Rheumatoid arthritis auxiliary classification model training method and classification method

By performing feature-level fusion of structured and unstructured data on rheumatoid arthritis data, combining BERT model and support vector machine, the data is heterogeneity problem is solved, the diagnostic accuracy is improved, and resources are optimized when data is updated Utilization, efficient diagnostic support is achieved.

CN118940131APending Publication Date: 2024-11-12SHAANXI NORMAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410984283.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Data on rheumatoid arthritis disease usually contain data from different sources and structures, which leads to complex data analysis and model construction, and it is difficult for the existing technology to effectively integrate structured and unstructured data, resulting in low diagnostic accuracy.

Method used

A training method for rheumatoid arthritis assisted classification model based on multi-source heterogeneous data fusion is proposed. By conducting feature engineering and naming entity recognition and word segmentation on structured data, generating a domain vocabulary list, combining with BERT model Classify with the support vector machine, realize feature-level fusion of data, and introduce incremental learning to optimize the model.

Benefits of technology

It effectively solves the problem of data heterogeneity, improves the diagnostic accuracy, avoids feature fuzzy information and uncertainty, and only selective training of new data when data is updated, improving resource utilization efficiency and avoiding repeated calculations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940131A_ABST
    Figure CN118940131A_ABST
Patent Text Reader

Abstract

The invention discloses a rheumatoid arthritis auxiliary classification model training method and a classification method, and the method comprises the steps: carrying out the feature engineering of structural data in the data of a rheumatoid arthritis patient, and obtaining a feature matrix; using a BiLSTM + CRF method to complete named entity recognition on unstructured data, performing word segmentation according to a result to generate a rheumatoid arthritis field word segmentation word list, performing word segmentation and encoding, and classifying obtained feature codes by using a BERT + label embedding method; fusing the feature matrix of the structured data and the confidence-based classification result of the unstructured data to obtain a fused feature matrix; classifying the fusion feature matrix by using a support vector machine to obtain a classification result; and during each time of data updating, performing named entity recognition on unstructured data, updating a word segmentation word list, calculating marginal loss based on the fused feature samples, including the samples corresponding to the marginal loss greater than a threshold value into a support vector set, updating the support vector set and corresponding labels, and training and updating a support vector machine.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention integrates knowledge from the fields of medicine, information science and computer application, and invents a rheumatoid arthritis auxiliary classification model training method and a classification method based on multi-source heterogeneous data fusion. Background Art

[0002] Rheumatoid arthritis (RA) is a systemic immune rheumatic disease characterized by polyarticular, symmetrical, erosive joint inflammation of the small joints of the hands and feet. The incidence of RA is as high as 0.3% to 1%. It is a difficult-to-treat disease with a recurrent course, lingering and high disability rate. It can affect multiple systems of the body and seriously endanger human health. With the advancement of information technology, the demand for using data science to strengthen medical care and disease management is rapidly increasing. In the diagnosis of rheumatic and immune diseases, computer-aided diagnosis also faces many problems, such as: rheumatoid arthritis disease data usually contains data from different sources and structures, which makes data analysis and model construction complicated. Summary of the invention

[0003] The present invention proposes a rheumatoid arthritis auxiliary classification model training method and a classification method, which divides rheumatoid arthritis data into structured data and unstructured data, processes the two parts of data separately, and performs feature-level fusion, and performs classification and diagnosis based on the fused features to assist doctors in identification and improve accuracy.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is: a rheumatoid arthritis auxiliary classification model training method based on multi-source heterogeneous data fusion, comprising the following steps:

[0005] For the feature engineering of structured data in multi-source heterogeneous data of rheumatoid arthritis, data cleaning and feature encoding are performed on the structured data, and the data categories and feature attributes of the structured data are extracted to obtain the feature matrix of the structured data;

[0006] Extract unstructured data from multi-source heterogeneous data of rheumatoid arthritis, annotate named entities, use BiLSTM+CRF for named entity recognition, and concatenate named entity phrases matching the symptoms of rheumatoid arthritis;

[0007] Based on the obtained named entity phrases, the text data is segmented to generate a domain vocabulary of rheumatoid arthritis; the original text data is segmented and encoded using the domain vocabulary of rheumatoid arthritis to construct an encoded representation of the text data;

[0008] The encoded representation of text data is used as input, and the BERT model + label embedding method is used for encoding and classification to obtain the classification results and confidence levels of the unstructured data part.

