Electronic medical record text classification model training method and device, electronic device, and medium

Abstract

The application provides an electronic medical record text classification model training method and device, electronic equipment and medium, and belongs to the technical field of computers. The method comprises the following steps: in the case that the clustering result of the newly added electronic medical record text data set is not one of the plurality of class clusters, performing multiple iteration training on the first classification model based on the electronic medical record text data set which is not one of the plurality of class clusters; in the case that the number of iteration training reaches the iteration number threshold and the model training index in each iteration training is normal, obtaining the second classification model obtained through the last iteration training in the multiple iteration training; and in the case that the model performance index of the second classification model is better than that of the first classification model, updating the first classification model to the second classification model. The electronic medical record text classification model training method provided by the application can improve the transferability and applicability of the classification model, and further improve the accuracy of electronic medical record text classification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and in particular to an electronic medical record text classification model training method and device, electronic equipment and medium. BACKGROUND

[0002] With the development of health information in recent years, health medical big data is formed in the scene of clinical scientific research, health medical service and the like. Among them, the electronic medical record as the main component of the health medical big data plays a significant role. The classification of the electronic medical record can further reduce the time cost of obtaining the medical record of the related disease, so as to further realize the system management of the hospital to the medical record of the patient.

[0003] At present, the classification of the electronic medical record is divided into two kinds, the first kind is electronic medical record classification, that is, the classification of the electronic medical record type of the electronic medical record (EMR) record, the types include admission record, discharge record, operation record and image examination record, etc.; the second kind is the classification of the electronic medical record text, that is, the classification of the text content in the admission record, for example, which can be divided into chief complaint, history of present illness, past history, etc. Among them, the electronic medical record text classification is more important, the classification process usually adopts a machine learning model, has high dependence on training data, poor migration and stability, and for complex and variable electronic medical record texts, for example, the daily record of a hospital can contain admission record and course record texts, which cannot guarantee the accuracy of the classification result of the electronic medical record text. SUMMARY

[0004] The present application provides an electronic medical record text classification model training method and device, electronic equipment and medium, to solve the problem of poor migration and stability of the electronic medical record text classification model in the prior art, resulting in poor classification result accuracy.

[0005] The present application provides an electronic medical record text classification model training method, comprising:

[0006] In the network text content related to the disease obtained based on the crawler program or big data, the text information with a similarity greater than or equal to a similarity threshold value to the text content contained in the first electronic medical record text data set is screened, and a network text data set is constructed;

[0007] Based on the target key information and the preset electronic medical record template, an extended electronic medical record text data set is generated, the target key information is determined based on the first electronic medical record text data set, wherein the target key information includes the association between the disease type and the patient's age, the association between the disease symptoms and the patient's environment, the association between the treatment method and the disease type, and the association between the diagnosis basis and the disease type.

[0008] performing K-Means clustering processing on the first electronic medical record text data set and the extended electronic medical record text data set to obtain electronic medical record text data of multiple class clusters;

[0009] In a case where a clustering result of the newly added electronic medical record text data set is not one of the multiple class clusters, performing multiple iteration training on a first classification model based on the electronic medical record text data set which is not one of the multiple class clusters; wherein the first classification model is obtained by training based on the first electronic medical record text data set;

[0010] In a case where a number of iteration training reaches an iteration threshold and a model training index in each iteration training is normal, obtaining a second classification model obtained by last iteration training in the multiple iteration training;

[0011] In a case where a model performance index of the second classification model is better than that of the first classification model, updating the first classification model to the second classification model, and the second classification model is used for text classification on an electronic medical record text to be classified.

[0012] In some embodiments, before the multiple iteration training on the first classification model based on the electronic medical record text data set which is not one of the multiple class clusters, the method further comprises:

[0013] performing preprocessing on an initial electronic medical record text data set to obtain a preprocessed initial electronic medical record text data set;

[0014] performing classification annotation on the preprocessed initial electronic medical record text data set to obtain the first electronic medical record text data set;

[0015] training an initial classification model based on the first electronic medical record text data set to obtain the first classification model.

[0016] In some embodiments, the multiple iteration training on the first classification model based on the electronic medical record text data set which is not one of the multiple class clusters comprises:

[0017] performing classification annotation on the electronic medical record text data set which is not one of the multiple class clusters to obtain a second electronic medical record text data set;

[0018] In a case where a data amount of the second electronic medical record text data set exceeds a quantity threshold, performing the multiple iteration training on the first classification model based on the second electronic medical record text data set and the first electronic medical record text data set.

[0019] In some embodiments, after the first classification model is updated to the second classification model, the method further comprises:

[0020] inputting the electronic medical record text to be classified into the second classification model for text classification to obtain classification result data of the electronic medical record text to be classified output by the second classification model;

[0021] based on the classification result data, judging the accuracy of the classification result data;

[0022] in the case where the accuracy of the classification result data is greater than or equal to the accuracy threshold and the similarity between the classification result data and the training data of the second classification model is less than the similarity threshold, determining the classification result data as the newly added electronic medical record text data set.

[0023] In some embodiments, after the second classification model obtained in the last iteration training in the plurality of iteration trainings is obtained, the method further comprises:

[0024] in the case where the model performance index of the second classification model is not better than the second classification model, classifying the electronic medical record text to be classified based on the first classification model.

[0025] In some embodiments, after the first classification model is trained for a plurality of iterations, the method further comprises:

[0026] in the case where the model training index of the classification model obtained in any iteration training in the plurality of iteration trainings is abnormal, stopping the plurality of iteration trainings;

[0027] in the case where the plurality of iteration trainings are stopped, classifying the electronic medical record text to be classified based on the first classification model.

