Training methods, devices, electronic equipment, and storage media for medical order classification models
By constructing a medical order classification model based on a medical order dictionary and medical entities, the problem of low efficiency and accuracy of manual classification was solved, and efficient medical order classification specific to hospitals was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING UNISOUND INFORMATION TECH CO LTD
- Filing Date
- 2022-09-06
- Publication Date
- 2026-07-17
AI Technical Summary
In the medical environment, the efficiency and accuracy of medical order classification are limited by manual processing, making it difficult to meet the rapid and diverse medical needs.
By acquiring the target hospital's medical order dictionary and medical entities, a set of medical terms is constructed. Similarity calculation and medical knowledge graph are used to expand the entities, train the initial medical order classification model, and generate a medical order classification model specific to the target hospital.
It enables rapid classification of medical orders for different hospitals, improving the efficiency and accuracy of medical order classification and overcoming the limitations of manual classification.
Smart Images
Figure CN115456069B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a training method, apparatus, electronic device and storage medium for a medical order classification model. Background Technology
[0002] In a medical setting, medical orders are instructions from doctors to patients regarding diet, medication, laboratory tests, and other aspects. With the increasing number of patients and the diversification of diseases, manually extracting and classifying medical orders has significant limitations in both efficiency and accuracy. Summary of the Invention
[0003] In order to solve the above-mentioned technical problems, or at least partially solve the above-mentioned technical problems, this application provides a training method, device, electronic device and storage medium for a medical order classification model.
[0004] According to one aspect of the embodiments of this application, a method for training a medical order classification model is provided, comprising:
[0005] Obtain a set of medical terms for training the medical order classification model, wherein the set of medical terms is obtained based on the medical order dictionary and medical entities corresponding to the target hospital;
[0006] Obtain the annotation information corresponding to each target medical term in the medical terminology set;
[0007] An initial medical order classification model is trained using the medical terminology set and the annotation information, so that the initial medical order classification model outputs the prediction result corresponding to each target medical term in the medical terminology set;
[0008] If the prediction result matches the annotation information, the initial medical order classification model is determined as the target medical order classification model corresponding to the target hospital.
[0009] Furthermore, obtaining the set of medical terms used to train the medical order classification model includes:
[0010] Obtain the medical order dictionary and current medical entity corresponding to the target hospital, wherein the medical order dictionary includes at least one medical term;
[0011] The current medical entity is stored in the database, and candidate medical entities that match the current medical entity are retrieved from the database;
[0012] Calculate the first similarity between each medical term and the candidate medical entity, and identify the medical terms whose first similarity is less than a first preset similarity as the target medical terms;
[0013] A set of medical terms is constructed using target medical terms to train a medical order classification model.
[0014] Furthermore, the step of retrieving candidate medical entities from the database that match the medical term includes:
[0015] Calculate the second similarity between medical entities in the database and medical terms included in the medical order dictionary;
[0016] The medical entities with a second similarity greater than the second preset similarity are identified as the candidate medical entities.
[0017] Furthermore, the step of constructing a medical terminology set for training the medical order classification model using target medical terms includes:
[0018] Randomly select one target medical term from the target medical terms as the first target medical term, and construct a non-homogeneous term set based on the first target medical term;
[0019] Obtain non-homogeneous terms corresponding to the first target medical term from the second target medical term, wherein the second target medical term is a medical term other than the first target medical term from the target medical term;
[0020] The non-homogeneous terms are added to the non-homogeneous term set to obtain an updated non-homogeneous term set, and the updated homogeneous term set is used as the medical term set for training the medical order classification model.
[0021] Furthermore, the step of obtaining non-homogeneous terms corresponding to the first target medical terms from the second target medical terms includes:
[0022] Calculate the cosine similarity between the first target medical term and the second target medical term;
[0023] The second target medical term whose cosine similarity is less than a preset threshold is determined as the non-homogeneous term corresponding to the first target medical term.
[0024] Furthermore, the method also includes:
[0025] Entity recognition is performed on medical terms that do not match the prediction results and the annotation information to obtain an initial entity set;
[0026] Based on the medical knowledge graph, the synonyms and subordinate entities of each entity in the entity set are obtained. The initial entity set is expanded using the synonyms and subordinate entities of each entity to obtain the target entity set.