[0009] The classification results of unstructured data are combined with the feature matrix of structured data to obtain a new fused feature matrix, and the new fused feature matrix is ​​used as input for final classification using a support vector machine;

[0010] Perform named entity recognition on the unstructured data classification part of the newly added data, and update the rheumatoid arthritis field vocabulary; calculate the marginal loss of the data samples after feature-level fusion of multi-source heterogeneous data. When the loss is greater than the given threshold τ, the data samples are included in the support vector set, and the support vector set and the corresponding labels are updated; train and update the support vector machine.

[0011] Furthermore, the multi-source heterogeneous data of rheumatoid arthritis disease include electronic medical records, laboratory values, and medical test results.

[0012] Furthermore, unstructured data from multi-source heterogeneous data of rheumatoid arthritis disease is extracted, named entities are annotated, BiLSTM+CRF is used for named entity recognition, and named entity phrases matching the symptoms of rheumatoid arthritis are concatenated, including:

[0013] Label each entity in the extracted rheumatoid arthritis symptom data, and use the BIO labeling method to label the entity's starting position, internal position, and non-entity position;

[0014] Take the text data and the corresponding annotated dataset as input and use the BiLSTM+CRF model to perform named entity recognition tasks;

[0015] Re-join the identified named entities to obtain entity phrases.

[0016] Furthermore, based on the obtained named entity phrases, the text data is segmented to generate a domain vocabulary of rheumatoid arthritis; the original text data is segmented and encoded using the domain vocabulary of rheumatoid arthritis, and the encoding representation of the text data is constructed, including:

[0017] Generate a vocabulary list in the field of rheumatoid arthritis according to the entity phrases after the named entities are concatenated, and each vocabulary phrase has a unique position in the vocabulary list in the field of rheumatoid arthritis;

[0018] Perform named entity recognition on the unstructured text raw data, and reassemble the entity content obtained by named entity recognition into entity phrases;

[0019] The obtained entity phrases are taken as input and encoded using the rheumatoid arthritis domain vocabulary. Each word is mapped to the corresponding vector representation to construct the encoded representation of the text data.

[0020] Furthermore, named entity recognition is performed on the original data of the unstructured text. When the entity content obtained by the named entity recognition is reassembled into entity phrases, the splicing is performed according to the set rule Entity = (B + I) or N, N = {1, 2, 3, 4, 5}. Specifically, corresponding to the rule when generating the vocabulary, the digital position information is processed separately according to the rule when splicing; the rule when generating the vocabulary is:

[0021] Entity=B+I

[0022] Where B and I represent the parts marked as B and I in the entity annotation respectively.

[0023] Furthermore, the encoded representation of the text data is used as input, and the BERT model + label embedding method is used for encoding and classification. The classification results and confidence levels of the unstructured data are obtained, including:

[0024] The encoded representation of the text data is used as input, and the input is encoded using the pre-trained BERT model. After being encoded by the multi-layer Transformer of the BERT model, the semantic representation of the text is obtained. The BERT model is used to capture rich information in the text, such as contextual information, and convert the text into a high-dimensional vector representation;

[0025] Embed the label of each category to obtain the embedding vector of the label;

[0026] Combine the high-dimensional vector representation of the text extracted by the BERT model with the embedding vector of the label. By performing an element-by-element product of the two vectors, the combined vector will contain both text features and label information.

[0027] The combined vector is input into the fully connected layer for confidence-based classification of unstructured data to obtain the classification result and its confidence.