[0028] The present application also provides an electronic medical record text classification model training device, comprising:

[0029] a data acquisition module, configured to filter, from network text content related to diseases obtained based on a crawler program or big data, text information having a similarity greater than or equal to a similarity threshold to text content contained in the first electronic medical record text data set, and construct a network text data set;

[0030] a data expansion module, configured to generate an expanded electronic medical record text dataset based on target key information and a preset electronic medical record template, the target key information being determined based on the first electronic medical record text dataset, wherein the target key information comprises an association between a disease type and a patient age, an association between a disease symptom and an environment in which the patient is located, an association between a treatment method and a disease type, and an association between a diagnosis basis and a disease type;

[0031] a data clustering module, configured to perform K-Means clustering processing on the first electronic medical record text dataset and the expanded electronic medical record text dataset to obtain a plurality of clusters of electronic medical record text data;

[0032] a model training module, configured to, in a case where a clustering result of a newly added electronic medical record text dataset is not one of the plurality of clusters, perform multiple iteration training on a first classification model based on the electronic medical record text dataset that is not one of the plurality of clusters, the first classification model being trained based on the first electronic medical record text dataset;

[0033] a model obtaining module, configured to, in a case where a number of iteration training reaches an iteration number threshold and a model training index in each iteration training is normal, obtain a second classification model obtained in a last iteration training of the multiple iteration training;

[0034] a model updating module, configured to, in a case where a model performance index of the second classification model is better than that of the first classification model, update the first classification model to the second classification model, the second classification model being used for text classification on an electronic medical record text to be classified.

[0035] The present application also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the electronic medical record text classification model training method according to any one of the above when executing the program.

[0036] The present application also provides a non-transitory computer readable storage medium, having a computer program stored thereon, wherein the computer program is executable by a processor to implement the electronic medical record text classification model training method according to any one of the above.

[0037] The present application also provides a computer program product, comprising a computer program, wherein the computer program is executable by a processor to implement the electronic medical record text classification model training method according to any one of the above.

[0038] The electronic medical record text classification model training method, device, electronic equipment and medium provided by the application, by performing K-Means clustering processing on the first electronic medical record text data set and the extended electronic medical record text data set, obtaining electronic medical record text data of multiple class clusters, using an electronic medical record text data set that is not one of the multiple class clusters to automatically iteratively train the first classification model, obtaining a second classification model, reducing the processing amount of the electronic medical record text data, reducing the amount of manual participation in establishing the classification model, and being able to timely classify the electronic medical record text according to the second classification model after iteration, being adaptable to complex medical text classification scenarios, thereby improving the transferability of the classification model, having good universality, and greatly improving the accuracy of electronic medical record text classification and recognition, and being applicable to classification and recognition of various types of complex electronic medical record texts, reducing the cost of manual classification and optimizing the cost of the classification model. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0040] Figure 1 is one of the flowcharts of the electronic medical record text classification model training method provided by the application;

[0041] Figure 2 is a structural schematic diagram of the electronic medical record text classification system provided by the application;

[0042] Figure 3 is the second flowchart of the electronic medical record text classification model training method provided by the application;

[0043] Figure 4 is a structural schematic diagram of the electronic medical record text classification model training device provided by the application;

[0044] Figure 5 is a structural schematic diagram of the electronic equipment provided by the application. DETAILED DESCRIPTION

[0045] In order to make the objects, technical solutions and advantages of the application more clear, the technical solutions in the application will be described clearly and completely below with reference to the drawings in the application. Obviously, the described embodiments are some embodiments of the application, but not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0046] The application provides an electronic medical record text classification model training method, device, electronic equipment and medium. Figures 1-5 The application provides an electronic medical record text classification model training method, device, electronic equipment and medium.

[0047] Figure 1 FIG. 1 is a flowchart of an electronic medical record text classification model training method provided by the application. Referring to FIG. 1, the electronic medical record text classification model training method provided by the application comprises the following steps. Figure 1

[0048] Step 110: in the network text content related to diseases obtained based on a crawler program or big data, text information with a similarity greater than or equal to a similarity threshold value to the text content contained in the first electronic medical record text data set is screened, and a network text data set is constructed.

[0049] Step 120: an extended electronic medical record text data set is generated based on target key information and a preset electronic medical record template, the target key information is determined based on the first electronic medical record text data set, and the target key information comprises an association relationship between a disease type and a patient age, an association relationship between a disease symptom and an environment in which a patient is located, an association relationship between a treatment method and a disease type, and an association relationship between a diagnosis basis and a disease type.

[0050] Step 130: the first electronic medical record text data set and the extended electronic medical record text data set are subjected to K-Means clustering processing, and electronic medical record text data of multiple class clusters is obtained.

[0051] Step 140: in a case where the clustering result of the newly added electronic medical record text data set is not one of the multiple class clusters, the first classification model is subjected to multiple iteration training based on the electronic medical record text data set which is not one of the multiple class clusters; the first classification model is trained based on the first electronic medical record text data set.

[0052] Step 150: in a case where the number of iteration training reaches an iteration number threshold value and the model training index in each iteration training is normal, a second classification model obtained in the last iteration training in the multiple iteration training is acquired.

[0053] Step 160: in a case where the model performance index of the second classification model is better than that of the first classification model, the first classification model is updated to the second classification model, and the second classification model is used for text classification on an electronic medical record text to be classified.

[0054] ​It should be noted that the execution subject of the electronic medical record text classification model training method provided by the present invention can be an electronic device, a component in an electronic device, an integrated circuit, or a chip. The electronic device can be a mobile electronic device or a non-mobile electronic device. For example, the mobile electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, a wearable device, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc. The non-mobile electronic device can be a server, a network attached storage (NAS), a personal computer (PC), a television (TV), a teller machine, or a self-service machine, etc., and the present invention does not make specific limitations.

[0055] The following takes the example of a computer executing the electronic medical record text classification model training method provided by the present invention to explain the technical solution of the present invention in detail.

[0056] In related technologies, electronic medical records contain a wealth of medical knowledge, including disease names, symptom descriptions, treatment plans, surgical procedures, imaging examinations, and demographic information during a patient's diagnosis and treatment. Data can be divided into structured and unstructured data based on its composition. Structured data can be directly used for medical research, but because structured fields are predefined, they contain less medical information or knowledge and are therefore less commonly used in clinical research. Unstructured data, on the other hand, is mostly textual data that records or summarizes the clinical diagnosis and treatment process. It covers a wide range of content, is highly timely, has high research value, and implicitly contains semantic information such as correlation, parallelism, or causality, playing a crucial supporting role in medical research.

