A medical data collection method based on a medical question-answering model
Through the data collection method based on the medical Q&A big model, the problem of inefficient data collection between multiple management tools is solved, and the intelligent Q&A rapid data acquisition is achieved, which improves the data collection efficiency.
Patent Information
- Application Number
- CN202411163486.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-23
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-08-23
AI Technical Summary
When collecting medical data, hospital managers and medical staff need to select appropriate tools from multiple different management tools for data collection. When they are not familiar with the operation, they need to spend a lot of time familiarizing themselves with the operations of each tool, resulting in inefficient data collection.
Using the data collection method based on the medical question and answer model, by obtaining the text of the question to be queried, first look up the Q&A results of whether there are similar questions from the historical question and answer library. If it does not exist, determine the query method and call the medical question and answer model. Obtain Q&A results from knowledge base or tool system to realize intelligent Q&A data collection.
Without manually operating multiple management tools, users can quickly and quickly obtain the required medical data, improving data collection efficiency and reducing the need for familiarity with the operation of each tool.
Smart Images

Figure CN118689967B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a medical data collection method based on a medical question-and-answer big model. Background Art
[0002] With the development of network technology, many hospitals now use a variety of medical management tools, such as medical information systems, electronic medical record management systems, medical quality control systems, telemedicine management systems, etc. The current medical management tools have the following characteristics:
[0003] (1) Medical data closure: Because hospital data involves patients’ personal information and medical condition information, medical management tools are usually closed and cannot be connected to external networks in order to protect patient privacy.
[0004] (2) Diversification of medical management tools: As hospitals pay more attention to operational management, a large number of hospital quality management tools have emerged in the medical field, such as performance management, operational management, single disease management, cost performance allocation and other tools, which help hospital managers conduct relevant statistical analysis;
[0005] (3) Diversification of hospital management indicators: Due to the complexity of medical care, there are thousands of medical management indicators in hospital management. Each indicator has different application scenarios and has its own threshold and range.
[0006] Therefore, when current users (referring to internal hospital personnel, such as hospital managers or medical staff) collect medical data, they may need to select appropriate tools from a large number of management tools for data collection. Sometimes, multiple tools may also need to be involved in joint queries. This requires users to be extremely familiar with the operation of each management tool in order to quickly complete the collection. If the operation is not familiar, it will take a lot of time to familiarize themselves with the operation of each management tool, which is extremely time-consuming. Summary of the invention
[0007] The purpose of the present invention is to provide a medical data collection method based on a medical question-answering big model to improve the problems existing in the prior art.
[0008] The embodiments of the present invention can be implemented as follows:
[0009] In a first aspect, the present invention provides a medical data collection method based on a medical question-answering big model, comprising:
[0010] Get the question text of the question to be queried:
[0011] Based on the question text, searching the historical question and answer database for a question and answer result of the question to be queried;
[0012] If it does not exist, the query method of the question to be queried is determined based on the question text, and the medical question and answer big model is called based on the query method to obtain the question and answer result of the question to be queried, and the query method is used to indicate whether the question and answer result is obtained based on a pre-built medical knowledge base or by querying the question and answer result from a medical tool system.
[0013] Optionally, the historical question and answer library stores historical question vectors and historical question and answer results of multiple historical question texts;
[0014] The step of searching, based on the question text, whether there is a question and answer result for the question to be queried from the historical question and answer library comprises:
[0015] Based on the question text, selecting a target similarity calculation method from a plurality of preset similarity calculation methods;
[0016] Convert the question text into a question vector;
[0017] Calculating the similarity between the question vector and each of the historical question vectors using the target similarity calculation method;
[0018] If there is no historical question vector whose similarity is greater than the first set threshold, it is determined that there is no question and answer result of the question to be queried in the historical question and answer database;
[0019] If there is at least one historical question vector whose similarity is greater than the first set threshold, the historical question and answer result corresponding to the historical question vector with the greatest similarity is used as the question and answer result of the question to be queried.
[0020] Optionally, the plurality of preset similarity calculation methods include a Jaccard similarity algorithm and a Euclidean distance algorithm;
[0021] The step of selecting a target similarity calculation method from a plurality of preset similarity calculation methods based on the question text includes:
[0022] If the question text does not contain a date and a medical code string, the Jaccard similarity algorithm is used as the target similarity calculation method;
[0023] If the question text contains a date and / or a medical code character string, the Euclidean distance algorithm is used as the target similarity calculation method.
[0024] Optionally, the step of determining a query method for the question to be queried based on the question text, and invoking the medical question-and-answer big model based on the query method to obtain a question-and-answer result for the question to be queried includes:
[0025] Obtaining a question vector corresponding to the question text;
[0026] Inputting the question vector into a pre-trained classification model to determine the query mode;
[0027] If the query method is a knowledge base query, the medical question and answer big model is called to search the medical knowledge base for the question and answer result of the question to be queried;
[0028] If the query mode is a tool query, the medical question-answering big model is called to perform intent analysis on the question text to obtain standardized format data;
[0029] sending the standardized format data to a data query tool so that the data query tool generates a query statement based on the standardized format data to obtain query response data from a medical tool system;
[0030] The question and answer result table returned by the data query tool is used as the question and answer result of the question to be queried; the question and answer result table is obtained by the data query tool through conversion based on the query response data.
[0031] Optionally, the step of calling the medical question-answering big model to perform intent analysis on the question text to obtain standardized format data includes:
[0032] Calling the medical question-answering big model to perform intent analysis on the question text to determine the query object, query time, and indicator data;
[0033] If there is a specified quantitative indicator in the indicator data, each specified quantitative indicator in the indicator data is used as an indicator to be queried;
[0034] If there is an abstract indicator in the indicator data, multiple quantitative indicators corresponding to the abstract indicator are determined from an indicator mapping library to obtain multiple indicators to be queried; the indicator mapping library includes multiple preset abstract indicators and multiple quantitative indicators corresponding to each of the preset abstract indicators;
[0035] Based on the query object, query time and each indicator to be queried, standardized format data is generated.
