not limited to the above. Furthermore, the information to be extracted may be information contained in papers, books, guidelines, manuals, laws and regulations, official gazettes, court records, minutes, public relations documents, newspapers, press releases, diaries, work records, answers, or applications. Furthermore, the information to be extracted may be information contained in web entries or posts to SNS (social networking services). The source of information extraction is not limited to the above.

[0029] Here, a specific example of the configuration of the information processing device 10 will be described. Fig. 3 is a diagram showing an example of the configuration of the information processing device 10. The information processing device 10 basically includes an acquisition unit 11, a generation unit 12, and a conversion unit 13. The information processing device 10 may also include, for example, an extraction unit 14, an output unit 15, and a storage unit 16.

[0030] The acquisition unit 11 acquires a sentence to be structured. The sentence to be structured is a sentence that is the source of conversion when converted into structured text. The sentence to be structured is, for example, a sentence that indicates the content of the information to be extracted. For example, when extracting candidate clinical trial patients, the sentence to be structured is, for example, a sentence that indicates the selection criteria for the clinical trial patients and information about the treatment. For example, when extracting candidate clinical trial patients, when a medical institution's staff member checks information about the treatment, the sentence to be structured is, for example, a sentence that indicates the selection criteria for the clinical trial patients.

[0031] The acquiring unit 11 acquires sentences to be structurized, for example, from the terminal device 20. The acquiring unit 11 also acquires data including information to be extracted, for example, from the data management device 30. For example, when extracting candidate clinical trial patients, the data including information to be extracted is information related to the patient's treatment. Furthermore, the data including information to be extracted is not limited to information related to the patient's treatment.

[0032] Figure 4 shows an example of information related to a patient's treatment. The example in Figure 4 is a sentence written in the doctor's findings column included in the electronic medical record, which is part of the information related to the patient's treatment. The example sentence in Figure 4 lists the details of the treatment along with the date indicating the date of the consultation. For example, the sentence "2 / 19: Suspected lung cancer. Imaging diagnosis required" in the example sentence in Figure 4 indicates that the patient was suspected of having lung cancer during the consultation on February 19th, and that confirmation by imaging diagnosis was required.

[0033] The generation unit 12 generates an intermediate representation to be used for structuring a sentence based on the relationships between terms included in the sentence. The intermediate representation is, for example, text containing information necessary for structuring the sentence. The information necessary for structuring the sentence is, for example, labels used for structuring the sentence. The intermediate representation is, for example, text obtained by adding the results of semantic understanding of the sentence to the sentence as information necessary for structuring the sentence. In other words, the intermediate representation is, for example, text obtained by converting the sentence into a format including information indicating the relationships between terms based on the results of semantic understanding. For example, the intermediate representation is text to which information indicating the relationships between terms included in the sentence is added. The information indicating the relationships between terms is, for example, labels attached to terms. Labels attached to terms are also called tags. Semantic understanding refers to, for example, understanding the relationships between terms included in a sentence. In other words, the result of semantic understanding is the result of analyzing the relationships between terms included in the sentence. The relationships between terms in semantic understanding are, for example, at least one of an inclusion relationship and a causal relationship between terms. The relationships between terms may also be understood based on knowledge in a field related to the sentence. For example, if the text is a selection criterion for clinical trial patients, the relationship between terms in semantic understanding is understood based on knowledge in the medical field.

[0034] The intermediate representation is, for example, text obtained by converting a sentence into a format that includes information indicating the relationships between terms based on the results of semantic understanding. For example, the intermediate representation is text to which information indicating the relationships between terms included in the sentence is added. The information indicating the relationships between terms is, for example, labels attached to terms. Labels attached to terms are also called tags.

[0035] The generation unit 12 generates an intermediate representation through, for example, multiple processing steps. The multiple processing steps are performed, for example, as follows: The generation unit 12 extracts terms from a sentence using, for example, a morphological analysis method. Then, the generation unit 12 adds labels to the terms included in the sentence using, for example, a dictionary. The generation unit 12 also identifies causal relationships in time series between terms. Then, the generation unit 12 generates an intermediate representation by adding the causal relationships in time series to the labels.

[0036] The generation unit 12 may also identify a period during which a treatment was performed based on the causal relationships between terms. The period during which a treatment was performed is used, for example, to identify a range of the target sentence for which structured text is to be generated. The period during which a treatment was performed is, for example, a period during which a series of actions was performed. For example, if the sentence is information regarding treatment, the generation unit 12 may identify a treatment period based on the causal relationships between terms. The treatment period is, for example, a period from the start of diagnosis to the completion of treatment. By generating an intermediate representation and converting it into structured text based on the range corresponding to the treatment period, it may be easier, for example, for the extraction unit 14 to compare the selection criteria with the structured text. The treatment period is not limited to the above. The period during which a treatment was performed is also not limited to the above.

[0037] The dictionary includes, for example, data associating terms in the field in which the target sentence is used with the content of the label for each term. When the target sentence is used in the medical field, the dictionary includes, for example, data associating terms in the medical field with the content of the label for each term. The content of the label is, for example, information indicating the classification of the term. The content of the label may include, for example, the part of speech of the term. The classification of the term is, for example, information indicating the item to which the term applies, such as the situation in which the term is used in the medical field, disease name, period, subject, guideline name, standard, drug name, or condition. The situation in which the term is used is, for example, the situation in which the term is used, such as examination, medication, and treatment. The situation in which the term is used is not limited to the above. Furthermore, the classification of the term is not limited to the above.