[0057] Due to the wide variety of medical information systems and the lack of unified data recording standards across them, extracting entities and relationships from this data can cause problems for subsequent applications. For example, some electronic medical record texts are not marked as admission records, discharge records, or surgical nursing records. If you want to find medical records of people who have recovered from lung cancer in surgical records, without classifying the electronic medical records, the extracted data will have different meanings and will not be able to provide effective support for further clinical data mining, such as disease-related factors, exploring potential connections between diseases and other factors, and disease treatment plans and prognosis relationships. Therefore, before conducting structured data governance, it is necessary to classify the electronic medical record text.

[0058] In the present invention, the first electronic medical record text data set may be a large amount of historical electronic medical record text data obtained from a medical information system;

[0059] In actual implementation, the historical electronic medical record text data can be expanded. The expansion manner is as follows:

[0060] (1) part of the historical electronic medical record text data is preprocessed, and part of the historical electronic medical record text data is not preprocessed;

[0061] (2) data expansion is performed by using big data or a crawler program as mentioned in step 110. The objects collected by the big data or the crawler program include but are not limited to: papers, cases, information about health or some medical related application programs, and other disease related network text contents.

[0062] Taking the crawler program as an example, the received crawler task is a disease related topic, and a uniform resource locator (URL) task queue is initialized; the URL task queue stores URL data to be crawled. A crawler task is taken out from the URL task queue, and disease related network text content is obtained.

[0063] Taking big data as an example, in the disease related big data, a vector similarity algorithm is used to obtain electronic medical record related network text content.

[0064] Then, in the obtained network text content, text information with a similarity to the data in the first electronic medical record data set exceeding a similarity threshold can be obtained, so as to construct a network text data set. The similarity calculation can use a vector similarity algorithm, and the specific algorithm includes but is not limited to a cosine similarity algorithm, a Euclidean distance algorithm or a Pearson correlation coefficient algorithm.

[0065] In step 120, semantic analysis is performed on the first electronic medical record text data set to determine the association relationship between different data and obtain target key information. The target key information can include the association relationship between the disease type and the patient age, the association relationship between the disease symptoms and the environment where the patient is located, the association relationship between the treatment method and the disease type, and the association relationship between the diagnosis basis and the disease type. The target key information can also include the association relationship between the disease type and the patient gender.

[0066] For example, the association between disease type and patient age can be manifested as: the disease types of infants can include jaundice, pediatric pneumonia, or hand-foot-mouth disease, etc.; the disease types of adolescents can include bone tumors, and the disease types of the elderly can include myocardial infarction or osteoporosis, etc. The association between disease symptoms and the environment in which the patient is located can be manifested as: mining personnel living in a mining area are prone to pneumoconiosis between the ages of 35-45, white-collar workers living in an industrial park are prone to lumbar disc herniation after the age of 25, etc. The association between treatment methods and disease types can be manifested as: essential amino acid therapy, maintaining positive nitrogen balance, correcting water, electrolyte disorders, and acidosis, expansion diuresis, dialysis treatment, or kidney transplantation, etc. are treatment methods for chronic renal failure. The association between diagnosis basis and disease type can be manifested as: sudden onset after strenuous activity, metastatic right lower abdominal pain, mild nausea, right lower abdominal appendix point fixation and obvious tenderness, mild abdominal wall muscle tension and rebound pain, etc. can be considered as appendicitis. The above-mentioned association is only used as an example.

[0067] In actual execution, based on the target key information and the preset electronic medical record template, an extended electronic medical record text data set can be generated. The preset electronic medical record template can be generated according to the template elements selected by the user, and the template elements are generated according to the preset electronic medical record template format. The relevant content of the target key information can be filled in the preset electronic medical record template, so that the electronic medical record text data can be generated, and then the extended electronic medical record text data set can be obtained.

[0068] In step 130, in the case of a large number of electronic medical record text data, the calculation amount of text classification is large; therefore, the K-means algorithm is used for clustering first; the first electronic medical record text data set and the extended electronic medical record text data set are clustered to obtain electronic medical record text data of multiple clusters;

[0069] The K-means algorithm (K-means clustering algorithm, K-means) is a clustering analysis algorithm for iterative solution. The steps are as follows: the first electronic medical record text data set and the extended electronic medical record text data set are divided into K groups, then K electronic medical record text data are randomly selected as initial cluster centers, then the distance between each electronic medical record text data and each seed cluster center is calculated, and each electronic medical record text data is assigned to the cluster center closest to it. The cluster centers and the electronic medical record text data assigned to them represent a cluster. The cluster centers of the cluster are recalculated according to the existing objects in the cluster after each assignment of an electronic medical record text data. This process will be repeated until a certain termination condition is met. The termination condition can be that no (or a minimum number of) objects are reassigned to different clusters, no (or a minimum number of) cluster centers change, or the error sum of squares is locally minimized.

[0070] In step 140, in the case that the clustering result of the newly added electronic medical record text data set is not one of the multiple clusters, a first classification model is trained based on the electronic medical record text data set which is not one of the multiple clusters.

[0071] In some embodiments, before step 140, the electronic medical record text classification model training method further comprises:

[0072] The initial electronic medical record text data set is preprocessed to obtain a preprocessed initial electronic medical record text data set.

[0073] The preprocessed initial electronic medical record text data set is classified and labeled to obtain a first electronic medical record text data set.

[0074] Based on the first electronic medical record text data set, the initial classification model is trained to obtain the first classification model.

[0075] In actual execution, the initial electronic medical record text data set can be de-identified electronic medical record text data collected from a public data set, or electronic medical record text data of any patient obtained from a medical information system. The content of the electronic medical record text data is not specifically limited by the present application.

[0076] Then, based on the pre-stored data preprocessing function, the initial electronic medical record text data set is preprocessed to obtain a preprocessed initial electronic medical record text data set.