[0027] The target medical order classification model is further trained using the target entity set.
[0028] Furthermore, the step of further training the target medical order classification model using the target entity set includes:
[0029] By embedding pseudo-medical terms generated using a medical order template into the target entity set, an updated target entity set is obtained.
[0030] The updated target entity set is used to continue training the target medical order classification model.
[0031] According to another aspect of the embodiments of this application, a training device for a medical order classification model is also provided, comprising:
[0032] The first acquisition module is used to acquire a set of medical terms for training the medical order classification model, wherein the set of medical terms is obtained based on the medical order dictionary and medical entities corresponding to the target hospital.
[0033] The second acquisition module is used to acquire the annotation information corresponding to each target medical term in the medical term set;
[0034] The prediction module is used to train an initial medical order classification model using the medical terminology set and the annotation information, so that the initial medical order classification model outputs the prediction result corresponding to each target medical term in the medical terminology set;
[0035] The processing module is used to determine the initial medical order classification model as the target medical order classification model corresponding to the target hospital when the prediction result matches the annotation information.
[0036] According to another aspect of the embodiments of this application, a storage medium is also provided, the storage medium including a stored program that executes the above steps when the program is run.
[0037] According to another aspect of the embodiments of this application, an electronic device is also provided, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; wherein: the memory is used to store computer programs; and the processor is used to execute the steps in the above method by running the programs stored in the memory.
[0038] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the steps in the above-described method.
[0039] Compared with the prior art, the technical solution provided in this application has the following advantages: the method provided in this application can train different medical order classification models for different hospitals, so that each hospital can quickly classify the content of medical orders, thereby solving the problem that manual classification has great limitations in efficiency and accuracy. Attached Figure Description
[0040] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0042] Figure 1 A flowchart illustrating a method for a medical order classification model provided in this application embodiment;
[0043] Figure 2 A flowchart illustrating a method for a medical order classification model, as provided in another embodiment of this application;
[0044] Figure 3 A block diagram of a device for a medical order classification model provided in an embodiment of this application;
[0045] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0046] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0047] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another similar entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0048] This application provides a training method, apparatus, electronic device, and storage medium for a medical order classification model. The method provided in this embodiment can be applied to any desired electronic device, such as a server, terminal, or other electronic device. For ease of description, it will be referred to as an electronic device below.
[0049] According to one aspect of the embodiments of this application, a method embodiment for training a medical order classification model is provided. Figure 1 A flowchart illustrating a training method for a medical order classification model provided in this application embodiment is shown below. Figure 1 As shown, the method includes:
[0050] Step S11: Obtain the set of medical terms used to train the medical order classification model. The medical term classification set is obtained based on the medical order dictionary and medical entities corresponding to the target hospital.
[0051] The method provided in this application embodiment is applied to a smart terminal capable of data processing. The smart terminal can be a computer, smartphone, tablet, etc. In this application embodiment, step S11, obtaining the set of medical terms used to train the medical order classification model, includes the following steps A1-A4:
[0052] Step A1: Obtain the medical order dictionary and target medical entity corresponding to the target hospital, wherein the medical order dictionary includes at least one medical term.
[0053] In this embodiment, the smart terminal obtains a medical order dictionary corresponding to the target hospital uploaded by the user. The medical order dictionary includes multiple categories of medical terms, such as: symptom-related medical terms, test-related medical terms, diagnostic medical data, etc. Moreover, each type of medical term can include multiple medical terms.
[0054] In this embodiment, after obtaining the medical order dictionary, the smart terminal extracts medical terms from the dictionary and performs entity recognition on the medical terms to obtain target medical entities. These medical entities can include entities related to medicine, such as symptoms, surgery, signs, departments, diseases, medical devices, treatment devices, and drugs. Examples include "fever," "head fever," "gastroenterology," and "minimally invasive surgery."
[0055] Step A2: Store the target medical entity in the database and retrieve candidate medical entities that match the medical terminology from the database.