[0038] The label may include information regarding the inclusion relationship between terms. For example, the information regarding the inclusion relationship between terms is information indicating the hierarchical relationship in terms of meaning. For example, given the terms "lung cancer" and "non-small cell lung cancer," "lung cancer" is a term that includes "non-small cell lung cancer." In this case, "lung cancer" is a higher-ranking term and a lower-ranking term than "non-small cell lung cancer." The label may also include information indicating a chronological order. For example, when adding a label to the sentence "After administering drug A, the administration was discontinued due to side effects, and drug B was administered," the generation unit 12 adds a label indicating that "drug A" and "administration" chronologically precede "drug B" and "administration."

[0039] The generation unit 12 may identify a chronological causal relationship between terms by performing a reverse lookup from a description of an outcome in the target text. Reverse lookup, for example, refers to identifying a relationship between an earlier action in the chronological order based on an action performed later in the chronological order. The generation unit 12 may also identify a chronological causal relationship based on the action indicated by a term and the causal relationship between the actions. The generation unit 12, for example, identifies a term related to an outcome in the target text. Then, the generation unit 12 traces back the text, starting from a term related to the outcome, and identifies a term that has a causal relationship with the term related to the outcome. The generation unit 12 traces back the text from the identified term, and identifies a term that has a causal relationship with the term that was the starting point. The relationship between the action and the chronological order is set, for example, as a rule. By repeating this process, the generation unit 12 extracts, for example, a series of terms that have a causal relationship. Then, the generation unit 12 identifies a chronological causal relationship based on the extraction result of a series of terms that have a causal relationship.

[0040] When the target text is information about treatment, the description of the outcome in the target text is, for example, a sentence or term indicating the treatment result. For example, in a clinical trial, the final stage of a series of treatment actions may be described as the treatment result. Therefore, the generation unit 12 can identify the chronological causal relationship between terms by tracing back the text based on the causal relationship between terms, starting from, for example, a term indicating the treatment result. For example, a description of medication written before a description of the treatment result is an action that occurred chronologically before the confirmation of the treatment result. Furthermore, for example, a description of an examination written before a description of medication is an action that occurred chronologically before the medication. For example, if the description of the outcome includes a description of the treatment result indicating the cure of the disease and further includes a description of the medication implementation indicating the administration of the drug, the generation unit 12 determines that the administration of the drug, which is the medication implementation, occurs chronologically before the cure of the disease, which is the treatment result. In this way, by identifying causal relationships over time, the accuracy of understanding sentences that include two or more actions, such as "After administering drug A, the growth of the tumor was suppressed," can be improved when understanding the meaning of a sentence that presents information about treatment, for example.

[0041] The generation unit 12 generates, as an intermediate representation text, text to which a label including information indicating a causal relationship in time series is added. That is, the generation unit 12 generates an intermediate representation of a target sentence based on, for example, the information indicating the causal relationship in time series. The generation unit 12, for example, adds a label including information indicating the causal relationship in time series to a term. The generation unit 12, for example, adds information indicating the causal relationship in time series to an already added label. The generation unit 12 may also replace an already added label with a label including information indicating the causal relationship in time series. The generation unit 12 may also add, for example, an additional label including information indicating the causal relationship in time series to an already added label. In this case, two or more labels with different roles may be added to one term. The generation unit 12, for example, adds a label including information indicating a stage of treatment to a term. The stage of treatment is, for example, a stage of treatment described in a sentence among several stages of treatment performed during a period from the start to the end of treatment. For example, the label may include information indicating each stage of diagnosis initiation, examination initiation, disease state confirmation, medication administration, or treatment result. When the label is information indicating a stage in treatment, the label may include information indicating a chronological causal relationship between terms. For example, terms labeled with diagnosis initiation, examination initiation, disease state confirmation, and medication administration are actions that occurred chronologically earlier than terms labeled with treatment result.

[0042] The generation unit 12 may generate the intermediate representation using a generative model. The generative model is, for example, a machine learning model that extracts terms from a sentence using natural language processing and assigns labels to the extracted terms. The generative model generates an intermediate representation of a target sentence by, for example, assigning labels containing information indicating a causal relationship in time series to terms included in the target sentence. The generative model is generated, for example, by learning the relationship between the sentence and a dictionary and the assigned labels. The generative model is generated, for example, by further learning the causal relationship in time series between terms or labels. The generative model may also be generated by further learning the inclusion relationship between terms or labels. The generative model may also be generated by further learning the causal relationship between terms or labels. The generative model is generated, for example, by deep learning using a neural network. The generative model is generated using, for example, Word2Vec. The algorithm used to train the generative model is not limited to the above. The generative model may also be generated using a trained large-scale language model. The large-scale language model may be, for example, GPT-2 (Generative Pre-trained Transformer-2), GPT-3, or GPT-4. Alternatively, the large-scale language model may be T5 (Text-to-Text Transfer Transformer), BERT (Bidirectional Encoder Representations from Transformers), RoBERTa (Robustly optimized BERT approach), or ELECTRA (Efficiently Learning an Encoder that Classifies Token Replacements Accurately). The generative model may be generated, for example, in a system external to the information processing device 10.

[0043] FIG. 5 shows an example of an intermediate representation for the sentence of the doctor's findings shown in the example of FIG. 4. The generation unit 12 extracts terms from the target document, for example, based on a dictionary. For example, the generation unit 12 extracts the term "suspected lung cancer" from the sentence "suspected lung cancer. Imaging diagnosis required." For example, the generation unit 12 extracts the terms "chest X-ray" and "lung cancer in the right lower lobe" from the sentence "chest X-ray confirmed lung cancer in the right lower lobe." For example, the generation unit 12 extracts the term "cisplatin administration" from the sentence "hospitalized. cisplatin administration." In addition, the generation unit 12 extracts the term "tumor disappearance."