[0077] The preprocessing can include at least one of the following: removing special tags, removing special symbols, Extensible Markup Language (XML) format parsing, etc.

[0078] Removing special tags means removing XML format tags in the text. For example, the XML format tag is <field label="smoking history">, and after removing the tag, "smoking history" is generated.

[0079] Removing special symbols means removing meaningless identifiers in the electronic medical record text to make the text content more standardized, including removing redundant line breaks, spaces, etc.

[0080] XML parsing refers to parsing electronic medical record texts stored in XML format. Because the XML formats used in electronic medical record texts in different electronic medical record systems are different, XML parsing of electronic medical record texts has the problem of low accuracy of parsing results. Based on the experience of managing medical system data, the system will internally build XML format templates of different electronic medical record system suppliers to improve the accuracy of XML parsing. For cases where the accuracy is too low, subsequent manual correction, bidding or adding templates are used to improve the accuracy of preprocessing.

[0081] Examples of electronic medical record texts in XML format:

[0082] The structured document content is: "{smoking history: smoking age [5] years,

[10] cigarettes per day.}"

[0083] The XML format is as follows:

[0084] <field label="smoking history">

[0085] <field label="smoking age" unit="years">5

[0086] <text>undefined< / text>

[0087] <field label="per day" id="number of cigarettes smoked per day" unit="cigarettes">10

[0088] <text> .< / text>

[0089]

[0090] After obtaining the preprocessed initial electronic medical record text data set, the preprocessed initial electronic medical record text data set is classified and labeled. The process of classification and labeling is as follows:

[0091] First, the preprocessed initial electronic medical record text data set is labeled with different electronic medical record types. Different electronic medical record types include: admission record, consultation record, discharge record, death record, nursing record, daily progress record or labor record, etc.

[0092] The labeling of electronic medical record types does not affect the subsequent text classification and labeling. The electronic medical record type labeling performed here is mainly to record the source of the text classification model training data.

[0093] Further, the preprocessed initial electronic medical record text data set can be subjected to paragraph type labeling according to a word library related to the content of the electronic medical record text paragraph, for example, the text labeling types corresponding to the admission record type electronic medical record record include chief complaint, history of present illness, past history, menstrual history, marital history, family history, personal history, and respective corresponding time, etc.

[0094] It can be understood that different electronic medical record types correspond to different text labeling types, which can be determined according to actual conditions, and the present application does not make specific limitations here.

[0095] After the classification labeling of the preprocessed initial electronic medical record text data set is completed, a first electronic medical record text data set is obtained;

[0096] Then, the pre-constructed initial classification model is trained based on the first electronic medical record text data set to obtain a first classification model.

[0097] The pre-constructed initial classification model can be a classification model algorithm constructed based on FastText, TextCNN, BiLSTM, etc. neural network, or a classification model algorithm constructed based on Transformer, BERT pretrain-finetune and prompt paradigm, etc. The network structure of the initial classification model can be constructed according to actual conditions, and the present application does not make specific limitations here.

[0098] In step 140, the accuracy of the first classification model can be evaluated in real time, and new electronic medical record text data sets are continuously collected. For example, the similarity between the new electronic medical record text data set and the classified and labeled first electronic medical record text data set can be matched, when the similarity is lower than the similarity threshold, it is determined that the new electronic medical record text data set is valid data, and the valid data is labeled, the labeled valid data and the first electronic medical record text data set together constitute new sample data, which is used to iteratively train the first classification model. When the similarity is higher than the similarity threshold, it is determined that the new electronic medical record text data set is invalid data. The similarity threshold can be set according to actual needs, and the present application does not make specific limitations here.

[0099] It can be understood that the first classification model is an initial electronic medical record text classification model iteratively trained.

[0100] In some embodiments, step 140 can include:

[0101] The electronic medical record text data set that is not one of the plurality of class clusters is classified and labeled to obtain a second electronic medical record text data set;

[0102] In the case that the data quantity of the second electronic medical record text dataset exceeds the quantity threshold, the first classification model is iteratively trained based on the second electronic medical record text dataset and the first electronic medical record text dataset.

[0103] In the case that the data quantity of the second electronic medical record text dataset exceeds the quantity threshold, the first classification model is iteratively trained based on the second electronic medical record text dataset and the first electronic medical record text dataset.

[0104] Then, the data quantity of the classified second electronic medical record text dataset is monitored in a timely manner. In the case that the data quantity of the second electronic medical record text dataset exceeds the quantity threshold, the second electronic medical record text dataset and the first electronic medical record text dataset are imported to iteratively train the first classification model. The quantity threshold and the number of iterations in a single iteration training can be set according to actual requirements, which are not specifically limited in the present application.

[0105] The iterative training refers to that the existing classification model is trained again according to the new labeled dataset as long as the newly added electronic medical record text dataset meets the similarity condition and the data quantity condition.

[0106] In step 150, the iterative training process of the first classification model can be monitored,

[0107] In actual execution, the iterative training process is monitored based on the log of the iterative training process. The model training indicators recorded in the log of the iterative training process include at least one of the following: the number of iterations, the current accuracy, the learning rate, the missed detection rate, and the false detection rate.

[0108] Optionally, the monitoring program can read the log of the iterative training process in a timely manner, obtain the number of iterations, the current accuracy, the learning rate, the missed detection rate, the false detection rate, and other model training indicators, and determine whether the current accuracy, the learning rate, the missed detection rate, and the false detection rate are abnormal and whether the number of model iterations is greater than the iteration threshold.

[0109] The learning rate can control the step size of adjusting the model weight parameter in each iteration training, so that the loss function of the classification model converges to a minimum value.

[0110] The accuracy is the percentage of the correct electronic medical record text data quantity in the classification result to the electronic medical record text data quantity input in the training process.

[0111] The missed detection rate is the percentage of the unclassified electronic medical record text data quantity to the electronic medical record text data quantity input in the training process.

[0112] The false detection rate is the percentage of the amount of electronic medical record text data with a wrong classification result in the amount of electronic medical record text data input into the model during the training process.