[0036] Optionally, the medical knowledge base includes a plurality of content vectors corresponding to each of a plurality of medical documents;
[0037] The step of calling the medical question-answering big model to search the medical knowledge base for the question-answering result of the question to be queried comprises:
[0038] Calculating the similarity between the question vector and each content vector in the medical knowledge base, and taking all content vectors whose similarity exceeds a second set threshold as target content vectors;
[0039] Obtaining target medical documents corresponding to each target content vector;
[0040] The medical question-answering big model is called to generate the question-answering result of the question to be queried based on all target medical documents and the question text.
[0041] Optionally, the construction process of the medical knowledge base includes:
[0042] Access to multiple medical literature;
[0043] For any of the medical documents, segment each paragraph of the medical document according to a preset segmentation length to obtain a plurality of text segments;
[0044] For text segments whose length is shorter than the preset segmentation length, padding is performed;
[0045] Convert each of the text segments into a content vector to obtain a plurality of content vectors corresponding to the medical document;
[0046] Each of the medical documents is traversed to obtain a plurality of content vectors corresponding to each of the medical documents, thereby completing the construction of the medical knowledge base.
[0047] In a second aspect, the present invention provides a medical data collection device based on a medical question-answering big model, comprising:
[0048] The acquisition module is used to obtain the question text of the question to be queried:
[0049] Query module for:
[0050] Based on the question text, searching the historical question and answer database for a question and answer result of the question to be queried;
[0051] If it does not exist, the query method of the question to be queried is determined based on the question text, and the medical question and answer big model is called based on the query method to obtain the question and answer result of the question to be queried, and the query method is used to indicate whether the question and answer result is obtained based on a pre-built medical knowledge base or by querying the question and answer result from a medical tool system.
[0052] In a third aspect, the present invention provides an electronic device comprising: a memory and a processor, wherein the memory stores a software program, and when the electronic device is running, the processor executes the software program to implement the medical data collection method based on the medical question and answer big model as described in the first aspect above.
[0053] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the medical data collection method based on the medical question-and-answer big model described in the first aspect.
[0054] Compared with the prior art, the embodiment of the present invention provides a medical data collection method based on a medical question-and-answer big model, which first obtains the question text of the question to be queried, and then searches the historical question-and-answer library based on the question text to see whether there is a question-and-answer result for the question to be queried; if not, the query method of the question to be queried is determined based on the question text, and the medical question-and-answer big model is called based on the query method to obtain the question-and-answer result of the question to be queried. The query method is used to indicate whether the question-and-answer result is obtained based on a pre-built medical knowledge base or the question-and-answer result is queried from a medical tool system. In this way, users can collect medical data through intelligent question-and-answer methods without manual work, which is fast and quick. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0056] Figure 1 One of the flow charts of a medical data collection method based on a medical question and answer big model provided in an embodiment of the present invention.
[0057] Figure 2 The second flowchart of a medical data collection method based on a medical question and answer big model provided in an embodiment of the present invention.
[0058] Figure 3 The third flowchart of a medical data collection method based on a medical question and answer big model provided in an embodiment of the present invention.
[0059] Figure 4 This is an example diagram of a training sample adopted by an embodiment of the present invention.
[0060] Figure 5 An example diagram of standardized format data provided by an embodiment of the present invention.
[0061] Figure 6 An example diagram of another standardized format data provided by an embodiment of the present invention.
[0062] Figure 7A schematic diagram of the structure of a medical data collection device based on a medical question and answer big model provided in an embodiment of the present invention.
[0063] Figure 8 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0064] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.
[0065] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0066] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.
[0067] In addition, the terms “first”, “second”, etc., if used, are merely used to distinguish between the descriptions and should not be understood as indicating or implying relative importance.
[0068] It should be noted that, in the absence of conflict, the features in the embodiments of the present invention may be combined with each other.
[0069] As mentioned in the background technology, when users (referring to internal hospital personnel, such as hospital managers or medical staff) collect medical data, they may need to select appropriate tools from a variety of management tools to collect data. Sometimes, they may also need to use multiple tools for joint query, which requires users to be very familiar with the operation of each management tool in order to quickly complete the collection. If they are not familiar with the operation, it will take a lot of time to familiarize themselves with the operation of each management tool, which is extremely time-consuming. In addition, the medical data collected at one time may include multiple indicators, which requires users to spend a lot of time to integrate.
[0070] Based on the discovery of the above technical problems, the inventors have proposed the following technical solutions to solve or improve the above problems through creative work. It should be noted that the defects in the solutions in the above prior art are the results obtained by the inventors after practice and careful research. Therefore, the discovery process of the above problems and the solutions proposed in the embodiments of the present application for the above problems below should all be the contributions made by the inventors to the present application in the process of invention and creation, and should not be understood as technical contents known to those skilled in the art.
[0071] An embodiment of the present invention provides a medical data collection method based on a medical question-and-answer big model, which can be applied to electronic devices, and the electronic devices may be but are not limited to personal computers, laptops, smart phones, smart tablets, servers, etc.
[0072] Please refer to Figure 1 , Figure 1 A flowchart of a medical data collection method based on a medical question-answering big model provided in an embodiment of the present invention is provided. The medical data collection method based on the medical question-answering big model comprises the following steps S101 to S103:
[0073] S101, obtaining the question text of the question to be queried.
[0074] In this embodiment, the question to be queried can be input by voice or text input by the user.
[0075] Optionally, taking the execution subject as a server as an example, the question to be queried can be the user's voice input in a personal terminal (mobile phone, tablet or personal computer), and then the personal terminal performs voice recognition on the user's input voice to obtain the question text and upload it to the server for query, or the question to be queried can be the user's text input on the personal terminal, and the personal terminal directly uploads the question text of the question to be queried input by the user to the server.
[0076] S102. Based on the question text, search the historical question and answer database to see whether there is a question and answer result for the question to be queried.
[0077] In this embodiment, the historical question and answer database can store historical question and answer results of multiple historical question texts. In order to reduce the computing power consumption caused by calling the medical question and answer big model, you can first directly search from the historical question and answer database to determine whether there is a historical question text similar or identical to the question to be queried. If so, the historical question and answer results of the historical question text can be directly used as the question and answer results of the question to be queried; if not, the medical question and answer big model still needs to be called.