[0044] For example, the generation unit 12 adds a label of "diagnosis started" to "suspected lung cancer." For example, the generation unit 12 adds a label of "examination performed" to "chest X-ray." Furthermore, the generation unit 12 adds a label of "condition confirmed" to "lung cancer in right lower lobe." Furthermore, the generation unit 12 adds a label of "medication performed" to "cisplatin administration." Furthermore, in the sentence "tumor disappears," the generation unit 12 adds a label of "treatment result" to "tumor disappears." Then, for example, the generation unit 12 generates text in which labels are added to each term included in the sentence, as the intermediate representation shown in the example of FIG. 5.

[0045] In the example of the intermediate representation of FIG. 5 , the generation unit 12 may identify the chronological relationships among "diagnosis start," "examination execution," "condition confirmation," "medication execution," and "treatment result" based on the added labels. When identifying the chronological relationships, the generation unit 12 generates, as an intermediate representation, labels added to each of the terms included in the sentence and information indicating the chronological relationships between the terms. When identifying the chronological relationships, the generation unit 12 may identify the period from "examination execution" to "treatment result" as the treatment period. The chronological relationship between the labels is set, for example, as a rule. When identifying the treatment period, the generation unit 12 generates an intermediate representation including information indicating the treatment period. In the example of the intermediate representation of FIG. 5 , the generation unit 12 may extract "tumor disappearance" as a description related to the outcome, and identify the chronological causal relationship of the terms included in the sentence by tracing back the document starting from "tumor disappearance."

[0046] The conversion unit 13 converts the target text into structured text based on the intermediate representation. Structuring, for example, refers to converting related information contained in a text into a state in which the relationships are clear. The structured text is, for example, data that associates items of the target text with the content of each item. Items indicate, for example, the classification of information to be structured. The classification of information indicates, for example, what the information is related to. For example, when extracting candidate clinical trial patients, "age," "disease name," "medication history," and "medication history" correspond to the items. Furthermore, the items may be hierarchical. For example, in the "medication history" item, the items "drug name," "medication period," and "dosage" may be associated below "medication history." The hierarchical structure may also have three or more levels. The items used when extracting candidate clinical trial patients are not limited to those described above. Furthermore, the conditions corresponding to the items correspond to the content of each item. For example, when extracting candidate clinical trial patients, when structuring a text that indicates the selection criterion "age 18 years or older," "age" corresponds to an item. Furthermore, in a sentence indicating a selection criterion such as "age must be 18 years or older," "18 years or older" corresponds to the content of an item. When structuring a sentence indicating selection criteria, a state in which the relationships are clear means a state in which the correspondence between the criteria and the conditions of each criterion is clear by associating each criterion included in the selection criteria, which are items, with the conditions of each criterion, which are the content of the item. For example, the conversion unit 13 converts the sentence indicating the selection criteria into a state in which the relationships are clear by associating "age" as an item with "18 years or older" as the content of the item. In this way, the conversion unit 13 converts, for example, a sentence indicating the selection criteria into a state in which the relationships between the items corresponding to each criterion included in the selection criteria and the conditions of each item are clear.

[0047] The conversion unit 13 converts a sentence into structured text, for example, based on labels attached to terms included in the sentence. The conversion unit 13 converts a sentence into structured text, for example, according to set rules. The rules, for example, define whether a label corresponds to an item or the content of the item. The rules may also define the format of the structured text. When the structured text is tabular data, the rules may define the table format and the relationship between the position in the table and the label. For example, the conversion unit 13 identifies, among terms included in the sentence, terms labeled with an item-related label and terms in the sentence that are semantically related to the term. For example, when extracting candidate clinical trial patients, a term semantically related to "age" is "18 years or older." The conversion unit 13, for example, extracts, as items, terms labeled with an item-related label based on the results of the term identification. The conversion unit 13 may also extract, as items, phrases including terms labeled with an item-related label. The conversion unit 13 also extracts, for example, terms labeled with a label relating to the content of an item as the content of the item. Then, the conversion unit 13 converts the target sentence into structured text by, for example, associating terms labeled with a label relating to the item with terms labeled with the content of the item.

[0048] The conversion unit 13 may convert a sentence into structured text using a structured model. The structured model is, for example, a machine learning model that takes an intermediate representation of a target sentence as input and converts the sentence into structured text. The structured model is generated, for example, by learning the relationship between the intermediate representation and the structured text. For example, the structured model is generated by learning the relationship between a sentence labeled with terms included therein and the structured text.

[0049] The structured model may take an intermediate representation of a target sentence and a structured sentence format as input and convert the sentence into structured text. For example, when converting selection criteria into structured text, the structured model is generated by learning the relationship between the intermediate representation of the sentence indicating the selection criteria and the structured selection criteria format and the structured selection criteria. For example, the structured model may take an intermediate representation of the sentence indicating the selection criteria as input and output structured selection criteria. In this case, the structured model is generated by learning the relationship between the intermediate representation of the sentence indicating the selection criteria and the structured selection criteria. The structured model is generated by deep learning using a neural network, for example. The structured model is generated in a system external to the information processing device 10, for example.

[0050] The selection criteria format is information that defines how to divide the selection criteria into items when structuring the selection criteria. The selection criteria format is set, for example, for each disease to which the drug being investigated is administered. In clinical trials, for example, evaluation items are often defined for each disease. For this reason, the structured model can, for example, use the items to be extracted as selection criteria as input along with the text indicating the selection criteria, and extract the conditions corresponding to the items from the text indicating the selection criteria.