[0113] In this step, when the number of times of iterative training of the model is greater than the iteration threshold and the model training indicators in each iterative training are normal, the iterative training of the model ends, and a new version of the classification model obtained in the last iteration training, i.e., the second classification model, can be acquired.

[0114] In some embodiments, after step 140, the electronic medical record text classification model training method further includes:

[0115] In the case that the model training indicators of the classification model acquired in any one of the multiple iterative trainings are abnormal, the multiple iterative trainings are stopped.

[0116] In the case that the multiple iterative trainings are stopped, the first classification model is used to classify the electronic medical record text to be classified.

[0117] In actual execution, if the model training indicators of any one of the multiple iterative trainings are abnormal, i.e., one or more of the current accuracy, learning rate, false omission rate and false detection rate are abnormal, it is determined that the iterative training of the model is abnormal, and the iterative training process is terminated, i.e., the iterative training of the model ends.

[0118] When it is determined that the iterative process is abnormal, the new version of the classification model is not released, and the original first classification model is used to classify the electronic medical record text to be classified.

[0119] In step 160, the effects of the new version of the model and the previous version of the model are verified, i.e., it is determined whether the model performance indicators of the second classification model are better than those of the first classification model.

[0120] In the case that the model performance indicators of the second classification model are better than those of the first classification model, the second classification model is released, the first classification model is updated to the second classification model, and then the second classification model can be used to classify the electronic medical record text to be classified.

[0121] In actual execution, the model performance indicators can include but are not limited to model performance gain, and in the case that the model performance gain exceeds the gain threshold, it is determined that the model performance indicators of the second classification model are better than those of the first classification model. Of course, the model performance indicators can be other parameters, which are not specifically limited in the present application.

[0122] In some embodiments, after step 150, the electronic medical record text method further includes:

[0123] In the case that the model performance indicator of the second classification model is not superior to the second classification model, the first classification model is used to classify the electronic medical record text to be classified.

[0124] In actual execution, if the model performance indicator of the second classification model is not superior to the second classification model, the iterative training process is terminated, i.e., the model iterative training is completed.

[0125] When the model performance indicator of the second classification model is not superior to the second classification model, the new version of the classification model is not released, and the original first classification model is used to classify the electronic medical record text to be classified.

[0126] It can be understood that after obtaining the second classification model, the second classification model can be iteratively trained according to the newly added electronic medical record text data set. When the new version of the classification model obtained after iteration is superior to the classification model released in the last version, the new version of the model is released, and then the electronic medical record text to be classified is classified based on the new version of the classification model.

[0127] In some embodiments, after step 160, the electronic medical record text method further comprises:

[0128] The electronic medical record text to be classified is input into the second classification model for text classification to obtain classification result data of the electronic medical record text to be classified output by the second classification model;

[0129] Based on the classification result data, the accuracy of the classification result data is determined.

[0130] In the case that the accuracy of the classification result data is greater than or equal to the accuracy threshold and the similarity between the classification result data and the training data of the second classification model is less than the similarity threshold, the classification result data is determined as the newly added electronic medical record text data set.

[0131] In actual execution, the electronic medical record text to be classified is input into the second classification model for text classification to obtain classification result data of the electronic medical record text to be classified output by the second classification model.

[0132] The classification result data can include paragraph splitting results of the electronic medical record text, specifically, the content and corresponding type of the relevant paragraph in the electronic medical record text, and the starting position and ending position of the relevant paragraph.

[0133] The accuracy of the classification result data is the percentage of the amount of electronic medical record data with correct classification result data in the total amount of electronic medical record data to be classified.

[0134] In a case where the accuracy of the classification result data is greater than or equal to the accuracy threshold and the similarity between the classification result data and the training data of the second classification model is less than the similarity threshold, the classification result data is determined as new sample data, that is, a newly added electronic medical record text data set.

[0135] Then the newly added electronic medical record text data set can be classified and labeled to form a new classified and labeled electronic medical record text data set for subsequent iterative training of the classification model.

[0136] The electronic medical record text classification model training method provided by the application reduces the processing amount of electronic medical record text data, reduces the amount of manual participation in establishing a classification model, updates model training data in a timely manner, and can classify electronic medical record text in a timely manner according to the second classification model after iteration, which is suitable for complex medical text classification scenarios, thereby improving the transferability of the classification model and having good universality. The accuracy of electronic medical record text classification and recognition is greatly improved, and the method can be applied to the classification and recognition of various types of complex electronic medical record text, reducing the cost of manual classification and optimizing the cost of the classification model.

[0137] The electronic medical record text classification system provided by the application is described below.

[0138] Figure 2 is a structural schematic diagram of the electronic medical record text classification system provided by the application. Referring to Figure 2 The electronic medical record text classification system 200 provided by the application includes an electronic medical record data acquisition module 210, an electronic medical record preprocessing module 220, an electronic medical record text classification module 230, a sampling quality inspection labeling module 240, and a model iteration module 250.

[0139] The electronic medical record data acquisition module 210 is configured to connect a database, acquire electronic medical record text to be classified, and perform data input.

[0140] The electronic medical record preprocessing module 220 is configured to perform preprocessing operations on the original electronic medical record text data set. The preprocessing operations can include removing special tags, removing special symbols, XML format analysis, and HTML format analysis.

[0141] The electronic medical record text classification module 230 is configured to call an electronic medical record text classification model and output classification result data, that is, a paragraph splitting result of the electronic medical record text, for example, including the type of a relevant paragraph in the electronic medical record text and the starting and ending positions of the paragraph.

[0142] The sampling quality inspection and labeling module 240 is configured to perform quality inspection on the classification result data obtained after text classification, and determine the accuracy of the classification result data. In the case that the accuracy of the classification result data is higher than the accuracy threshold X1 and the similarity of the classification result data to the training data of the current classification model is lower than the similarity threshold X2, the classification result data is taken as new sample data, the new sample data is classified and labeled to form a new labeled data set, and the new labeled data set is used for subsequent iterative training of the model.

[0143] The model iteration module 250 is configured to determine the number of the new labeled data set, and when the data volume is higher than the data volume threshold, the new labeled data set is automatically included in the training data set, the classification model is iteratively trained, and model training monitoring and model publishing are performed.