[0028] Furthermore, incremental learning is introduced into multi-source heterogeneous data fusion and classification, and incremental support vector machine classification is adopted. The specific steps of incremental support vector machine are as follows:

[0029] Get new data and divide it into structured data and unstructured data according to the data structure;

[0030] Perform named entity recognition on unstructured data, and add newly appeared entity phrases to the domain vocabulary of rheumatoid arthritis based on the recognition results to update the domain vocabulary of rheumatoid arthritis;

[0031] The confidence-based classification result of the unstructured data is combined with the feature matrix of the structured data to obtain a new feature matrix, and the marginal loss of the feature matrix is ​​calculated. If the marginal loss is greater than the set threshold τ, the data sample is included in the support vector set, and the support vector set and the corresponding label are updated;

[0032] According to the new set of support vectors and corresponding labels, the support vector machine model and its parameters are updated using the stochastic gradient descent method.

[0033] The present invention also provides a computer device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, the rheumatoid arthritis auxiliary classification model training method described in the present invention can be implemented.

[0034] At the same time, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the rheumatoid arthritis auxiliary classification model training method of the present invention can be implemented.

[0035] The present invention also provides a rheumatoid arthritis auxiliary classification method, which obtains multi-source heterogeneous data of rheumatoid arthritis diseases to be classified, divides the multi-source heterogeneous data of rheumatoid arthritis diseases to be classified into unstructured data and structured data, and uses the unstructured data and structured data as inputs of the rheumatoid arthritis auxiliary classification model of the present invention to obtain classification results and their confidence levels.

[0036] Compared with the prior art, the present invention has at least the following beneficial effects:

[0037] The present invention performs feature-level fusion of multi-source heterogeneous data on rheumatoid arthritis data to solve the heterogeneity problem between data and achieve better integration effect; unstructured text data completes named entity recognition through BiLSTM+CRF, and generates a specific vocabulary for the field of rheumatoid arthritis based on named entity segmentation; BERT+label embedding is used to classify unstructured text data, and the relationship between features is represented by the distance information of vectors in space, thereby avoiding fuzzy information and uncertainty relationship between features. Incremental learning is introduced into multi-source heterogeneous data fusion and classification. Whenever there is new data, only new data needs to be selectively trained, and there is no need to retain a large amount of training data. While using resources more efficiently, repeated calculations are avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of an auxiliary classification method for rheumatoid arthritis based on incremental multi-source heterogeneous data fusion.

[0039] Figure 2 This is a flow chart of an auxiliary classification method for rheumatoid arthritis based on multi-source heterogeneous data fusion. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Figure 2 The flowchart of a rheumatoid arthritis auxiliary classification method based on multi-source heterogeneous data fusion is shown. In order to deeply integrate structured data and unstructured data information and provide a more comprehensive feature representation, feature-level fusion is performed on multi-source heterogeneous data of rheumatoid arthritis. Feature-level fusion is an intermediate-level fusion. First, feature extraction is performed on the original data, and then the feature data is comprehensively analyzed and processed; considerable information compression is achieved and real-time processing is facilitated. On the basis of fused features, auxiliary classification of rheumatoid arthritis is further realized. Specifically, the following steps are included:

[0042] Step 1: Collect data and determine the multi-source heterogeneous relationships of data.

[0043] Collect multi-source heterogeneous data on the disease by consulting medical records, reading professional literature, and consulting doctors and experts. The collected disease-related information includes the physical symptoms, medical test results, and diagnosis results of patients with rheumatoid arthritis; consulting medical records can supplement the knowledge obtained from other channels from the perspective of actual cases, making the collected information more complete.

[0044] The structured data includes the patient’s personal information (gender, age) and medical test results or laboratory values, including rheumatoid factor, erythrocyte sedimentation rate, etc. The unstructured data includes some text descriptions of the patient’s condition, including the patient’s current medical history, specialist conditions, etc.

[0045] Step 2: Structured data feature engineering. Perform data cleaning and feature encoding processing operations on the structured data, extract the data categories and feature attributes of the structured data, and obtain the feature matrix of the structured data;

[0046] Data cleaning (processing missing values): fill in missing values ​​or delete samples / features with more missing values;

[0047] Feature encoding: label encoding of feature attributes and category labels;

[0048] Feature standardization: Numerical features (such as rheumatoid factor) are standardized so that their mean is 0 and their variance is 1.