[0051] The structured model may be a large-scale language model. For example, the conversion unit 13 converts the target document into structured text based on an intermediate representation of the target sentence and the format of the structured text. The format of the structured text is, for example, a label attached to a term extracted as an item, a label attached to a term associated with the item as the content of the item, and information specifying an output format. The information specifying the format of the structured text is also called a prompt.

[0052] The structured text may be an artificial language, such as a computer program. A computer program may contain commands that cause a computer to perform an operation. The structured text may also be prompts used as input for a large-scale model.

[0053] The large-scale language model may be, for example, GPT-2, GPT-3, or GPT-4. Alternatively, the large-scale language model may be, for example, T5, BERT, RoBERTa, or ELECTRA. The structured model is generated, for example, in a system external to the information processing device 10. The process of generating structured selection criteria using the structured model may be performed in a system external to the information processing device 10.

[0054] FIG. 6 shows an example of a sentence indicating criteria for inclusion of a patient in a clinical trial, among the selection criteria. In the example of the inclusion criteria for a clinical trial shown in FIG. 6, the inclusion criteria include a plurality of criteria. The generation unit 12 generates an intermediate representation from the sentence for each criterion included in the inclusion criteria, for example. Then, the conversion unit 13 converts the sentence of the example inclusion criteria shown in FIG. 6 into structured text, for example, based on the intermediate representation.

[0055] FIG. 7 shows an example of structured text of the inclusion criteria. FIG. 7 is an example of structured text of the sentence showing the inclusion criteria shown in FIG. 6. In the example of structured text in FIG. 7, the items of the inclusion criteria are associated with the conditions for each item. In the example of structured inclusion criteria in FIG. 7, the sentence showing the inclusion criteria is structured using items such as "age," "diagnosis," "lesion size," "number of lesions," and "white blood cell count," which correspond to the criteria included in the sentence showing the inclusion criteria. For example, the sentence "Must be 20 years of age or older at the time of screening" in the example of structured inclusion criteria in FIG. 7 is converted into structured text by using "age" as an item and associating the condition "20 years of age or older" with that item.

[0056] FIG. 8 shows an example of a sentence indicating a patient exclusion criteria from a clinical trial, which is one of the selection criteria. In the example of the exclusion criteria from a clinical trial shown in FIG. 8, the exclusion criteria include multiple criteria. The generation unit 12, for example, generates an intermediate representation from the sentence for each criterion included in the exclusion criteria. Then, the conversion unit 13, for example, converts the sentence of the example exclusion criteria shown in FIG. 8 into structured text based on the intermediate representation.

[0057] FIG. 9 shows an example of structured text of the exclusion criteria. FIG. 9 is an example of structured text of the exclusion criteria shown in FIG. 8. In the example of structured text in FIG. 9, the items of the exclusion criteria are associated with the conditions for each item. In the example of structured exclusion criteria in FIG. 9, the sentence indicating the exclusion criteria is structured using items such as "history of current illness," "medical equipment," and "condition," which correspond to the criteria included in the sentence indicating the exclusion criteria. For example, the sentence "has or has a history of interstitial pneumonia" in the example of sentence indicating the exclusion criteria in FIG. 8 is converted into structured text by using "history of current illness" as an item and associating the condition "interstitial pneumonia" with that item in the example of structured exclusion criteria in FIG. 9.

[0058] Fig. 10 shows an example of structured text of information about treatment. In the example shown in Fig. 10, for example, each item of information about treatment is associated with information recorded in an electronic medical record. In the example of structured text of information about treatment in Fig. 10, for example, "gender" is used as an item, and the description "male" is associated with this item to convert it into structured text. Furthermore, the example of structured text of information about treatment in Fig. 10 is used, for example, in a process of extracting whether or not information about a patient's treatment complies with the selection criteria for a clinical trial.

[0059] The extraction unit 14 extracts information to be extracted based on, for example, structured text. When extracting candidate clinical trial patients, the extraction unit 14 extracts, for example, information indicating whether or not information about the treatment of each patient conforms to the selection criteria for the clinical trial as the information to be extracted. When extracting candidate clinical trial patients, the extraction unit 14 may extract, for example, information that a person in charge needs to confirm in order to determine whether or not the patient conforms to the clinical trial as the information to be extracted. The extraction unit 14 extracts the information to be extracted using, for example, an extraction model. The extraction model is, for example, a machine learning model that uses structured text as input and extracts the information to be extracted. The extraction unit 14 may use the extraction model to extract information indicating whether or not information matches between two or more types of structured text. In this case, the extraction model is, for example, a machine learning model that uses two or more types of structured text as input and extracts information indicating whether or not information matches. For example, when extracting candidate clinical trial patients, the extraction unit 14 extracts information indicating whether or not information about the treatment conforms to the selection criteria. The extraction unit 14, for example, receives structured selection criteria and structured information on treatment as input and extracts information indicating whether the information on treatment conforms to the selection criteria. A match may include a match that can be considered to occur in consideration of an inclusion relationship. For example, if the structured selection criteria require "lung cancer" and the information on treatment includes a description of "non-small cell lung cancer," which is a more detailed description of "lung cancer," the extraction unit 14 determines that the selection criteria and the information on treatment conform.