[0078] S103. If the question and answer result of the question to be queried does not exist in the historical question and answer database, the query method of the question to be queried is determined based on the question text, and the medical question and answer big model is called based on the query method to obtain the question and answer result of the question to be queried.
[0079] In this embodiment, the query mode can be used to indicate obtaining a question and answer result based on a pre-built medical knowledge base or querying a question and answer result from a medical tool system.
[0080] The embodiment of the present invention provides a medical data collection method based on a medical question-and-answer big model, which first obtains the question text of the question to be queried, and then searches the historical question-and-answer database based on the question text to see whether there is a question-and-answer result for the question to be queried; if not, the query method of the question to be queried is determined based on the question text, and the medical question-and-answer big model is called based on the query method to obtain the question-and-answer result of the question to be queried, and the query method is used to indicate whether the question-and-answer result is obtained based on a pre-built medical knowledge base or the question-and-answer result is queried from a medical tool system. In this way, users can collect medical data through intelligent question-and-answer methods without manual work, which is fast and quick.
[0081] In an optional implementation, the historical question and answer database may store historical question vectors and historical question and answer results of multiple historical question texts. Figure 1 Based on Figure 2 The sub-steps of the above step S102 may include S1021~S1025.
[0082] S1021. Based on the question text, select a target similarity calculation method from a plurality of preset similarity calculation methods.
[0083] Optionally, the preset similarity calculation method may include the Jaccard similarity algorithm and the Euclidean distance algorithm. Accordingly, the sub-steps of step S1021 may include the following two steps:
[0084] (1) If the question text does not contain the date and medical code string, the Jaccard similarity algorithm is used as the target similarity calculation method;
[0085] (2) If the question text contains date and / or medical code strings, the Euclidean distance algorithm is used as the target similarity calculation method.
[0086] Usually, in the medical field, some diagnosis names and surgical names have common medical codes in the field. For example, the common medical code for the surgical name "post-thyroid surgery" is "Z98.800". If the question text carries a string with numbers such as a medical code and / or a date, then the order of characters in the question text is very important. The calculation of similarity should consider using a calculation method that is sensitive to the character order, so that the calculated similarity is more accurate.
[0087] The advantage of the Jaccard similarity algorithm is that it is insensitive to the order of characters. Its calculation formula is as follows:
[0088]
[0089] in, These are two vectors representing the similarity calculation. It is the Jaccard similarity between the two, also known as the Jaccard distance.
[0090] The advantage of the Euclidean distance algorithm is that it is sensitive to the order of characters. Its calculation formula is as follows:
[0091]
[0092] in, Represents the first The similarity calculated by the Euclidean distance algorithm is usually called the Euclidean distance.
[0093] For example, "2012" and "2021" contain the same four numbers, but the numbers are arranged in a different order. The similarity between the two vectors calculated using the Jaccard similarity algorithm is 100%, but in essence the meanings they express are completely different. In other words, in the case of dates, the similarity obtained using the Jaccard similarity algorithm, which is insensitive to character order, is completely inaccurate.
[0094] S1022. Convert the question text into a question vector.
[0095] In this embodiment, the encoding method for converting the question text into the question vector can adopt existing technology, such as the word2vec algorithm, and the detailed conversion process is not described here.
[0096] S1023. Calculate the similarity between the question vector and each historical question vector using a target similarity calculation method.
[0097] S1024: If there is no historical question vector with a similarity greater than a first set threshold, it is determined that there is no question and answer result for the question to be queried in the historical question and answer database.
[0098] S1025. If there is at least one historical question vector whose similarity is greater than a first set threshold, the historical question and answer result corresponding to the historical question vector with the greatest similarity is used as the question and answer result of the question to be queried.
[0099] In this embodiment, the first set threshold value can be set flexibly. For example, the first set threshold value can be set to 80% or 85%, etc., which is not limited in this embodiment of the present invention.
[0100] Because the question and answer results obtained by the medical question and answer big model have a certain degree of randomness, that is, the answers to the same question twice may be different. By setting up a historical question and answer library, the same question does not need to be answered twice by calling the medical question and answer big model. Instead, only the answers in the historical question and answer library need to be called, which ensures the stability and accuracy of the results fed back to users, while also reducing the computing power consumption caused by calling the medical question and answer big model.
[0101] In an optional implementation, a classification model can be pre-trained to determine the query method. Figure 1 Based on Figure 3 The process of "determining the query method of the question to be queried based on the question text, and calling the medical question and answer big model based on the query method to obtain the question and answer result of the question to be queried" in the above step S103 may include the following sub-steps S1031~S1036.
[0102] S1031. Obtain a question vector corresponding to the question text.
[0103] In this embodiment, the obtained question vector is obtained by conversion in the above step S1022.
[0104] S1032: Input the question vector into a pre-trained classification model to determine the query method.
[0105] In this embodiment, the input of the classification model is a question vector, and the output is a probability value between 0 and 1. If the probability value is greater than 0.5, the query method is determined to be a tool query; if the probability value is less than or equal to 0.5, the query method is determined to be a knowledge base query.
[0106] Optionally, the classification model may adopt but is not limited to: Logistic Regression, k-Nearest Neighbors algorithm, decision tree, support vector machine, Naive Bayes algorithm, neural network model, etc.
[0107] Before training the classification model, you need to collect data and construct a training set. The constructed training set can include several training problem samples and their labels, where the label represents a tool query or a knowledge base query. Taking the neural network model as an example, you can pre-build a neural network classification model, whose output layer uses a sigmoid function, and then use the constructed training set to train the pre-built neural network classification model to obtain a trained classification model.
[0108] For an optional example, see Figure 4 , Figure 4 This is an example diagram of a training problem sample adopted by an embodiment of the present invention. Figure 4 In the example, the column "Independent variables" includes the six training problem samples given as examples, and the column "Dependent variables" contains the labels of the six training problem samples. The label 0 represents the knowledge base query, and the label 1 represents the query. The column "Prediction results" contains the prediction results of the six training problem samples after one training is completed. This example is only an example and is not limited here.
[0109] Before calling the medical question-and-answer big model, the classification model is used to determine the query method, ensuring that the query method is accurately determined based on the user's inquiry needs, so that subsequent calls to the medical question-and-answer big model can be accurately queried according to the determined query method, achieving accurate and fast query.