[0144] The model training monitoring refers to monitoring the model training indicators in the model iterative training process. When the model monitoring is abnormal, the model training is ended. When the model monitoring is not abnormal, the iteration number of the model is obtained, and when the iteration number is greater than the iteration number threshold, the algorithm model iteration is ended, the new version of the classification model obtained after iteration is compared with the last version of the classification model published, and when the model performance indicators of the new version of the classification model are better than those of the last version of the classification model, the new version of the classification model is published.

[0145] Figure 3 FIG. 2 is a flowchart of a method for training an electronic medical record text classification model according to an embodiment of the present application. The method comprises the following steps: Figure 3 The electronic medical record text classification model training method provided by the present application can be applied to the electronic medical record text classification system described above, and the method comprises the following steps:

[0146] Step 301: Preprocessing the original electronic medical record text data set;

[0147] Step 302: Classifying and labeling the original electronic medical record text data set;

[0148] Step 303: Generating an initial electronic medical record text classification model according to the labeled data set;

[0149] Step 304: Extending the original electronic medical record text data set based on big data or a crawler program to obtain extended electronic medical record text data. Then, the original electronic medical record text data set and the extended electronic medical record text data set are subjected to K-Means clustering processing to obtain electronic medical record text data of multiple class clusters; the electronic medical record text data set that is not in one of the multiple class clusters is taken as new electronic medical record text data set, and is labeled to generate a new labeled data set;

[0150] Step 305, the data amount of the newly labeled data set is monitored, and it is judged whether the data amount of the newly labeled data set is greater than the data amount threshold X3; when the data amount of the newly labeled data set is greater than the data amount threshold X3, step 306 is entered; when the data amount of the newly labeled data set is less than the data amount threshold X3, step 304 is returned to;

[0151] Step 306, the newly labeled data set and the original labeled data set are imported, and model iterative training is performed;

[0152] Step 307, the model training process is monitored, and it is judged whether the model training is abnormal; when the model training is not abnormal, step 308 is entered; when the model training is abnormal, step 309 is entered;

[0153] Step 308, the iteration number of the training model is acquired in real time, and it is judged whether the iteration number is greater than the threshold X4; when the iteration number is greater than the iteration number threshold X4, step 309 is entered; when the iteration number is less than the iteration number threshold X4, step 306 is returned to continue model iterative training;

[0154] Step 309, the model iteration is ended;

[0155] Step 310, the effect of classification of the new version model and the last version model is verified, and it is judged whether the model performance index of the new version model is better than that of the last version model;

[0156] Step 311, when the new version model is better than the last version model published before, the new version model is published to the system for use. When the new version model is not better than the last version model published before, the last version model is used for text classification.

[0157] The application provides an electronic medical record text classification model training method and an electronic medical record text classification system based on automatic iteration for the electronic medical record text classification scene, provides a classification model algorithm that can be automatically iterated, can be applied to various types of complex electronic medical record texts, improves the accuracy of electronic medical record classification and recognition, thereby reducing the cost of artificial classification and the cost of model optimization, and has good universality and migration.

[0158] The electronic medical record text classification model training device provided by the application is described below, and the electronic medical record text classification model training device described below can be correspondingly referred to the electronic medical record text classification model training method described above.

[0159] Figure 4 It is a structural schematic view of the electronic medical record text classification model training device provided by the application. Referring to Figure 4 , the electronic medical record text classification model training device provided by the application comprises:

[0160] The data acquisition module 410 is configured to filter, from the disease-related network text content based on a crawler program or big data acquisition, text information having a similarity greater than or equal to a similarity threshold to text content contained in the first electronic medical record text dataset, and construct a network text dataset.

[0161] The data expansion module 420 is configured to generate an expanded electronic medical record text dataset based on target key information and a preset electronic medical record template, the target key information being determined based on the first electronic medical record text dataset, wherein the target key information includes an association between a disease type and a patient age, an association between a disease symptom and an environment in which a patient is located, an association between a treatment method and a disease type, and an association between a diagnosis basis and a disease type.

[0162] The data clustering module 430 is configured to perform K-Means clustering processing on the first electronic medical record text dataset and the expanded electronic medical record text dataset to obtain a plurality of clusters of electronic medical record text data.

[0163] The model training module 440 is configured to, in a case where a clustering result of a newly added electronic medical record text dataset is not one of the plurality of clusters, perform multiple times of iterative training on a first classification model based on the electronic medical record text dataset that is not one of the plurality of clusters, the first classification model being trained based on the first electronic medical record text dataset.

[0164] The model acquisition module 450 is configured to, in a case where a number of times of iterative training reaches an iteration threshold and a model training index in each time of iterative training is normal, acquire a second classification model obtained in a last time of the multiple times of iterative training.

[0165] The model updating module 460 is configured to, in a case where a model performance index of the second classification model is better than that of the first classification model, update the first classification model to the second classification model, the second classification model being used for text classification on a to-be-classified electronic medical record text.

[0166] The electronic medical record text classification model training device provided by the application reduces the processing amount of electronic medical record text data, reduces the amount of manual participation in establishing a classification model, and can classify electronic medical record text in a timely manner according to the second classification model after iteration, adapt to complex medical text classification scenarios, thereby improving the transferability of the classification model and having good universality; the accuracy of electronic medical record text classification and identification is greatly improved, and the device can be applied to the classification and identification of various types of complex electronic medical record text, thereby reducing the cost of manual classification and optimizing the cost of the classification model.

[0167] In some embodiments, the device further comprises:

[0168] The preprocessing module is configured to preprocess an initial electronic medical record text data set to obtain a preprocessed initial electronic medical record text data set before the first classification model is trained by multiple iterations based on the newly added electronic medical record text data set and the first electronic medical record text data set.

[0169] The preprocessed initial electronic medical record text data set is classified and labeled to obtain the first electronic medical record text data set.

[0170] The initial classification model is trained based on the first electronic medical record text data set to obtain the first classification model.