[0060] The extraction unit 14, for example, compares text in which the selection criteria are structured with text in which information about treatment is structured. Then, the extraction unit 14 extracts information indicating whether the information about treatment matches the items included in the selection criteria. For example, the extraction unit 14 extracts information indicating whether the information about treatment matches the selection criteria. The extraction unit 14 may determine that information corresponding to the selection criteria is not included in the information about treatment. For example, when the extraction unit 14 determines that information corresponding to the selection criteria is not included in the information about treatment, it outputs information indicating that the corresponding information is not described.

[0061] The extraction model is generated, for example, by learning the relationship between structured text and information to be extracted. The extraction model is also generated by learning the relationship between two or more types of structured text and whether or not information between the texts matches. The extraction model is generated, for example, by converting the structured text and the information to be extracted into feature vectors and learning the relationship between the feature vectors. The extraction model may also be generated by converting two types of structured text into feature vectors and learning the relationship between the feature vectors. The feature vector extraction model may also be generated by deep learning using a neural network. The learning algorithm for extracting the extraction model is not limited to the above. The extraction model may also be generated, for example, in a system external to the information processing device 10.

[0062] The extraction unit 14 may calculate a score indicating the reliability of the extracted information. For example, the extraction unit 14 calculates a score indicating the reliability of the estimation by the extraction model. For example, when extracting information indicating whether or not information related to treatment conforms to selection criteria, the extraction unit 14 sets the score to the reliability of the determination of whether or not the conditions indicated by the selection criteria match the information related to treatment. The reliability of the estimation by the extraction model may be, for example, the probability that the structured text matches the information to be extracted. The reliability of the estimation by the extraction model may be, for example, the probability that two types of structured text match. For example, when extracting information indicating whether or not information related to treatment conforms to selection criteria, the extraction unit 14 outputs the probability that the conditions indicated by the selection criteria match the information related to treatment as a score. The reliability of the estimation by the extraction model may be calculated based on the distance between feature vectors obtained by converting two pieces of information. Furthermore, when one item includes multiple conditions, the extraction unit 14 may calculate the percentage of matching conditions as a score.

[0063] The extraction unit 14 may use a large-scale language model to extract information to be extracted, using the structured text as a prompt. The information to be extracted is, for example, information included in information about treatment. For example, when selecting patients to be included in a clinical trial, the extraction unit 14 extracts information indicating whether or not the selection criteria are met from the information about treatment based on the structured text. The information about treatment is, for example, records of the patient's medical care recorded in an electronic medical record. The information to be extracted is not limited to information included in the information about treatment.

[0064] The extraction unit 14 may use the intermediate representation to extract information referenced by the user from the text to be structured. For example, in the medical field, the user may be a person in charge of extracting candidate clinical trial patients, a doctor, a nurse, a physical therapist, a psychotherapist, or a nutritionist. The user is not limited to the above. For example, the extraction unit 14 uses the intermediate representation to generate a summary of the text to be structured as information to be referenced by the user. For example, the extraction unit 14 inputs a summary instruction and the intermediate representation to the large-scale language model. Then, the extraction unit 14 regards the summary generated by the large-scale language model as information extracted from the text to be structured.

[0065] Fig. 11 shows an example of a summary extracted from a target text by the extraction unit 14. The example summary in Fig. 11 is extracted using structured text converted using the example intermediate representation shown in Fig. 5. By referring to the example summary in Fig. 11, a person looking at the summary can easily understand the name of the patient's illness and the progress of treatment.

[0066] The output unit 15, for example, outputs the information extracted by the extraction unit 14. When candidate clinical trial patients are extracted, the output unit 15, for example, outputs information extracted from information related to treatment by the extraction unit 14 using structured text. The output unit 15 may also output information indicating whether or not the information related to treatment complies with the selection criteria. The output unit 15 outputs the information extracted by the extraction unit 14 to, for example, a terminal device 20.

[0067] When the extraction unit 14 calculates a score indicating the likelihood of the extracted information, the output unit 15 outputs, for example, the score calculated by the extraction unit 14. The output unit 15 may further output a score indicating the likelihood of conversion into structured text by the conversion unit 13. The score indicating the likelihood of conversion into structured text is calculated by the conversion unit 13, for example, based on the degree of match between terms included in the structured text and terms included in a dictionary.

[0068] The output unit 15 may output the structured text. For example, the output unit 15 outputs the structured text converted by the conversion unit 13. The output unit 15 may also output intermediate representation text together with the structured text. Furthermore, when the extraction unit 14 extracts a summary of the information to be extracted, the output unit 15 may output the extracted summary.

[0069] FIG. 12 shows an example of extraction results indicating whether a patient meets the selection criteria. In the example of the extraction results in FIG. 12, the items included in the selection criteria, the criteria for each item, and the extraction results by the extraction unit 14 are associated as matching results. The matching results indicate whether the selection criteria match the information about the patient's treatment. Items with a matching result of "○" indicate that the criteria set forth in the selection criteria match the information about the patient's treatment. Items with a matching result of "×" indicate that the criteria set forth in the selection criteria do not match the information about the patient's treatment. Furthermore, items with a matching result of "not stated" indicate that information corresponding to the criteria set forth in the selection criteria is not included in the information about the patient's treatment. A person in charge of extracting candidate clinical trial patients can determine whether each patient is suitable as a candidate clinical trial patient by, for example, referring to the matching results shown in FIG. 12.