[0056] In this embodiment, the smart terminal stores the obtained medical entities in a database, and simultaneously retrieves multiple candidate medical entities matching medical terms from all medical entities stored in the database. It should be noted that the medical entities stored in the database are identified through medical terms in the corresponding medical order dictionaries of each hospital. The medical order dictionary corresponding to the target hospital and the current medical entity are obtained, wherein the medical order dictionary includes at least one medical term.
[0057] Specifically, obtaining candidate medical entities from the database includes: calculating the second similarity between medical entities in the database and medical terms included in the medical order dictionary; and identifying N medical entities with a second similarity greater than a second preset similarity as candidate medical entities, where N is an integer greater than 0.
[0058] Step A3: Calculate the first similarity between each medical term and the candidate medical entity, and identify the medical terms with the first similarity less than the first preset similarity as the target medical terms.
[0059] In this embodiment of the application, each medical term t is calculated. dict The process of determining the first similarity with candidate medical entities can be as follows: t is calculated based on the longest common subsequence similarity algorithm. dict The first similarity with the top N candidate medical entities. If there is a candidate medical entity with a first similarity less than a first preset similarity, it is retained and identified as the target medical data. The longest common subsequence similarity is calculated as follows:
[0060] Medical terminology dict and candidate medical entity e train This can be represented as X = (x1, x2, x3, ..., x n ) and Y = (y1, y2, y3..., y m ), where m is greater than 0 and n is greater than 0. X i =(x0,x1,x3,...,x i ) and Y j =(y0,y1,y2,...,y jx0 and y0 are empty characters. 0 ≤ i ≤ n, 0 ≤ j is less than or equal to m.
[0061] The calculation formula is as follows:
[0062]
[0063]
[0064] In the formula, LCS represents the length of the longest common subsequence, and S(X,Y) represents the first similarity between X and Y.
[0065] In this embodiment of the application, a first similarity between each medical term and a candidate medical entity is calculated based on the above formula, and medical terms with a first similarity less than a first preset similarity are identified as target medical terms.
[0066] Step A4: Construct a set of medical terms for training the medical order classification model using the target medical terms.
[0067] In this embodiment of the application, step A4, constructing a set of medical terms for training the medical order classification model using target medical terms, includes the following steps A401-A403:
[0068] Step A401: Randomly select one target medical term from the target medical terms as the first target medical term, and construct a non-homogeneous term set based on the first target medical term.
[0069] Step A402: Obtain non-homogeneous terms corresponding to the first target medical terms from the second target medical terms, wherein the second target medical terms are medical terms other than the first target medical terms in the target medical terms.
[0070] Step A403: Add non-homogeneous terms to the non-homogeneous term set to obtain an updated non-homogeneous term set, and use the updated homogeneous term set as the medical term set for training the medical order classification model.
[0071] In this embodiment of the application, an initial medical terminology set is constructed based on the target medical terminology. A homogeneous data extraction tool is used to group the homogeneous data in the initial medical terminology set. A term is randomly assigned to each group to form a non-homogeneous terminology set, which is the final medical terminology set.
[0072] Specifically, obtaining non-homogeneous terms corresponding to the first target medical term from the second target medical term includes: calculating the cosine similarity between the first target medical term and the second target medical term; and identifying the second target medical term with a cosine similarity less than a preset threshold as the non-homogeneous term corresponding to the first target medical term.
[0073] It should be noted that the homogeneous data extraction tool first obtains the vector representations of two medical terms based on the bag-of-words model of word frequency, and then calculates the similarity score of the two medical terms vectors based on cosine similarity. If the score is higher than the threshold, the two medical terms are considered homogeneous medical terms.
[0074] Step S12: Obtain the annotation information corresponding to each target medical term in the medical terminology set.
[0075] In this embodiment, the smart terminal detects editing operations performed on each target medical term in the medical terminology set, and determines the annotation information corresponding to the target medical term based on the editing operations. It should be noted that the annotation information is used to indicate the category corresponding to the target medical term.
[0076] Step S13: Train the initial medical order classification model using the medical terminology set and annotation information, so that the initial medical order classification model outputs the prediction result corresponding to each target medical term in the medical terminology set.