[0110] S1033. If the query method is a knowledge base query, the medical question and answer big model is called to search the medical knowledge base for the question and answer results of the question to be queried.
[0111] Optionally, the construction process of the medical knowledge base may include the following steps S001 to S005:
[0112] S001. Obtain multiple medical documents;
[0113] In this embodiment, each acquired medical document may be preprocessed, that is, the title, table and picture in the document may be deleted.
[0114] S002. For any medical document, segment each paragraph of the medical document according to a preset segmentation length to obtain multiple text segments.
[0115] In this embodiment, for a medical document, each paragraph of the medical document is segmented according to a preset segmentation length to obtain multiple text segments. This ensures that the content of a text segment belongs to one paragraph, and does not cause a cross-paragraph situation where a text segment belongs to two adjacent paragraphs, thereby avoiding inaccurate question and answer results queried from the medical knowledge base due to the cross-paragraph situation.
[0116] Optionally, the preset segmentation length may be flexibly set, for example, the preset segmentation length may be 20 characters or 30 characters, which is only an example and is not limited here.
[0117] S003. Perform padding processing on the text segments whose length is shorter than the preset segmentation length.
[0118] In this embodiment, since the number of words in each paragraph of a medical document is different, when a paragraph is segmented, the length of the last segmented text segment may be less than the preset segmentation length.
[0119] However, due to the complex structure and correlation of medical data, including the connection between patients, the doctor's diagnostic process, etc., data segmentation itself will cause some information loss. In order to retain data information as much as possible, to make the content of each text segment complete, and to prevent data information loss or information confusion due to segmentation, it is necessary to fill in this special text segment whose length is less than the preset segmentation length so that the length of all text segments after filling is the preset segmentation length.
[0120] S004. Convert each text segment into a content vector to obtain multiple content vectors corresponding to the medical document.
[0121] S005. Traverse each medical document to obtain multiple content vectors corresponding to each medical document, and complete the construction of the medical knowledge base.
[0122] In this embodiment, for each acquired medical document, the above steps S002 to S004 are executed to finally obtain a medical knowledge base. When new medical documents are subsequently introduced, the above steps S002 to S004 are also executed to expand the content of the medical knowledge base.
[0123] Therefore, the pre-built medical knowledge base may include multiple content vectors corresponding to multiple medical documents. Correspondingly, the sub-steps of step S1033 may include S10331 to S10333.
[0124] S10331. Calculate the similarity between the question vector and each content vector in the medical knowledge base, and use all content vectors whose similarity exceeds a second set threshold as target content vectors.
[0125] In this embodiment, when calculating the similarity between the question vector and each content vector in the medical knowledge base, the similarity calculation method adopted is the target similarity calculation method determined in the above step S1021 to ensure the accuracy of the similarity calculation.
[0126] S10332. Obtain the target medical document corresponding to each target content vector.
[0127] In this embodiment, each acquired target medical document is complete, including all text paragraphs, titles, charts and tables.
[0128] S10333. Call the medical question-answering big model to generate question-answering results for the question to be queried based on all target medical documents and question texts.
[0129] In this embodiment, the medical question-answering model can extract the question-answering results of the query question from all target medical documents based on the question text, and then return them to the user.
[0130] Optionally, the medical question-answering model can use large language models (LLMs), such as but not limited to ChatGPT, GPT-4, Chatglm3-6b, etc. The medical question-answering model has high requirements on the server and is not suitable for segmented industries. Through quantitative adjustment, it is suitable for low-configuration servers. Therefore, the medical question-answering model uses the int4 quantization method, that is, the parameter type is converted from float32 to int4 type, which can save memory and improve the model response speed.
[0131] S1034. If the query method is tool query, the medical question and answer big model is called to perform intent analysis on the question text to obtain standardized format data.
[0132] In this embodiment, the standardized format data may be a JSON format data, including information such as the query object, query time, and indicators to be queried.
[0133] Optionally, the sub-steps of step S1034 may include S10341~S10344.
[0134] S10341. Call the medical question-answering big model to perform intent analysis on the question text to determine the query object, query time, and indicator data.
[0135] In this embodiment, the indicator data may include specified quantitative indicators and / or abstract indicators. For example, medical efficiency or medical quality, which cannot be quantified by a single value but require a comprehensive representation of multiple quantitative indicators, are abstract indicators.
[0136] S10342. If there are specified quantitative indicators in the indicator data, each specified quantitative indicator in the indicator data is used as an indicator to be queried.
[0137] S10343. If there are abstract indicators in the indicator data, multiple quantitative indicators corresponding to the abstract indicators are determined from the indicator mapping library to obtain multiple indicators to be queried.
[0138] In this embodiment, the indicator mapping library may include a plurality of preset abstract indicators and a plurality of quantitative indicators corresponding to each preset abstract indicator. Optionally, the indicator mapping library may be stored in the medical knowledge base or may exist independently.
[0139] S10344. Generate standardized format data based on the query object, query time, and each indicator to be queried.
[0140] In this embodiment, the standardized format data in JSON format conforms to the input parameter format of the data query tool, that is, it includes three parameters: query object, query time, and query index.
[0141] S1035. Send the standardized format data to the data query tool, so that the data query tool generates a query statement based on the standardized format data to obtain query response data from the medical tool system.
[0142] In this embodiment, the data query tool can generate SQL (Structured Query Language) query statements based on standardized format data to obtain query response data from the corresponding medical tool system. The query response data is also in JSON format, including the indicator value of each indicator to be queried.
[0143] S1036. Use the question and answer result table returned by the data query tool as the question and answer result of the question to be queried.
[0144] In this embodiment, the question and answer result table is obtained by converting the query response data by the data query tool.