[0049] Step 3: Named Entity Recognition (NER). Unstructured data is extracted from the electronic medical record. First, the named entities are labeled. Then, BiLSTM+CRF is used for named entity recognition. The named entity phrases matching the symptoms of rheumatoid arthritis are concatenated. The steps are as follows:

[0050] Step 3.1, label each entity in the extracted rheumatoid arthritis symptom data. Use the BIO (Begin, Inside, Outside) labeling method to label the entity's starting position, internal position, and non-entity position;

[0051] The medical entity types of rheumatoid arthritis are subdivided according to the needs of annotation, including:

[0052] 1) Position: describes the approximate position of a small part of the body, such as "left", "left hand", etc.

[0053] 2) Body (body parts): such as "shoulder joint", "metacarpophalangeal joint", etc.;

[0054] 3) Number (specific numerical position): mainly describes the specific position of the joints of the hands and feet, such as "each" in each metacarpophalangeal joint and "the 2nd to 4th" in the 2nd to 4th proximal interphalangeal joints;

[0055] 4) Function: describes some functions or activities that the body can normally complete, such as "dorsiflexion and flexion function" of the left wrist, "lifting activity" of the right shoulder, etc.

[0056] 5) Degree: describes the degree to which the patient is affected, such as "mild" tenderness, "slight" swelling, "significant" limitation of movement, etc.

[0057] 6) Signs: Describe the patient's symptoms, such as "tenderness", "swelling", "restriction", etc.;

[0058] 7) Check (check items): such as "bone friction sound", "floating patella test", etc.;

[0059] 8) Result: Describes the result of the test item, such as "positive", "-", etc.

[0060] Step 3.2, select a model suitable for the named entity recognition task, take the text data and the corresponding annotation dataset as input, and use the BiLSTM+CRF model to perform the named entity recognition task. Taking "The wrist joint is slightly swollen and tender." as an example, the result of entity recognition based on the annotation should be {"wrist":"B-Body","joint":"I-Body","slightly":"B-Degree","swollen":"B-Signs","swelling":"I-Signs","、":"O","pressure":"B-Signs","pain":"I-Signs",".":"O"};

[0061] Step 3.3, re-join the entities identified in step 3.2 to obtain entity phrase sequences, such as ["wrist joint", "slightly", "swelling", "tenderness"] after joining in the above example;

[0062] Step 4: Word segmentation and encoding. Based on the named entity phrases obtained in step 3, the text data is segmented to generate a vocabulary in the field of rheumatoid arthritis; the rheumatoid arthritis vocabulary is used to segment and encode the original text data;

[0063] Step 4.1, generating a rheumatoid arthritis domain vocabulary according to the concatenated entity phrases, wherein each vocabulary phrase has a unique position in the rheumatoid arthritis domain vocabulary;

[0064] In addition to the digital location information, each entity is processed and reassembled according to the subdivision rules during annotation. The subdivision rules are described as follows:

[0065] Entity=B+I

[0066] Where B and I represent the parts marked as B and I in the entity annotation respectively;

[0067] For example, "wrist joint" is spliced ​​into an entity phrase, which occupies a single row in the vocabulary and is encoded as one word. The types of digital position information are complex, such as "2-4", "first", "second, third, fifth", "each", etc. In order to avoid redundant information caused by segmenting only by named entities, these numbers are encoded separately, that is, numbers 1 to 5 are located in five different rows in the vocabulary and are represented by different codes.

[0068] Step 4.2, perform named entity recognition on the unstructured text raw data according to steps 3.1-3.2, and reassemble the meaningful content obtained by named entity recognition (i.e., the content marked with B and I in the BIO annotation method) into entity phrase data according to the set rules. The set rules are described as follows:

[0069] Entity=(B+I)or N,N={1,2,3,4,5};

[0070] Specifically, in order to correspond to the rules when generating the vocabulary in 4.1, the digital position information is processed separately according to the rules during splicing, that is, the digital position information is refined into specific and detailed digital information, such as "2-4" corresponds to "2", "3", "4", "the 2nd" corresponds to "2", "the 2nd, 3rd, 5th" corresponds to "2", "3", "5", and "each" corresponds to "1", "2", "3", "4", "5", etc.