[0070] The memory unit 16 stores, for example, data related to the process of converting a sentence into structured text. The memory unit 16 stores, for example, a sentence to be structured. The memory unit 16 also stores, for example, an intermediate representation of the sentence to be structured. The memory unit 16 also stores, for example, a text obtained by structuring a target document. The memory unit 16 also stores, for example, a dictionary used for extracting terms from a sentence and adding labels. The memory unit 16 also stores, for example, data related to the process of extracting information using structured text from a sentence. The memory unit 16 stores, for example, a generative model. The memory unit 16 also stores, for example, a structured model. The memory unit 16 also stores, for example, an extraction model. The generative model, the structured model, and the extraction model may each be stored in a storage means other than the memory unit 16.

[0071] The terminal device 20 is, for example, a terminal device used by a person who extracts information. The terminal device 20 acquires a target sentence to be converted into structured text. The target sentence is, for example, input to the terminal device 20 by an operation of the person who extracts information. The terminal device 20 outputs the target document to, for example, the acquisition unit 11 of the information processing device 10.

[0072] The terminal device 20 acquires the information extraction result from, for example, the output unit 15 of the information processing device 10. Then, the terminal device 20 outputs the information extraction result to a display device (not shown). Furthermore, the terminal device 20 acquires structured text converted from the target document from, for example, the output unit 15 of the information processing device 10. Then, the terminal device 20 outputs the structured text converted from the target document to a display device (not shown).

[0073] When extracting patients who meet the selection criteria for clinical trial patients, the terminal device 20 is, for example, a terminal device used by a person who extracts candidate clinical trial patients. The terminal device 20, for example, acquires the selection criteria for clinical trial patients. The selection criteria for clinical trial patients are input into the terminal device 20, for example, by the person who extracts candidate clinical trial patients. The terminal device 20 then outputs the selection criteria for clinical trial patients to the acquisition unit 11 of the information processing device 10.

[0074] The terminal device 20 acquires a matching result regarding conformance with the selection criteria from, for example, the output unit 15 of the information processing device 10. Then, the terminal device 20 outputs the matching result to a display device (not shown). The terminal device 20 may also acquire structured selection criteria from, for example, the output unit 15 of the information processing device 10. The terminal device 20 outputs the structured selection criteria to a display device (not shown).

[0075] The data management device 30 stores information to be extracted. The data management device 30 stores, for example, information related to the patient's treatment. The information related to the patient's treatment stores, for example, electronic medical record data. The information related to the treatment may be test data. The information related to the treatment may also be medical accounting data or medical receipts. The data management device 30 outputs the information related to the patient's treatment to, for example, the acquisition unit 11 of the information processing device 10.

[0076] The process of converting a target sentence into structured text will now be described with reference to Fig. 13, which is a diagram showing an example of the operational flow of the process of converting a target sentence into structured text.

[0077] The acquisition unit 11 acquires a sentence to be structured (step S11). The acquisition unit 11 acquires the sentence to be structured from the terminal device 20, for example.

[0078] When the sentence to be structured is acquired, the generation unit 12 generates an intermediate representation to be used for structuring based on the relationships between terms contained in the sentence (step S12).

[0079] When intermediate representations have been generated from all of the target sentences (Yes in step S13), the conversion unit 13 converts the target sentences to be structurized into structured text based on the intermediate representations (step S14).

[0080] Once converted into structured text, the output unit 15 outputs the structured text (step S15).

[0081] In step S13, if there is a sentence for which intermediate information has not been generated (No in step S13), the process returns to step S12, and the generation unit 12 generates an intermediate representation to be used for structuring based on the relationships between terms contained in the sentence.

[0082] This section describes a process for extracting information indicating whether or not information related to treatment conforms to the selection criteria when selecting candidate clinical trial patients. Figure 14 is a diagram showing an example of the operational flow of the process for extracting information indicating whether or not information related to treatment conforms to the selection criteria.

[0083] The acquisition unit 11 acquires a sentence indicating the selection criteria for clinical trial patients (step S21). The acquisition unit 11 acquires the sentence indicating the selection criteria for clinical trial patients, for example, from the terminal device 20. The sentence indicating the selection criteria for clinical trial patients is input into the terminal device 20, for example, by a person in charge of extracting candidate clinical trial patients.

[0084] The acquiring unit 11 also acquires information about the treatment of each patient (step S22). The acquiring unit 11 acquires information about the treatment of each patient from the data management device 30, for example.

[0085] When the sentence indicating the selection criteria and the information on the treatment are acquired, the generation unit 12 generates an intermediate representation to be used for structuring based on the relationship between terms included in the sentence (step S23). The generation unit 12 generates an intermediate representation for each of the sentence indicating the selection criteria and the information on the treatment.

[0086] Once an intermediate representation is generated from the sentence indicating the selection criteria and all information regarding the treatment (Yes in step S24), the conversion unit 13 converts the sentence indicating the selection criteria and information regarding the treatment into structured text based on the intermediate representation (step S25).

[0087] Once the sentence indicating the selection criteria and the information regarding treatment have been converted into structured text, the extraction unit 14 extracts, for example, information indicating whether the information regarding treatment complies with each of the criteria included in the selection criteria based on the structured text (step S26).

[0088] Once information indicating whether the treatment information conforms to the selection criteria for all target patients is extracted (Yes in step S27), the output unit 15 outputs information indicating whether the treatment information conforms to each of the criteria included in the selection criteria (step S28).

[0089] In step S24, if there is a sentence for which intermediate information has not been generated (No in step S24), the process returns to step S23, and the generation unit 12 generates an intermediate representation to be used for structuring based on the relationship between terms contained in the sentence.

[0090] Also, in step S27, if there is a patient for whom information extraction has not been completed (No in step S27), the process returns to step S26, and the extraction unit 14 extracts information indicating whether or not the information regarding treatment complies with each of the criteria included in the selection criteria, for example, based on structured text.