[0145] In the indicator mapping library, the preset abstract indicators may include but are not limited to: medical quality, medical benefits, medical efficiency, workload, three-level public, basic information, DRGs (Diagnosis Related Groups), CN-DRGs (i.e. Chinese Diagnosis Related Groups), ICU (Intensive Care Unit) information, surgery information (including operation), surgery information (excluding operation), three-level review, medical technology income information, operating costs, medical insurance payment information, maternal and child indicators, cost information, etc. The corresponding quantitative indicators are as follows:
[0146] (1) The quantitative indicators corresponding to the abstract indicator of "medical quality" include: the number of unplanned readmissions within 31 days for the same disease, the rate of unplanned readmissions within 31 days for the same disease, the number of cured and improved patients, the rate of cured and improved patients, the number of deaths on the day after surgery, the number of deaths within 24 hours after surgery, etc.; medical efficiency, including indicators such as the mean length of stay, the expected mean length of stay, the O / E value of the length of stay (O represents the actual value, E represents the expected value), the number of extreme lengths of stay, the total length of stay, the median length of stay, the standard deviation of the length of stay, the opportunities for improvement in the length of stay, the number of additional admissions, the proportion of hospitalizations exceeding 30 days, and the number of days a ventilator is used, etc.;
[0147] (2) The quantitative indicators corresponding to the abstract indicator of “medical benefit” include: total cost, mean total cost, expected mean total cost, total cost O / E, total cost improvement opportunity, median total cost, standard deviation of total cost, total medical service cost, mean medical service cost, expected mean medical service cost, O / E value of medical service cost, room for improvement of medical service cost, standard deviation of medical service cost, proportion of medical service cost, medical service cost per bed day, total drug cost, mean drug cost, expected mean drug cost, drug cost O / E value, opportunities for improvement in drug costs, total drug share, drug share (excluding Chinese herbal medicines), antibiotic share, Chinese herbal medicine share, Chinese medicine share, total consumables costs, average consumables costs, expected average consumables costs, O / E value of consumables costs, opportunities for improvement in consumables costs, consumables share (excluding drug costs), consumables share (including drug costs), total surgery costs, average surgery costs, average surgical consumables costs (including all patients), average surgical consumables costs (only surgical patients), expected average surgical consumables costs, surgical consumables Cost O / E, surgical consumables cost improvement opportunities, surgical consumables ratio, total auxiliary examination costs, auxiliary examination cost average, auxiliary examination cost ratio, test and inspection cost average, test and inspection cost expected average, test and inspection cost O / E, test and inspection cost, blood product costs, TCM treatment cost ratio, out-of-pocket cost ratio, medical service income, medical service income ratio, TCM treatment cost ratio, TCM treatment cost ratio, Chinese medicine decoction piece costs, Chinese medicine decoction piece costs average, Chinese medicine decoction piece ratio in total costs, Chinese medicine decoction piece ratio in total medicine costs, decoction-free granule costs , the proportion of decoction-free granules in total expenses, the proportion of decoction-free granules in total drug expenses, Chinese patent medicine expenses, the average Chinese patent medicine expenses, the proportion of Chinese patent medicines in total expenses, the proportion of Chinese patent medicines in total drug expenses, western medicine expenses, the average western medicine expenses, the proportion of western medicine in total expenses, the proportion of western medicine in total drug expenses, drug expenses, the proportion of drugs, the average cost of disposable medical materials for examination, the average cost of disposable medical materials for treatment, treatment expenses, the average cost of treatment expenses, the proportion of treatment expenses, examination fees, the proportion of examination fees, inspection fees, the proportion of inspection fees, pathological diagnosis fees, clinical diagnosis fees, antimicrobial drug expenses, etc.;
[0148] (3) The quantitative indicators corresponding to the abstract indicator of “medical efficiency” include: mean length of stay, expected mean length of stay, O / E value of length of stay, number of extreme length of stay, total length of stay, median length of stay, standard deviation of length of stay, opportunities for improvement in length of stay, number of additional admissions, percentage of hospitalization for more than 30 days, number of days of ventilator use, etc.;
[0149] (4) The quantitative indicators corresponding to the abstract indicator “workload” include: number of patients, ACMI, number of minimally invasive surgeries, proportion of discharged patients undergoing minimally invasive surgeries, number of level IV surgeries, proportion of level IV surgeries, number of surgical complications, incidence of complications in surgical patients, proportion of day surgeries to elective surgeries, proportion of basic diseases, proportion of difficult and severe patients, number of basic and difficult diseases in tertiary hospitals, proportion of basic and difficult diseases in tertiary hospitals, etc.;
[0150] (5) The quantitative indicators corresponding to the abstract indicator of "three-level public" include: the number of minimally invasive surgeries, the proportion of minimally invasive surgeries in discharged patients, the number of level 4 surgeries, the proportion of level 4 surgeries, the number of surgical complications, the incidence of complications in surgical patients, the proportion of day surgeries in elective surgeries, and the number of minimally invasive surgeries;
[0151] (6) The quantitative indicators corresponding to the abstract indicator “basic information” include: number of patients, ACMI (Adjusted Case Mix Index, i.e., risk-adjusted case mix index, which is the mean of all hospitalized cases AW), number of patients with ACMI between 0 and 3, number of patients with ACMI between 3 and 6, number of patients with ACMI between 6 and 10, percentage of patients with ACMI between 0 and 3, percentage of patients with ACMI between 3 and 6, percentage of patients with ACMI between 6 and 10, ACMP (Adjusted Case Mix Production, risk-adjusted total weight), percentage of basic diseases, percentage of patients with difficult and severe diseases, number of patients with basic and difficult diseases in tertiary hospitals, percentage of basic and difficult diseases in tertiary hospitals, CMI (Case Mix Production). The number of patients with CMI less than 0.5, the proportion of patients with CMI between 0.5 and 2, the low-risk mortality rate, the proportion of patients with CMI between 0.5 and 2, the high-risk treatment success rate, the number of patients with CMI between 2 and 10, the proportion of patients with CMI between 2 and 10, the number of patients with CMI greater than 10, the proportion of patients with CMI greater than 10, the number of out-of-town patients, the proportion of out-of-town patients, the number of transfers, the proportion of transfers, the coverage rate of difficult diseases, the number of patients treated with traditional Chinese medicine, the rate of traditional Chinese medicine treatment, the number of patients treated mainly with traditional Chinese medicine, the proportion of patients treated mainly with traditional Chinese medicine, the number of patients from outside the province, the proportion of patients from outside the province, etc.