[0071] Step 4.3: Take the entity data obtained in step 4.2 as input, encode it using the vocabulary in step 4.1, map each word to a corresponding vector representation (a word corresponds to its position in the vocabulary one by one), and construct an encoded representation of the text data;

[0072] Step 5: Classify unstructured data. The code obtained from the word segmentation encoding in step 4 is used as input, and the Bert model + label embedding method is used for encoding and classification.

[0073] Step 5.1, the encoded representation of the generated text data is taken as input;

[0074] Step 5.2: Use the pre-trained BERT model to encode the input, and then pass it through the model's multi-layer Transformer encoding to obtain the semantic representation of the text. The BERT model can capture rich information in the text, such as contextual information, and convert the text into a high-dimensional vector representation;

[0075] Step 5.3, embed the label of each category to obtain the label embedding vector;

[0076] Step 5.4: Based on the semantic representation of the text output by the BERT model, an additional label embedding vector is added to combine the text features extracted by the BERT model with the label embedding vector. The combined vector can contain both text features and label information by performing an element-by-element product of the two vectors.

[0077] In step 5.5, the combined vector containing text features and label information is input into the fully connected layer for classification to obtain the confidence-based classification result of the unstructured data.

[0078] Step 6: Combine the confidence-based classification results of the unstructured data with the feature matrix of the structured data to obtain a new feature matrix, and use the new feature matrix as input for final classification using a support vector machine. The kernel function of the support vector machine is a Gaussian kernel function (RBF, Radial Basis Function Kernel), and the formula is as follows:

[0079]

[0080] Among them, x and y are the feature vectors of two sample points, and γ is the hyperparameter of the Gaussian kernel function.

[0081] Figure 1 The method shown in the figure is an auxiliary classification method for rheumatoid arthritis based on incremental multi-source heterogeneous data fusion. Figure 2 Incremental learning is introduced on the basis of . When there is new data, only the new data needs to be selectively trained, and there is no need to retain a large amount of training data. This makes more efficient use of resources and avoids repeated calculations. The specific contents of the newly added steps are as follows:

[0082] Step 7: Whenever the data is updated, named entity recognition is performed on the classified part of the unstructured data in the new data, and the rheumatoid arthritis domain vocabulary is updated; in addition, the marginal loss is calculated for the data after feature-level fusion of multi-source heterogeneous data in the sample. When the loss is greater than the given threshold τ, the fused data is included in the support vector set, and the support vector set and the corresponding labels are updated; finally, the support vector machine is trained and updated.

[0083] Step 7.1, collect new data and divide it into structured data and unstructured data according to the data structure;

[0084] Step 7.2, perform named entity recognition on the unstructured data, and add the newly appeared entity phrases to the word segmentation vocabulary according to the recognition results, and update the domain word segmentation vocabulary;

[0085] Step 7.3, merge the confidence-based classification results of unstructured data with the feature matrix of structured data to obtain a new feature matrix, and calculate its marginal loss. For the new data sample (x, y), the calculation formula of the Squared Hinge Loss function is:

[0086] L(x,f(x))=max(0,1-yf(x)) 2

[0087] Where f(x) represents the predicted output and y represents the correct category;

[0088] If the marginal loss is greater than the set threshold τ, the sample is included in the support vector set and the label corresponding to the support vector set kernel is updated;

[0089] Step 7.4, based on the new set of support vectors and corresponding labels, use the incremental optimization algorithm - stochastic gradient descent to update the support vector machine model parameters γ with Gaussian kernel.