[0091] The information processing device 10 generates an intermediate representation to be used for structuring from the sentence to be structured based on the relationships between terms included in the sentence to be structured.The information processing device 10 then converts the sentence to be structured into structured text based on the intermediate representation.By converting into structured text based on the intermediate representation generated based on the relationships between terms, the information processing device 10 can improve the accuracy of conversion into structured text.

[0092] Furthermore, when extracting patients who meet the selection criteria for clinical trial patients, the information processing device 10 generates an intermediate representation from a sentence indicating the selection criteria. Then, the information processing device 10 converts the sentence indicating the selection criteria into structured text using the intermediate representation. The information processing device 10 also generates an intermediate representation from information about the treatment of each patient. Then, the information processing device 10 converts the information about the treatment of each patient into structured text using the intermediate representation. By generating structured text of the selection criteria and the information about the treatment based on the intermediate representation in this manner, the information processing device 10 can, for example, accurately convert the selection criteria and the information about the treatment into structured text. Furthermore, the information processing device 10 extracts, for each patient, information indicating whether the information about the treatment meets the selection criteria, for example, based on the structured text of the selection criteria and the structured text of the information about the treatment. For example, when determining whether the information about the treatment meets the selection criteria, the information processing device 10 can improve the accuracy of determining whether the information meets the selection criteria by comparing the accurately converted structured texts. Therefore, by using the information processing device 10, the accuracy of extracting information indicating whether the information about the treatment meets the selection criteria can be improved.

[0093] Furthermore, by generating structured text based on the intermediate representation generated from the target text, the target text can be converted into structured text that is optimized for the target text. Furthermore, by referencing information on patients who meet the selection criteria for clinical trials extracted based on the structured text, a person selecting patients who will be candidates for clinical trials can make appropriate decisions in selecting patients for clinical trials.

[0094] Each process in the information processing device 10 may be distributed and executed among multiple information processing devices connected via a network. For example, the process of converting a target sentence into structured text and the process of extracting information using the structured text may be executed by different information processing devices. It may be appropriately set which information processing device executes each process in the information processing device 10.

[0095] Each process in the information processing device 10 can be realized by executing a computer program on the computer. Fig. 15 shows an example of the configuration of a computer 100 that executes a computer program that performs each process in the information processing device 10. The computer 100 includes a CPU (Central Processing Unit) 101, a memory 102, a storage device 103, an input / output I / F (Interface) 104, and a communication I / F 105.

[0096] The CPU 101 reads and executes computer programs for performing each process from the storage device 103. The CPU 101 may be configured as a combination of multiple CPUs. The CPU 101 may also be configured as a combination of a CPU and another type of processor. For example, the CPU 101 may be configured as a combination of a CPU and a graphics processing unit (GPU). The memory 102 is configured with a dynamic random access memory (DRAM) or the like, and temporarily stores computer programs executed by the CPU 101 and data being processed. The storage device 103 stores computer programs executed by the CPU 101. The storage device 103 is configured with, for example, a non-volatile semiconductor storage device. Other storage devices such as a hard disk drive may also be used for the storage device 103. The input / output I / F 104 is an interface that accepts input from an operator and outputs display data, etc. The communication I / F 105 is an interface that transmits and receives data between the terminal device 20, the data management device 30, and other information processing devices. Furthermore, the terminal device 20 and the data management device 30 may also have the same configuration as the computer 100 .

[0097] The computer program used to execute each process can also be stored and distributed on a computer-readable recording medium that non-temporarily stores data. Examples of recording media that can be used include magnetic tapes for recording data and magnetic disks such as hard disks. Optical disks such as CD-ROMs (Compact Disc Read Only Memory) can also be used as recording media. Non-volatile semiconductor storage devices can also be used as recording media.

[0098] Some or all of the above-described embodiments can be described as, but are not limited to, the following supplementary notes.

[0099] [Supplementary Note 1] An information processing device comprising: an acquisition means for acquiring a sentence to be structured; a generation means for generating an intermediate representation to be used for structuring the sentence based on the relationship between terms included in the sentence; and a conversion means for converting the sentence into structured text based on the intermediate representation.

[0100] [Supplementary Note 2] The information processing device according to Supplementary Note 1, wherein the generating means generates the intermediate representation based on an inclusion relationship between terms included in the sentence.

[0101] [Supplementary Note 3] The information processing device according to Supplementary Note 1 or 2, wherein the generating means converts the sentence into the intermediate representation based on a time-series causal relationship between terms included in the sentence.

[0102] [Supplementary Note 4] The information processing device according to Supplementary Note 3, wherein the generating means identifies a chronological causal relationship between terms by performing a reverse lookup from a description of the result in the sentence.

[0103] [Supplementary Note 5] The information processing device according to Supplementary Note 4, wherein the generating means identifies a period during which the treatment was performed based on a description of the results in the text.

[0104] [Supplementary Note 6] The information processing device according to any one of Supplementary Notes 1 to 5, wherein the intermediate representation is text in which labels indicating relationships between terms included in the sentence are added to the sentence.

[0105] [Supplementary Note 7] The information processing device according to any one of Supplementary Notes 1 to 6, further comprising: an extraction unit that extracts the information from a sentence that is a target for information extraction based on the structured text; and an output unit that outputs the extracted information.

[0106] [Supplementary Note 8] The information processing device according to Supplementary Note 7, wherein the structured text is a prompt used for input to a large-scale language model.