[0152] (7) The quantitative indicators corresponding to the abstract indicator "DRGs" include: time efficiency, cost efficiency, CMI (Cornell Medical Index), CMP (Comprehensive Metabolic Panel), number of DRG groups, number of cases included, inclusion rate, DRG coverage rate, inclusion rate (medical insurance), mortality rate of low-risk group, number of deaths in low-risk group, number of people in low-risk group, percentage of people in low-risk group, number of people in high-risk group, percentage of people in high-risk group, MCC / CC (MCC stands for Major Complication & Comorbidity, which refers to major complications or comorbidities; CC stands for Complication & Comorbidity, which refers to complications or comorbidities) capture rate, etc.
[0153] (8) The quantitative indicators corresponding to the abstract indicator "CN-DRGs" include: time efficiency, cost efficiency, CMI, CMP, number of DRG groups, number of cases enrolled, enrollment rate, DRG coverage rate, mortality rate of low-risk group, number of deaths in low-risk group, number of people in low-risk group, percentage of people in low-risk group, number of people in high-risk group, percentage of people in high-risk group, etc.;
[0154] (9) The quantitative indicators corresponding to the abstract indicator “ICU information” include: number of ICU patients, average ICU hospitalization days, proportion of ICU patients, etc.;
[0155] (10) The quantitative indicators corresponding to the abstract indicator “surgical information (excluding operations)” include: number of surgical operations, number of operations per capita, number of surgeries, percentage of number of surgeries, number of surgical cases, number of operating tables, number of first-level surgeries, percentage of first-level surgeries, number of first-level surgeries, percentage of first-level surgeries, number of second-level surgeries, percentage of second-level surgeries, number of third-level surgeries, percentage of third-level surgeries, number of third-level surgeries, percentage of third-level surgeries, number of fourth-level surgeries, percentage of fourth-level surgeries , number of level 4 surgeries, percentage of level 4 surgeries, number of level 2 and 3 surgeries, percentage of level 2 and 3 surgeries, number of level 2 and 3 surgeries, percentage of level 2 and 3 surgeries, number of level 3 and 4 surgeries, percentage of level 3 and 4 surgeries, number of level 3 and 4 surgeries, percentage of level 3 and 4 surgeries, mean number of days before surgery, mean number of days after surgery, number of day surgeries, percentage of day surgeries, number of interventional treatments, number of elective surgeries, number of elective surgery tables, number of level 1 surgery tables, number of level 2 surgery tables, number of level 3 surgery tables, number of level 4 surgery tables, etc.;
[0156] (11) The quantitative indicators corresponding to the abstract indicator “surgical information (excluding operation)” include: number of surgeries per capita, number of surgeries, percentage of number of surgeries, number of surgeries, number of type I surgical site infections, type I surgical site infection rate, number of operating tables, number of first-level surgeries, percentage of first-level surgeries, number of first-level surgeries, percentage of first-level surgeries, number of second-level surgeries, percentage of second-level surgeries, number of third-level surgeries, percentage of third-level surgeries, number of fourth-level surgeries, percentage of fourth-level surgeries, number of second-level surgeries, percentage of fourth-level surgeries, number of second-level surgeries, percentage of fourth-level surgeries, number of second- and third-level surgeries, percentage of second- and third-level surgeries, number of third- and fourth-level surgeries, percentage of third- and fourth-level surgeries, number of elective surgeries, number of elective operating tables, number of first-level operating tables, number of second-level operating tables, number of third-level operating tables, number of fourth-level operating tables, etc.
[0157] (12) The quantitative indicators corresponding to the abstract indicator of “three-level review” include: number of surgical complications, incidence of surgical complications, etc.;
[0158] (13) The quantitative indicators corresponding to the abstract indicator “medical technology income information” include: (medical technology) surgical income, (medical technology) imaging income, (medical technology) inspection income, (medical technology) anesthesia income, (medical technology) other income, etc.;
[0159] (14) The quantitative indicators corresponding to the abstract indicator "operating cost" include: actual total cost, actual medical technology cost, surgical cost, imaging cost, laboratory test cost, other costs, total surplus, average surgical income, average surgical cost, surgical cost percentage, average imaging income, average imaging cost, imaging cost percentage, average laboratory test income, average laboratory test cost, laboratory test cost percentage, bed-day average cost, bed-day cost percentage, other average income, other average cost, other cost percentage, drug cost, drug cost percentage, consumables cost, consumables cost percentage, etc.;
[0160] (15) The quantitative indicators corresponding to the abstract indicator “medical insurance payment information” include: medical insurance payment score, medical insurance payment cost (forecast), medical insurance payment balance (forecast), medical insurance payment cost (actual), medical insurance payment balance (actual), etc.;
[0161] (16) The quantitative indicators corresponding to the abstract indicator “maternal and child indicators” include: the proportion of discharged patients undergoing surgery, the proportion of discharged patients undergoing minimally invasive surgery, the proportion of discharged patients undergoing high-level surgery, the incidence of surgical complications, the rate of surgical site infection of type I incisions, the average length of stay for cesarean section, the mortality rate of cesarean section, the average length of stay for pneumonia, the mortality rate of pneumonia, etc.;
[0162] (17) The quantitative indicators corresponding to the abstract indicator “cost information” include: medical income, average medical income, medical cost, average medical cost, medical surplus, average surplus per medical visit, medical surplus rate, medical service income, proportion of medical service income, clinical department operation (income), clinical department operation (cost), clinical department operation (surplus), clinical department operation (cost proportion), improveable bed-day cost, drug cost (income), drug cost (cost), drug cost (surplus), drug cost (cost proportion), drug cost (income proportion), improveable drug cost, consumables cost (income), consumables cost (cost), consumables cost (surplus), consumables cost (cost proportion), consumables cost (income proportion), improveable consumables cost, drug management (income), drug management (cost), drug management (surplus), drug management (cost proportion), consumables management (income), consumables management (cost), consumables management (surplus), consumables management (cost proportion), personnel expenses (income), personnel expenses (cost), personnel expenses Funds (surplus), personnel expenses (% of cost), numbness of hands (income), numbness of hands (cost), numbness of hands (surplus), numbness of hands (% of cost), blood fees (income), blood fees (cost), blood fees (surplus), blood fees (% of cost), examination (income), examination (cost), examination (surplus), examination (% of cost), treatment (income), treatment (cost), treatment (surplus), treatment (% of cost), vaccine fees (income), vaccine fees (cost), vaccine fees (surplus), vaccine fees (% of cost), diagnosis (income), diagnosis (cost), diagnosis (surplus), diagnosis (% of cost), radiotherapy fees (income), radiotherapy fees (cost), radiotherapy fees (surplus), radiotherapy fees (% of cost), testing (income), testing (cost), testing (surplus), testing (% of cost), surgery fees (income), surgery fees (cost), surgery fees (surplus), surgery fees (% of cost), others (income), others (cost), others (surplus), others (% of cost), etc.