[0090] In summary, the present invention proposes an auxiliary classification method for rheumatoid arthritis based on multi-source heterogeneous data fusion. First, the collected rheumatoid arthritis patient data is divided into structured data and unstructured data according to the structure; wherein, feature engineering is performed on the structured data, including data cleaning, feature encoding, etc., to obtain the feature matrix of the structured data; the unstructured data uses the BiLSTM+CRF method to complete named entity recognition, and the word segmentation is performed according to the NER result to generate a word segmentation vocabulary in the field of rheumatoid arthritis, and the word segmentation and encoding are performed based on this vocabulary. The feature encoding obtained is used as input using BERT+label The embedding method is used for classification; secondly, the feature matrix of structured data is merged with the confidence-based classification result of unstructured data to obtain a new fused feature matrix, completing the feature-level fusion of multi-source heterogeneous data; finally, the feature matrix is ​​used as input for classification using a support vector machine to obtain the final classification result; in addition, each time the data is updated, named entity recognition is performed on the unstructured data and the word segmentation vocabulary is updated, and then the marginal loss is calculated based on the fused feature sample. If it is greater than a given threshold, the sample is included in the support vector set, and the support vector set and the corresponding label are updated; finally, the model is trained and updated.

[0091] The present invention can also provide a computer device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, the rheumatoid arthritis auxiliary classification model training method described in the present invention can be implemented.

[0092] At the same time, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the rheumatoid arthritis auxiliary classification model training method of the present invention can be implemented.

[0093] The computer device may be a laptop computer, a desktop computer or a workstation.

[0094] The processor of the present invention may be a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC) or a readily available field programmable gate array (FPGA).

[0095] The memory of the present invention may be an internal storage unit of a laptop computer, a desktop computer or a workstation, such as a memory or a hard disk; or an external storage unit, such as a mobile hard disk or a flash memory card.

[0096] Computer-readable storage media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer-readable storage media may include: read-only memory (ROM), random access memory (RAM), solid-state drive (SSD) or optical disk, etc. Among them, random access memory may include resistance random access memory (ReRAM) and dynamic random access memory (DRAM).

[0097] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A rheumatoid arthritis auxiliary classification model training method, characterized in that: The following steps are involved: For the feature engineering of structured data in multi-source heterogeneous data of rheumatoid arthritis, data cleaning and feature encoding are performed on the structured data, and the data categories and feature attributes of the structured data are extracted to obtain the feature matrix of the structured data; Extract unstructured data from multi-source heterogeneous data of rheumatoid arthritis, annotate named entities, use BiLSTM+CRF for named entity recognition, and concatenate named entity phrases matching the symptoms of rheumatoid arthritis; Based on the obtained named entity phrases, the text data is segmented to generate a domain vocabulary of rheumatoid arthritis; the original text data is segmented and encoded using the domain vocabulary of rheumatoid arthritis to construct an encoded representation of the text data; The encoded representation of text data is used as input, and the BERT model + label embedding method is used for encoding and classification to obtain the classification results and confidence levels of the unstructured data part. The classification results of unstructured data are combined with the feature matrix of structured data to obtain a new fused feature matrix, and the new fused feature matrix is ​​used as input for final classification using a support vector machine; Perform named entity recognition on the unstructured data classification part of the newly added data, and update the rheumatoid arthritis field vocabulary; calculate the marginal loss of the data sample after the feature-level fusion of multi-source heterogeneous data. When the loss is greater than the given threshold τ, the data sample is included in the support vector set, and the support vector set and the corresponding label are updated; Train and update a support vector machine.

2. The rheumatoid arthritis auxiliary classification model training method according to claim 1, characterized in that: The multi-source heterogeneous data of rheumatoid arthritis disease include electronic medical records, laboratory values, and medical test results.

3. The rheumatoid arthritis auxiliary classification model training method according to claim 1, characterized in that: Extract unstructured data from multi-source heterogeneous data of rheumatoid arthritis, annotate named entities, use BiLSTM+CRF for named entity recognition, and concatenate named entity phrases matching rheumatoid arthritis symptoms, including: Label each entity in the extracted rheumatoid arthritis symptom data, and use the BIO labeling method to label the entity's starting position, internal position, and non-entity position; Take the text data and the corresponding annotated dataset as input and use the BiLSTM+CRF model to perform named entity recognition tasks; Re-join the identified named entities to get entity phrases.