[0107] [Appendix 9] The information processing device described in Appendix 7, wherein the text to be structured is selection criteria for patients to be clinical trial subjects and information regarding the treatment of each patient, and the extraction means extracts information indicating whether or not each patient complies with the selection criteria based on a text in which the selection criteria have been structured and a text in which the information regarding the treatment of each patient has been structured.

[0108] [Supplementary Note 10] The information processing device according to Supplementary Note 9, wherein the information indicating conformance to the selection criteria further includes a score indicating likelihood of extraction of the information indicating conformance.

[0109] [Supplementary Note 11] The information processing device according to Supplementary Note 9 or 10, wherein the selection criteria include at least one of a condition for inclusion in a clinical trial and a condition for exclusion from a clinical trial.

[0110] [Supplementary Note 12] The information processing device according to any one of Supplementary Notes 1 to 11, wherein the conversion means converts the sentence into the structured text using a structured model generated by machine learning.

[0111] [Supplementary Note 13] An information processing method comprising: acquiring a sentence to be structured; generating an intermediate representation to be used for structuring the sentence based on relationships between terms contained in the sentence; and converting the sentence into structured text based on the intermediate representation.

[0112] [Supplementary Note 14] A recording medium that non-temporarily records a program that causes a computer to execute the following processes: a process of acquiring a sentence to be structured; a process of generating an intermediate representation to be used for structuring the sentence based on the relationship between terms included in the sentence; and a process of converting the sentence into structured text based on the intermediate representation.

[0113] Furthermore, some or all of the configurations described in Supplementary Notes 2 to 12, which are dependent on Supplementary Note 1, may also be dependent on Supplementary Notes 13 and 14 in the same dependent relationship as Supplementary Notes 2 to 12. Furthermore, not limited to Supplementary Notes 1, 13, and 14, some or all of the configurations described as Supplements may be made dependent on various hardware, software, various recording means for recording software, or systems, within the scope of each of the above-mentioned embodiments.

[0114] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each embodiment can be combined with other embodiments as appropriate.

[0115] REFERENCE SIGNS LIST 10 Information processing device 11 Acquisition unit 12 Generation unit 13 Conversion unit 14 Extraction unit 15 Output unit 16 Storage unit 20 Terminal device 30 Data management device 100 Computer 101 CPU 102 Memory 103 Storage device 104 Input / output I / F 105 Communication I / F

Claims

1. An information processing apparatus comprising: an acquisition means for acquiring a text to be structured; a generation means for generating an intermediate representation used for structuring the text based on the relationships between terms included in the text; and a conversion means for converting the text into a structured text based on the intermediate representation.

2. The information processing apparatus according to claim 1, wherein the generation means generates the intermediate representation based on the inclusion relationship between terms included in the text.

3. The information processing apparatus according to claim 1 or 2, wherein the generation means converts the text into the intermediate representation based on the temporal causal relationship between terms included in the text.

4. The information processing apparatus according to claim 3, wherein the generation means identifies the temporal causal relationship between terms by backtracking from the description regarding the results in the text.

5. The information processing apparatus according to claim 4, wherein the generation means identifies the period during which the treatment was performed based on the description regarding the results in the text.

6. The information processing apparatus according to any one of claims 1 to 5, wherein the intermediate representation is a text obtained by adding labels indicating the relationships between terms included in the text to the text.

7. The information processing apparatus according to any one of claims 1 to 6, further comprising: an extraction means for extracting the information from the text to be the extraction target of the information based on the structured text; and an output means for outputting the extracted information.

8. The information processing apparatus according to claim 7, wherein the structured text is a prompt used for input to a large language model.

9. The text to be structured is the selection criteria for patients in a clinical trial and the information regarding the treatment of each patient. The extraction means extracts information indicating whether each patient meets the selection criteria based on the text obtained by structuring the selection criteria and the text obtained by structuring the information regarding the treatment of each patient. The information processing apparatus according to claim 7.

10. The information indicating whether each patient meets the selection criteria further includes a score indicating the likelihood of the extraction of the information indicating whether each patient meets the selection criteria. The information processing apparatus according to claim 9.

11. The information processing apparatus according to claim 9 or 10, wherein the selection criteria include at least one of inclusion conditions for inclusion in a clinical trial and exclusion conditions from a clinical trial.

12. The information processing apparatus according to any one of claims 1 to 11, wherein the conversion means uses a structured model generated by machine learning to convert the text into the structured text.

13. An information processing method, comprising: obtaining a text to be structured; generating an intermediate representation for use in structuring the text based on relationships between terms included in the text; and converting the text into structured text based on the intermediate representation.

14. The information processing method according to claim 13, wherein the intermediate representation is generated based on inclusion relationships between terms included in the text.

15. The information processing method according to claim 13 or 14, wherein the text is converted into the intermediate representation based on a temporal causal relationship between terms included in the text.

16. The information processing method according to claim 15, wherein a temporal causal relationship between terms is identified by backtracking from a description of results in the text.

17. The information processing method according to claim 16, wherein a period during which a treatment was performed is identified based on a description of results in the text.

18. The information processing method according to any one of claims 13 to 17, wherein the intermediate representation is text obtained by adding labels indicating relationships between terms included in the text to the text.

19. The information processing method according to any one of claims 13 to 18, comprising: extracting the information from a text to be an extraction target based on the structured text; and outputting the extracted information.

20. A non-transitory recording medium that non-transitorily records a program for causing a computer to execute a process of obtaining a text to be structured, a process of generating an intermediate representation for use in structuring the text based on relationships between terms included in the text, and a process of converting the text into structured text based on the intermediate representation.

Images