[0163] In an optional example, suppose the query question is: "Help me analyze the medical efficiency of each department in 2023 according to clinical disciplines", then the standardized format data generated by the medical question-answering model after parsing is as follows: Figure 5 As shown, in this example, the indicator data only includes the abstract indicator of "medical efficiency". Using the indicator mapping library, 11 indicators to be queried can be determined.
[0164] In an optional example, suppose the query question is: "Help me analyze the medical efficiency indicators of clinical disciplines in 2023, which must include indicators such as the number of patients, CMI, CCMI, etc." Then the standardized format data generated by the medical question-answering big model after parsing is as follows: Figure 6As shown, in this example, the indicator data includes the abstract indicator "medical efficiency" and the three specified quantitative indicators "number of patients, CMI, CCMI". Combined with the indicator mapping library, 14 indicators to be queried can be finally determined.
[0165] The above two are just examples and are not limiting here.
[0166] It should be noted that the execution order of each step in the above method embodiment is not limited to that shown in the drawings, and the execution order of each step shall be based on the actual application situation.
[0167] In order to execute the corresponding steps in the above method embodiment and each possible implementation method, an implementation method of a medical data collection device based on a medical question and answer big model is given below.
[0168] See also Figure 7 , Figure 7 The schematic diagram of the structure of the medical data collection device based on the medical question and answer big model provided by the embodiment of the present invention is shown. The medical data collection device 200 based on the medical question and answer big model includes: an acquisition module 210 and a query module 220.
[0169] The acquisition module 210 is used to obtain the question text of the question to be queried:
[0170] The query module 220 is used to: based on the question text, search from the historical question and answer library to see whether there is a question and answer result for the question to be queried; if not, determine the query method for the question to be queried based on the question text, and call the medical question and answer big model based on the query method to obtain the question and answer result for the question to be queried. The query method is used to indicate whether to obtain the question and answer result based on a pre-built medical knowledge base or to query the question and answer result from the medical tool system.
[0171] Optionally, the historical question and answer library stores historical question vectors and historical question and answer results of multiple historical question texts. When the query module 220 is used to search for question and answer results of the question to be queried in the historical question and answer library based on the question text, it can be specifically used to: select a target similarity calculation method from a plurality of preset similarity calculation methods based on the question text; convert the question text into a question vector; use the target similarity calculation method to calculate the similarity between the question vector and each historical question vector; if there is no historical question vector with a similarity greater than a first set threshold, it is determined that there is no question and answer result of the question to be queried in the historical question and answer library; if there is at least one historical question vector with a similarity greater than the first set threshold, the historical question and answer result corresponding to the historical question vector with the largest similarity is used as the question and answer result of the question to be queried.
[0172] Optionally, the plurality of preset similarity calculation methods include Jaccard similarity algorithm and Euclidean distance algorithm. The query module 220 is used to select a target similarity calculation method from a plurality of preset similarity calculation methods based on the question text, and specifically can be used to: if the question text does not contain a date and a medical code string, use the Jaccard similarity algorithm as the target similarity calculation method; if the question text contains a date and / or a medical code string, use the Euclidean distance algorithm as the target similarity calculation method.
[0173] Optionally, the query module 220 is used to determine the query method of the question to be queried based on the question text, and when calling the medical question and answer big model based on the query method to obtain the question and answer result of the question to be queried, it can be specifically used to: obtain the question vector corresponding to the question text; input the question vector into a pre-trained classification model to determine the query method; if the query method is a knowledge base query, call the medical question and answer big model to search the question and answer result of the question to be queried from the medical knowledge base; if the query method is a tool query, call the medical question and answer big model to perform intent analysis on the question text to obtain standardized format data; send the standardized format data to the data query tool, so that the data query tool generates a query statement based on the standardized format data to obtain query response data from the medical tool system; use the question and answer result table returned by the data query tool as the question and answer result of the question to be queried; the question and answer result table is obtained by the data query tool through conversion based on the query response data.
[0174] Optionally, when the query module 220 is used to call the medical question and answer big model to perform intent analysis on the question text to obtain standardized format data, it can be specifically used to: call the medical question and answer big model to perform intent analysis on the question text to determine the query object, query time and indicator data; if there are specified quantitative indicators in the indicator data, each specified quantitative indicator in the indicator data is used as the indicator to be queried; if there are abstract indicators in the indicator data, multiple quantitative indicators corresponding to the abstract indicators are determined from the indicator mapping library to obtain multiple indicators to be queried; the indicator mapping library includes multiple preset abstract indicators and multiple quantitative indicators corresponding to each preset abstract indicator; based on the query object, query time and each indicator to be queried, standardized format data is generated.
[0175] Optionally, the medical knowledge base includes multiple content vectors corresponding to multiple medical documents.
[0176] When the query module 220 is used to call the medical question and answer big model to search the question and answer results of the question to be queried from the medical knowledge base, it can be specifically used to: calculate the similarity between the question vector and each content vector in the medical knowledge base, and use all content vectors whose similarity exceeds a second set threshold as target content vectors; obtain the target medical document corresponding to each target content vector; call the medical question and answer big model to generate the question and answer results of the question to be queried based on all the target medical documents and the question text.
[0177] Optionally, the medical data collection device 200 based on the medical question and answer big model also includes a construction module 230, which is used to construct a medical knowledge base, specifically for: obtaining multiple medical documents; for any medical document, segmenting each paragraph of the medical document according to a preset segmentation length to obtain multiple text segments; for text segments whose length is less than the preset segmentation length, performing padding processing; converting each text segment into a content vector to obtain multiple content vectors corresponding to the medical document; traversing each medical document to obtain multiple content vectors corresponding to each medical document, and completing the construction of the medical knowledge base.