[0077] In this embodiment, an initial medical order classification model is trained using a medical terminology set and annotation information. Specifically, the terms from the medical terminology set are input into the initial medical order classification model to be trained. The initial medical order classification model extracts terminology features corresponding to each target medical term in the medical terminology set. Terminology features can be terminology keywords, etc. Then, based on the terminology features, the terminology category corresponding to the terminology features is predicted, and the terminology category is determined as the prediction result. Then, the prediction result is matched with the annotation information to determine the training result.
[0078] Step S14: If the prediction results match the annotation information, the initial medical order classification model is determined as the target medical order classification model corresponding to the target hospital.
[0079] In this embodiment, matching the prediction result with the annotation information can be achieved by calculating the similarity between the term category and the term category carried in the annotation information. If the similarity falls within a preset similarity range, the prediction result is determined to match the annotation information. This indicates that the initial medical order classification model has been successfully trained.
[0080] The method provided in this application can train different medical order classification models for different hospitals, enabling each hospital to quickly classify medical order content, thereby solving the problem that manual classification has significant limitations in efficiency and accuracy.
[0081] In the embodiments of this application, such as Figure 2 As shown, the method also includes the following steps B1-B3:
[0082] Step B1: Obtain medical terms that do not match the prediction results and annotation information, perform entity recognition, and obtain an initial entity set.
[0083] Step B2: Based on the medical knowledge graph, obtain the synonyms and subordinate entities of each entity in the entity set. Use the synonyms and subordinate entities of each entity to expand the initial entity set to obtain the target entity set.
[0084] Step B3: Continue training the target medical order classification model using the target entity set.
[0085] In this embodiment, medical terms whose prediction results do not match the annotation information (i.e., medical terms with abnormal predictions) are obtained, and entity recognition is performed on them to obtain the entity corresponding to each medical term with abnormal predictions, thus forming an initial entity set E. badcase The entity set E is obtained based on the medical knowledge graph. badcase Each entity e in badcase The initial entity set is expanded using the synonyms and subordinate entities of each entity to obtain the target entity set E. bc_badcase Finally, the target entity set is used to further train the target medical order classification model.
[0086] In the real-time example of this application, during the continued training process, the system first queries the database for entity e from the target entity set. bc_badcase Given N similar entities, determine whether entity e is included among them. bc_badcase If the string is present, determine whether the category tag of the medical order content containing the string is entity e. bc_badcase If the labeling information is not provided, then the classification label of the similar entity will be changed to entity e. bc_badcase The annotation information is modified to ensure more accurate calculation of the similarity between entities and their parent entities during subsequent training.
[0087] In this embodiment of the application, the target medical order classification model is further trained using the target entity set, including the following steps C1-C2:
[0088] Step C1: Embed pseudo-medical terms generated using the medical order template into the target entity set to obtain the updated target entity set.
[0089] Step C2: Continue training the target medical order classification model using the updated target entity set.
[0090] In this embodiment, the entities in the medical order content are first extracted, such as "appendectomy today". The surgery "appendectomy" is extracted, and the template "today's ____" is generated. Template extraction is performed for each category of the medical order. Then, for each category, the top N templates are retained according to frequency from high to low. Entity e is... bc_badcase Based on the entity category, a pseudo-data entry is generated by fitting the template from the above steps. For example, fitting "radical gastrectomy for gastric cancer" into the surgical template "Today's procedure ____" generates the pseudo-data entry "Today's procedure is radical gastrectomy for gastric cancer," with the category of this pseudo-data entry being "surgery." This pseudo-data entry is then added to the target entity set. The updated target entity set is then used to continue training the target medical order classification model.
[0091] It should be noted that by actively learning, data from new domain data that the old model is prone to making mistakes are selected as test data to guide model iteration; combining medical knowledge graphs to improve the scope of training data error correction and data augmentation; and using template mining and pseudo-data generation methods to effectively increase the size of the training dataset and improve the model's generalization ability in new domains.
[0092] This application also provides the following method, which includes: obtaining medical order information to be identified, the medical order information including: hospital identifier and medical order content; obtaining the medical order classification model corresponding to the current hospital identifier based on the previous correspondence between the preset hospital identifier and the medical order classification model; inputting the medical order content into the medical order classification model so that the medical order classification model detects the medical order features of the medical order content and determines the medical order category corresponding to the medical order features.