[0171] In some embodiments, the model training module 410 is specifically configured to:

[0172] The newly added electronic medical record text data set is classified and labeled to obtain a second electronic medical record text data set.

[0173] When the data amount of the second electronic medical record text data set exceeds a quantity threshold, the first classification model is trained by multiple iterations based on the second electronic medical record text data set and the first electronic medical record text data set.

[0174] In some embodiments, the device further comprises:

[0175] The first classification module is configured to input the electronic medical record text to be classified into the second classification model for text classification after the first classification model is updated to the second classification model, to obtain classification result data of the electronic medical record text to be classified output by the second classification model.

[0176] A judgment module, configured to judge the accuracy of the classification result data based on the classification result data;

[0177] When the accuracy of the classification result data is greater than or equal to the accuracy threshold and the similarity between the classification result data and the training data of the second classification model is less than the similarity threshold, the classification result data is determined as the newly added electronic medical record text data set.

[0178] In some embodiments, the apparatus further comprises:

[0179] The second classification module is used to classify the electronic medical record text to be classified based on the first classification model after obtaining the second classification model obtained by the last iterative training in the multiple iterative trainings, if the model performance index of the second classification model is not better than that of the second classification model.

[0180] In some embodiments, the model updating module 430 is further configured to:

[0181] After the first classification model is subjected to multiple iterative trainings, if a model training indicator of the classification model obtained in any iterative training in the multiple iterative trainings is abnormal, stopping the multiple iterative trainings;

[0182] When the multiple iterative trainings are stopped, the electronic medical record text to be classified is classified based on the first classification model.

[0183] Figure 5 An example of a physical structure diagram of an electronic device is shown below. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the electronic medical record text classification model training method, which includes:

[0184] In the network text content related to the disease obtained based on the crawler program or big data, text information having a similarity greater than or equal to a similarity threshold with the text content included in the first electronic medical record text dataset is screened to construct a network text dataset;

[0185] generate an extended electronic medical record text dataset based on target key information and a preset electronic medical record template, the target key information being determined based on the first electronic medical record text dataset, wherein the target key information comprises an association relationship between a disease type and a patient age, an association relationship between a disease symptom and an environment in which the patient is located, an association relationship between a treatment method and a disease type, and an association relationship between a diagnosis basis and a disease type;

[0186] perform K-Means clustering processing on the first electronic medical record text dataset and the extended electronic medical record text dataset to obtain electronic medical record text data of a plurality of clusters;

[0187] In a case where a clustering result of the newly added electronic medical record text dataset is not one of the plurality of clusters, perform multiple iteration training on a first classification model based on the electronic medical record text dataset that is not one of the plurality of clusters, wherein the first classification model is trained based on the first electronic medical record text dataset;

[0188] In a case where a number of iteration training reaches an iteration threshold and a model training index in each iteration training is normal, obtain a second classification model obtained in a last iteration training in the multiple iteration training;

[0189] In a case where a model performance index of the second classification model is better than the second classification model, update the first classification model to the second classification model, and the second classification model is used for text classification on an electronic medical record text to be classified.

[0190] In addition, the logic instructions in the memory 530 described above can be implemented in the form of a software functional unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0191] On the other hand, the present invention further provides a computer program product, comprising a computer program, which may be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the electronic medical record text classification model training method provided by the above methods, which includes:

[0192] In the network text content related to the disease obtained based on the crawler program or big data, text information having a similarity greater than or equal to a similarity threshold with the text content included in the first electronic medical record text dataset is screened to construct a network text dataset;

[0193] generating an extended electronic medical record text dataset based on target key information and a preset electronic medical record template, wherein the target key information is determined based on the first electronic medical record text dataset, wherein the target key information includes a correlation between disease type and patient age, a correlation between disease symptoms and patient environment, a correlation between treatment means and disease type, and a correlation between diagnostic basis and disease type;

[0194] Performing K-Means clustering on the first electronic medical record text dataset and the extended electronic medical record text dataset to obtain electronic medical record text data of multiple clusters;

[0195] When the clustering result of the newly added electronic medical record text dataset does not belong to one of the multiple clusters, performing multiple iterative training on a first classification model based on the electronic medical record text dataset that does not belong to one of the multiple clusters; wherein the first classification model is trained based on the first electronic medical record text dataset;

[0196] When the number of iterative training reaches the iteration number threshold and the model training indicators in each iterative training are normal, obtaining a second classification model obtained by the last iterative training in the multiple iterative trainings;

[0197] When the model performance index of the second classification model is better than that of the second classification model, the first classification model is updated to the second classification model, and the second classification model is used to perform text classification on the electronic medical record text to be classified.

[0198] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the electronic medical record text classification model training method provided by the above methods, the method comprising:

[0199] In the disease-related network text content based on the crawler program or big data acquisition, the text information with a similarity greater than or equal to a similarity threshold value is screened from the text content contained in the first electronic medical record text data set, and a network text data set is constructed;

[0200] Based on the target key information and the preset electronic medical record template, an extended electronic medical record text data set is generated, and the target key information is determined based on the first electronic medical record text data set, wherein the target key information includes an association relationship between a disease type and a patient age, an association relationship between a disease symptom and an environment in which a patient is located, an association relationship between a treatment method and a disease type, and an association relationship between a diagnosis basis and a disease type.

[0201] The first electronic medical record text data set and the extended electronic medical record text data set are subjected to K-Means clustering processing to obtain a plurality of class clusters of electronic medical record text data.

[0202] In a case where the clustering result of the newly added electronic medical record text data set is not one of the plurality of class clusters, a first classification model is subjected to multiple iteration training based on the electronic medical record text data set which is not one of the plurality of class clusters, and the first classification model is trained based on the first electronic medical record text data set.

[0203] In a case where the number of iteration training reaches an iteration number threshold value and the model training index in each iteration training is normal, a second classification model obtained in the last iteration training in the multiple iteration training is acquired.

[0204] In a case where the model performance index of the second classification model is better than that of the second classification model, the first classification model is updated to the second classification model, and the second classification model is used for text classification of a to-be-classified electronic medical record text.