[0178] Technical personnel in the relevant field can clearly understand that, for the convenience and brevity of description, the specific working process of the medical data collection device 200 based on the medical question and answer big model described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.
[0179] See also Figure 8 , Figure 8 The electronic device 300 includes a processor 310 , a memory 320 , and a bus 330 , wherein the processor 310 is connected to the memory 320 via the bus 330 .
[0180] The memory 320 may be used to store software programs, for example, software programs corresponding to the medical data collection device 200 based on the medical question-answering big model provided in the embodiment of the present invention. The processor 310 executes various functional applications and data processing to implement the medical data collection method based on the medical question-answering big model provided in the embodiment of the present invention by running the software programs stored in the memory 320.
[0181] Among them, the memory 320 can be but is not limited to: RAM (Random Access Memory), ROM (Read Only Memory), FLASH (Flash Memory), PROM (Programmable Read-Only Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electric Erasable Programmable Read-Only Memory), etc.
[0182] The processor 310 may be an integrated circuit chip with signal processing capability. The processor 310 may be a general-purpose processor, including: CPU (Central Processing Unit), NP (Network Processor), etc.; it may also be: DSP (Digital Signal Processing), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
[0183] Understandably, Figure 8 The structure shown is for illustration only. The electronic device 300 may also include Figure 8 More or fewer components as shown, or with Figure 8 Different configurations are shown. Figure 8 Each component shown in the figure can be implemented by hardware, software or a combination thereof.
[0184] The embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the medical data collection method based on the medical question-answering model disclosed in the above embodiment is implemented. The computer-readable storage medium can be, but is not limited to, various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a PROM, an EPROM, an EEPROM, a FLASH disk, or an optical disk.
[0185] In summary, the embodiment of the present invention provides a medical data collection method based on a medical question-and-answer big model, which first obtains the question text of the question to be queried, and then searches the historical question-and-answer library based on the question text to see whether there is a question-and-answer result for the question to be queried; if not, the query method of the question to be queried is determined based on the question text, and the medical question-and-answer big model is called based on the query method to obtain the question-and-answer result of the question to be queried, and the query method is used to indicate whether the question-and-answer result is obtained based on a pre-built medical knowledge base or the question-and-answer result is queried from a medical tool system. In this way, users can collect medical data through intelligent question-and-answer methods without manual work, which is fast and quick.
[0186] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims
1. A medical data collection method based on a medical question-answering big model, characterized in that: include: Get the question text of the question to be queried: Based on the question text, searching the historical question and answer database for a question and answer result of the question to be queried; The historical question and answer library stores historical question vectors and historical question and answer results of multiple historical question texts; If it does not exist, then obtain the question vector corresponding to the question text, and input the question vector into a pre-trained classification model to determine the query method of the question to be queried; wherein the output of the classification model is a probability value between 0 and 1, and if the probability value is greater than 0.5, then determine that the query method is a tool query; if the probability value is less than or equal to 0.5, then determine that the query method is a knowledge base query; If the query method is a knowledge base query, then the similarity between the question vector and each content vector in the medical knowledge base is calculated, and all content vectors whose similarity exceeds a second set threshold are used as target content vectors, wherein the medical knowledge base includes multiple content vectors corresponding to multiple medical documents respectively; the target medical document corresponding to each target content vector is obtained; the medical question and answer big model is called to generate the question and answer result of the question to be queried based on all the target medical documents and the question text; If the query method is a tool query, the medical question-answering big model is called to perform intent analysis on the question text to determine the query object, query time and indicator data; If there is a specified quantitative indicator in the indicator data, each specified quantitative indicator in the indicator data is used as the indicator to be queried; if there is an abstract indicator in the indicator data, multiple quantitative indicators corresponding to the abstract indicator are determined from the indicator mapping library to obtain multiple indicators to be queried; the indicator mapping library includes multiple preset abstract indicators and multiple quantitative indicators corresponding to each of the preset abstract indicators; Generate standardized format data based on the query object, query time and each indicator to be queried; sending the standardized format data to a data query tool so that the data query tool generates a query statement based on the standardized format data to obtain query response data from a medical tool system; The question and answer result table returned by the data query tool is used as the question and answer result of the question to be queried; the question and answer result table is obtained by the data query tool through conversion based on the query response data.
2. The method according to claim 1, characterized in that The step of searching, based on the question text, whether there is a question and answer result for the question to be queried from the historical question and answer library comprises: Based on the question text, selecting a target similarity calculation method from a plurality of preset similarity calculation methods; Convert the question text into a question vector; Calculating the similarity between the question vector and each of the historical question vectors using the target similarity calculation method; If there is no historical question vector whose similarity is greater than the first set threshold, it is determined that there is no question and answer result of the question to be queried in the historical question and answer database; If there is at least one historical question vector whose similarity is greater than the first set threshold, the historical question and answer result corresponding to the historical question vector with the greatest similarity is used as the question and answer result of the question to be queried.
3. The method according to claim 2, characterized in that The multiple preset similarity calculation methods include Jaccard similarity algorithm and Euclidean distance algorithm; The step of selecting a target similarity calculation method from a plurality of preset similarity calculation methods based on the question text includes: If the question text does not contain a date and a medical code string, the Jaccard similarity algorithm is used as the target similarity calculation method; If the question text contains a date and / or a medical code character string, the Euclidean distance algorithm is used as the target similarity calculation method.
4. The method according to any one of claims 1 to 3, characterized in that: The construction process of the medical knowledge base includes: Access to multiple medical literature; For any of the medical documents, segment each paragraph of the medical document according to a preset segmentation length to obtain a plurality of text segments; For text segments whose length is shorter than the preset segmentation length, padding is performed; Convert each of the text segments into a content vector to obtain a plurality of content vectors corresponding to the medical document; Each of the medical documents is traversed to obtain a plurality of content vectors corresponding to each of the medical documents, thereby completing the construction of the medical knowledge base.