[0093] Figure 3 This is a block diagram of a training device for a medical order classification model provided in an embodiment of this application. This device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. Figure 3 As shown, the device includes:
[0094] The first acquisition module 31 is used to acquire the set of medical terms corresponding to the training of the medical order classification model, wherein the set of medical terms classification is obtained based on the medical order dictionary and medical entities corresponding to the target hospital.
[0095] The second acquisition module 32 is used to acquire the annotation information corresponding to each target medical term in the medical terminology set;
[0096] Prediction module 33 is used to train an initial medical order classification model using a medical terminology set and annotation information, so that the initial medical order classification model outputs the prediction result corresponding to each target medical term in the medical terminology set.
[0097] The processing module 34 is used to determine the initial medical order classification model as the target medical order classification model corresponding to the target hospital when the prediction results are matched with the annotation information.
[0098] In this embodiment of the application, the first acquisition module is used to acquire the medical order dictionary and medical entities corresponding to the target hospital, wherein the medical order dictionary includes at least one medical term; store the medical entities in a database, and acquire candidate medical entities that match the medical terms from the database; calculate the first similarity between each medical term and the candidate medical entities, and determine those with the first similarity less than the first preset similarity as target medical terms; and construct a set of medical terms for training the medical order classification model using the target medical terms.
[0099] In this embodiment of the application, the first acquisition module 31 is used to calculate the second similarity between medical entities in the database and medical terms included in the medical order dictionary; and to determine medical entities with a second similarity greater than a second preset similarity as candidate medical entities.
[0100] In this embodiment of the application, the first acquisition module 31 is used to randomly select one target medical term from the target medical terms as the first target medical term, and construct a non-homogeneous term set based on the first target medical term;
[0101] Obtain non-homogeneous terms corresponding to the first target medical terms from the second target medical terms, wherein the second target medical terms are medical terms other than the first target medical terms in the target medical terms; add the non-homogeneous terms to the non-homogeneous term set to obtain an updated non-homogeneous term set, and use the updated homogeneous term set as the medical term set for training the medical order classification model.
[0102] In this embodiment of the application, the first acquisition module 31 is used to calculate the cosine similarity between the first target medical term and the second target medical term; and to determine the second target medical term with a cosine similarity less than a preset threshold as the non-homogeneous term corresponding to the first target medical term.
[0103] In this embodiment, the training device for the medical order classification model further includes: a data optimization module, used to obtain medical terms that do not match the prediction results and annotation information for entity recognition, and obtain an initial entity set; obtain synonyms and subordinate entities of each entity in the entity set based on the medical knowledge graph, expand the initial entity set using the synonyms and subordinate entities of each entity, and obtain a target entity set; and continue to train the target medical order classification model using the target entity set.
[0104] In this embodiment, the data optimization module is used to embed pseudo-medical terms generated using a medical order template into the target entity set to obtain an updated target entity set; and to continue training the target medical order classification model using the updated target entity set.
[0105] This application also provides an electronic device, such as... Figure 4 As shown, the electronic device may include: a processor 1501, a communication interface 1502, a memory 1503, and a communication bus 1504, wherein the processor 1501, the communication interface 1502, and the memory 1503 communicate with each other through the communication bus 1504.
[0106] Memory 1503 is used to store computer programs;
[0107] When the processor 1501 executes the computer program stored in the memory 1503, it implements the steps of the above embodiments.
[0108] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.
[0109] The communication interface is used for communication between the aforementioned terminal and other devices.
[0110] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.
[0111] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0112] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores instructions that, when executed on a computer, cause the computer to perform the training method of the medical order classification model described in any of the above embodiments.
[0113] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute the training method of the medical order classification model described in any of the above embodiments.
[0114] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive).
[0115] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.
[0116] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.