[0205] The device embodiments described above are only schematic, wherein the units shown as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment scheme. Those skilled in the art can understand and implement it without creative labor.

[0206] Those skilled in the art can clearly understand the technical solutions of the various embodiments from the above description of the embodiments, and the various embodiments can be implemented by means of software with the necessary general hardware platforms, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part of the prior art that makes a contribution, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0207] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for training an electronic medical record text classification model, characterized in that: include: In the network text content related to the disease obtained based on the crawler program or big data, text information having a similarity greater than or equal to a similarity threshold with the text content included in the first electronic medical record text dataset is screened to construct a network text dataset; generating an extended electronic medical record text dataset based on target key information and a preset electronic medical record template, wherein the target key information is determined based on the first electronic medical record text dataset, wherein the target key information includes a correlation between disease type and patient age, a correlation between disease symptoms and patient environment, a correlation between treatment means and disease type, and a correlation between diagnostic basis and disease type; Performing K-Means clustering on the first electronic medical record text dataset and the extended electronic medical record text dataset to obtain electronic medical record text data of multiple clusters; When the clustering result of the newly added electronic medical record text dataset does not belong to one of the multiple clusters, performing multiple iterative training on a first classification model based on the electronic medical record text dataset that does not belong to one of the multiple clusters; wherein the first classification model is trained based on the first electronic medical record text dataset; When the number of iterative training reaches the iteration number threshold and the model training indicators in each iterative training are normal, obtaining a second classification model obtained by the last iterative training in the multiple iterative trainings; When the model performance index of the second classification model is better than that of the first classification model, the first classification model is updated to the second classification model, and the second classification model is used to perform text classification on the electronic medical record text to be classified.

2. The electronic medical record text classification model training method according to claim 1, characterized in that: Before performing multiple iterative training on the first classification model based on the electronic medical record text dataset that is not a cluster among the multiple clusters, the method further includes: Preprocessing the initial electronic medical record text data set to obtain a preprocessed initial electronic medical record text data set; Classifying and labeling the preprocessed initial electronic medical record text dataset to obtain the first electronic medical record text dataset; Based on the first electronic medical record text dataset, an initial classification model is trained to obtain the first classification model.

3. The electronic medical record text classification model training method according to claim 1, characterized in that: The performing multiple iterative training on the first classification model based on the electronic medical record text dataset that is not a cluster among the multiple clusters includes: Classify and label the electronic medical record text dataset that does not belong to one of the multiple clusters to obtain a second electronic medical record text dataset; When the data volume of the second electronic medical record text dataset exceeds a quantity threshold, the first classification model is trained for multiple iterations based on the second electronic medical record text dataset and the first electronic medical record text dataset.

4. The electronic medical record text classification model training method according to any one of claims 1 to 3, characterized in that: After updating the first classification model to the second classification model, the method further includes: Inputting the electronic medical record text to be classified into the second classification model for text classification, and obtaining classification result data of the electronic medical record text to be classified output by the second classification model; Based on the classification result data, determining the accuracy of the classification result data; When the accuracy of the classification result data is greater than or equal to the accuracy threshold and the similarity between the classification result data and the training data of the second classification model is less than the similarity threshold, the classification result data is determined as the newly added electronic medical record text data set.

5. The electronic medical record text classification model training method according to any one of claims 1 to 3, characterized in that: After obtaining the second classification model obtained by the last iterative training in the multiple iterative trainings, the method further includes: When the model performance index of the second classification model is not better than that of the second classification model, the electronic medical record text to be classified is classified based on the first classification model.

6. The electronic medical record text classification model training method according to any one of claims 1 to 3, characterized in that: After the first classification model is iteratively trained for multiple times, the method further includes: If the model training indicator of the classification model obtained in any iterative training in the multiple iterative trainings is abnormal, stopping the multiple iterative trainings; When the multiple iterative trainings are stopped, the electronic medical record text to be classified is classified based on the first classification model.

7. An electronic medical record text classification model training device, characterized in that: include: a data acquisition module, configured to screen, from disease-related web text content acquired based on a crawler program or big data, text information having a similarity greater than or equal to a similarity threshold with the text content included in the first electronic medical record text dataset, and construct a web text dataset; a data expansion module, configured to generate an extended electronic medical record text dataset based on target key information and a preset electronic medical record template, wherein the target key information is determined based on the first electronic medical record text dataset, wherein the target key information includes a correlation between disease type and patient age, a correlation between disease symptoms and patient environment, a correlation between treatment means and disease type, and a correlation between diagnostic basis and disease type; a data clustering module, configured to perform K-Means clustering on the first electronic medical record text dataset and the extended electronic medical record text dataset to obtain electronic medical record text data of multiple clusters; a model training module for, when a clustering result of a newly added electronic medical record text dataset does not fall into one of the multiple clusters, iteratively training a first classification model based on the electronic medical record text dataset that does not fall into one of the multiple clusters; wherein the first classification model is trained based on the first electronic medical record text dataset; A model acquisition module is used to obtain a second classification model obtained by the last iterative training in the multiple iterative trainings when the number of iterative trainings reaches an iteration number threshold and the model training indicators in each iterative training are normal; The model updating module is used to update the first classification model to the second classification model when the model performance index of the second classification model is better than that of the first classification model. The second classification model is used to perform text classification on the electronic medical record text to be classified.

8. The electronic medical record text classification model training device according to claim 7, characterized in that: The device further comprises: a preprocessing module, configured to preprocess the initial electronic medical record text dataset before performing multiple iterative training on the first classification model based on the newly added electronic medical record text dataset and the first electronic medical record text dataset, to obtain a preprocessed initial electronic medical record text dataset; Classifying and labeling the preprocessed initial electronic medical record text dataset to obtain the first electronic medical record text dataset; Based on the first electronic medical record text dataset, an initial classification model is trained to obtain the first classification model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the electronic medical record text classification model training method as described in any one of claims 1 to 6 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the electronic medical record text classification model training method according to any one of claims 1 to 6 is implemented.

Images