4. The rheumatoid arthritis auxiliary classification model training method according to claim 1, characterized in that: Based on the obtained named entity phrases, the text data is segmented to generate a domain vocabulary of rheumatoid arthritis; The original text data is segmented and encoded using the rheumatoid arthritis domain vocabulary, and the encoding representation of the text data is constructed including: Generate a vocabulary list in the field of rheumatoid arthritis according to the entity phrases after the named entities are concatenated, and each vocabulary phrase has a unique position in the vocabulary list in the field of rheumatoid arthritis; Perform named entity recognition on the unstructured text raw data, and reassemble the entity content obtained by named entity recognition into entity phrases; The obtained entity phrases are taken as input and encoded using the rheumatoid arthritis domain vocabulary. Each word is mapped to the corresponding vector representation to construct the encoded representation of the text data.

5. The rheumatoid arthritis auxiliary classification model training method according to claim 4, characterized in that: Perform named entity recognition on the original data of unstructured text. When the entity content obtained by named entity recognition is reassembled into entity phrases, it is spliced ​​according to the set rule Entity = (B + I) or N, N = {1, 2, 3, 4, 5}. Specifically, corresponding to the rules when generating the vocabulary, the digital position information is processed separately according to the rules when splicing; the rules when generating the vocabulary are: Entity=B+I Where B and I represent the parts marked as B and I in the entity annotation respectively.

6. The rheumatoid arthritis auxiliary classification model training method according to claim 1, characterized in that: The encoded representation of text data is used as input, and the BERT model + label embedding method is used for encoding and classification. The classification results and confidence levels of the unstructured data are as follows: The encoded representation of the text data is used as input, and the input is encoded using the pre-trained BERT model. After being encoded by the multi-layer Transformer of the BERT model, the semantic representation of the text is obtained. The BERT model is used to capture rich information in the text, such as contextual information, and convert the text into a high-dimensional vector representation; Embed the label of each category to obtain the embedding vector of the label; Combine the high-dimensional vector representation of the text extracted by the BERT model with the embedding vector of the label. By performing an element-by-element product of the two vectors, the combined vector will contain both text features and label information. The combined vector is input into the fully connected layer for confidence-based classification of unstructured data to obtain the classification result and its confidence.

7. The rheumatoid arthritis auxiliary classification model training method according to claim 1, characterized in that: Incremental learning is introduced into multi-source heterogeneous data fusion and classification, and incremental support vector machine classification is adopted. The specific steps of incremental support vector machine are as follows: Get new data and divide it into structured data and unstructured data according to the data structure; Perform named entity recognition on unstructured data, and add newly appeared entity phrases to the domain vocabulary of rheumatoid arthritis based on the recognition results to update the domain vocabulary of rheumatoid arthritis; The confidence-based classification result of the unstructured data is combined with the feature matrix of the structured data to obtain a new feature matrix, and the marginal loss of the feature matrix is ​​calculated. If the marginal loss is greater than the set threshold τ, the data sample is included in the support vector set, and the support vector set and the corresponding label are updated; According to the new set of support vectors and corresponding labels, the support vector machine model and its parameters are updated using the stochastic gradient descent method.

8. A computer device, characterized in that: It includes a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, it can implement the rheumatoid arthritis auxiliary classification model training method described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that: A computer program is stored in a computer-readable storage medium, and when the computer program is executed by a processor, it can implement the rheumatoid arthritis auxiliary classification model training method as described in any one of claims 1-7.

10. A method for auxiliary classification of rheumatoid arthritis, characterized in that: Obtain multi-source heterogeneous data of rheumatoid arthritis diseases to be classified, divide the multi-source heterogeneous data of rheumatoid arthritis diseases to be classified into unstructured data and structured data, use the unstructured data and structured data as input of the rheumatoid arthritis auxiliary classification model obtained as claimed in any one of claims 1 to 7, and obtain a classification result and its confidence.