Claims
1. A training method for a medical order classification model, characterized in that, include: Obtain a set of medical terms for training the medical order classification model, wherein the set of medical terms is obtained based on the medical order dictionary and medical entities corresponding to the target hospital; Obtain the annotation information corresponding to each target medical term in the medical terminology set; An initial medical order classification model is trained using the medical terminology set and the annotation information, so that the initial medical order classification model outputs the prediction result corresponding to each target medical term in the medical terminology set; If the prediction result matches the annotation information, the initial medical order classification model is determined as the target medical order classification model corresponding to the target hospital; The method of constructing a medical terminology set for training a medical order classification model using target medical terms includes: Randomly select one target medical term from the target medical terms as the first target medical term, and construct a non-homogeneous term set based on the first target medical term; Obtain non-homogeneous terms corresponding to the first target medical term from the second target medical term, wherein the second target medical term is a medical term other than the first target medical term from the target medical term; The non-homogeneous terms are added to the non-homogeneous term set to obtain an updated non-homogeneous term set, and the updated homogeneous term set is used as the medical term set for training the medical order classification model. The step of obtaining non-homogeneous terms corresponding to the first target medical terms from the second target medical terms includes: Calculate the cosine similarity between the first target medical term and the second target medical term; The second target medical term with a cosine similarity less than a preset threshold is determined as the non-homogeneous term corresponding to the first target medical term; The step of obtaining the set of medical terms used to train the medical order classification model includes: Obtain the medical order dictionary and current medical entity corresponding to the target hospital, wherein the medical order dictionary includes at least one medical term; The current medical entity is stored in the database, and candidate medical entities that match the medical term are retrieved from the database; Calculate the first similarity between each medical term and the candidate medical entity, and identify the medical terms whose first similarity is less than a first preset similarity as the target medical terms; Calculate each medical term The process of determining the first similarity with candidate medical entities is as follows: The similarity is calculated based on the longest common subsequence similarity algorithm. The first similarity with the top N candidate medical entities; if there are candidate medical entities with a first similarity less than a first preset similarity, they are retained and identified as target medical data; the longest common subsequence similarity is calculated as follows: Medical terminology and candidate medical entities Represent as and Where m is greater than 0 and n is greater than 0; and , The null character; 0 ≤ i ≤ n, 0 ≤ j is less than or equal to m; The calculation formula is as follows: In the formula, LCS represents the length of the longest common subsequence, and S(X,Y) represents the first similarity between X and Y; Construct a set of medical terms for training a medical order classification model using target medical terms; The step of retrieving candidate medical entities from the database that match the medical term includes: Calculate the second similarity between medical entities in the database and medical terms included in the medical order dictionary; Medical entities with a second similarity greater than a second preset similarity are identified as candidate medical entities.
2. The method according to claim 1, characterized in that, The method further includes: Entity recognition is performed on medical terms that do not match the prediction results and the annotation information to obtain an initial entity set; Based on the medical knowledge graph, the synonyms and subordinate entities of each entity in the entity set are obtained. The initial entity set is expanded using the synonyms and subordinate entities of each entity to obtain the target entity set. The target medical order classification model is further trained using the target entity set.
3. The method according to claim 2, characterized in that, The step of further training the target medical order classification model using the target entity set includes: By embedding pseudo-medical terms generated using a medical order template into the target entity set, an updated target entity set is obtained. The updated target entity set is used to continue training the target medical order classification model.
4. A training device for a medical order classification model, employing the method described in any one of claims 1 to 3, characterized in that, include: The first acquisition module is used to acquire a set of medical terms for training the medical order classification model, wherein the set of medical terms is obtained based on the medical order dictionary and medical entities corresponding to the target hospital. The second acquisition module is used to acquire the annotation information corresponding to each target medical term in the medical term set; The prediction module is used to train an initial medical order classification model using the medical terminology set and the annotation information, so that the initial medical order classification model outputs the prediction result corresponding to each target medical term in the medical terminology set; The processing module is used to determine the initial medical order classification model as the target medical order classification model corresponding to the target hospital when the prediction result matches the annotation information.
5. A storage medium, characterized in that, The storage medium includes a stored program, wherein the program, when executed, performs the method steps of any one of claims 1 to 3.
6. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, communication interface, and memory communicate with each other through the communication bus; wherein: Memory, used to store computer programs; A processor for executing the method steps of any one of claims 1 to 3 by running a program stored in memory.