Syndrome type classification method and device based on syndrome element decomposition, equipment and medium

Abstract

The application provides a syndrome type classification method and device based on syndrome element decomposition, equipment and medium. The method analyzes the syndrome type name and syndrome type description of the to-be-classified syndrome type through a syndrome element prediction model, splits and predicts the main syndrome element and the secondary syndrome element of the to-be-classified syndrome type as the symptom characteristics of the to-be-classified syndrome type. The target feature vector of the to-be-classified syndrome type is subjected to similarity matching through the retrieval of the known feature vector of the known syndrome type, so that the known syndrome type similar to the feature of the to-be-classified syndrome type is searched in the known syndrome type library to complete the classification of the to-be-classified syndrome type. The application relates to the technical field of model prediction. The feature vector of the to-be-classified syndrome type is obtained through the syndrome element decomposition of the syndrome type, and the vector matching is performed with the known syndrome type, so that the matching between the syndrome types with different standard names and the same symptoms can be realized, the accuracy of the syndrome type matching is improved, and the classification accuracy between the syndrome types with different standards is improved.

Description

Technical Field

[0001] This application relates to the field of model prediction technology, and in particular to a method, apparatus, equipment and medium for syndrome type classification based on syndrome element decomposition. Background Technology

[0002] In the process of TCM diagnosis, there is no standardized symptom description for the diagnosis result. It is only necessary to conform to the patient's objective condition. Therefore, the granularity of TCM diagnosis is particularly precise and detailed. There are thousands of diagnostic results (i.e., syndrome types). The same disease can contain multiple subdivided syndrome types, and there is a dynamic transformation relationship between syndrome types.

[0003] Currently, there are inconsistencies in the standards for defining syndrome types in TCM books, textbooks, and actual clinical practice. When defining syndrome types, the conversion between different standards often results in a loss of accuracy. Furthermore, there are instances where different syndrome types have overlapping symptom meanings, meaning that TCM syndrome types have different diagnostic criteria. Requiring all TCM syndrome types to be converted and aligned using a single standard easily leads to insufficient accuracy in syndrome type conversion. Therefore, addressing the low accuracy of current syndrome type classifications based on different standards has become an urgent technical problem to be solved. Summary of the Invention

[0004] This application provides a method, apparatus, computer equipment, and storage medium for syndrome classification based on syndrome element decomposition, aiming to improve the accuracy of TCM syndrome classification according to different standards.

[0005] Firstly, this application provides a method for classifying syndrome types based on syndrome element decomposition, the method comprising the following steps:

[0006] Obtain the certificate type name and certificate type description of the certificate type to be classified. Based on the certificate element prediction model, the certificate type name and the certificate type description, perform certificate element prediction on the certificate type to be classified to obtain the main certificate element and secondary certificate element of the certificate type to be classified.

[0007] Based on the certificate type name, the certificate type description, the main certificate elements, and the secondary certificate elements, the target feature vector of the certificate type to be classified is obtained.

[0008] Based on the known feature vectors corresponding to known certificate types in the preset certificate type library, similarity matching is performed on the target feature vectors to determine the certificate type category of the certificate type to be classified.

[0009] Secondly, this application also provides a pattern classification device based on pattern element decomposition, the pattern classification device based on pattern element decomposition comprising:

[0010] The evidence element prediction module is used to obtain the evidence type name and evidence type description of the evidence type to be classified, and to perform evidence element prediction on the evidence type to be classified based on the evidence element prediction model, the evidence type name and the evidence type description, so as to obtain the main evidence element and secondary evidence element of the evidence type to be classified.

[0011] The feature vector acquisition module is used to obtain the target feature vector of the certificate type to be classified based on the certificate type name, the certificate type description, the main certificate element and the secondary certificate element;

[0012] The category prediction module is used to perform similarity matching on the target feature vector based on the known feature vectors corresponding to known certificate types in the preset certificate type library, and to determine the certificate type category of the certificate type to be classified.

[0013] Thirdly, this application also provides a computer device, the computer device including a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, it implements the steps of the above-described method for classifying syndrome types based on syndrome element decomposition.

[0014] Fourthly, this application also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of the above-described method for classifying syndrome types based on syndrome element decomposition.

[0015] This application provides a method, apparatus, device, and storage medium for syndrome classification based on syndrome element decomposition. The application includes obtaining the syndrome name and description of a syndrome to be classified; predicting the syndrome elements of the syndrome to be classified based on a syndrome element prediction model, the syndrome name, and the syndrome description to obtain the primary and secondary syndrome elements of the syndrome to be classified; obtaining the target feature vector of the syndrome to be classified based on the syndrome name, the syndrome description, the primary syndrome elements, and the secondary syndrome elements; and performing similarity matching on the target feature vector based on the known feature vectors corresponding to known syndromes in a preset syndrome type library to determine the syndrome type category. The method involves analyzing the syndrome name and description of the syndrome to be classified using a syndrome element prediction model to decompose and predict the primary and secondary syndrome elements of the syndrome to be classified, which serve as symptom features of the syndrome to be classified. By retrieving the known feature vectors of known syndromes, the method performs similarity matching on the target feature vector of the syndrome to be classified, thereby finding known syndromes in the known syndrome type library that are similar to the features of the syndrome to be classified, thus completing the classification of the syndrome to be classified. By using the above method, the feature vector of the syndrome to be classified is obtained by decomposing the syndrome elements of the syndrome. Syndrome matching is then performed using the syndrome elements as the essence of the syndrome. This method can achieve matching between syndromes with the same symptoms but different naming standards, which improves the accuracy of syndrome matching and solves the technical problem of low accuracy in syndrome classification under different standards. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 An embodiment of this application provides a certificate type classification system based on certificate element decomposition;

[0018] Figure 2 A flowchart illustrating a first embodiment of a pattern classification method based on pattern element decomposition provided in this application;

[0019] Figure 3 A flowchart illustrating a second embodiment of a pattern classification method based on pattern element decomposition provided in this application;

[0020] Figure 4 A flowchart illustrating a third embodiment of a syndrome type classification method based on syndrome element decomposition provided in this application;

[0021] Figure 5 A flowchart illustrating the fourth embodiment of a method for classifying syndrome types based on syndrome element decomposition provided in this application;

[0022] Figure 6 This is a schematic block diagram of a pattern classification device based on pattern element decomposition provided in an embodiment of this application;

[0023] Figure 7 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application.

[0024] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0027] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0028] The embodiments of this application provide a syndrome classification method based on syndrome element decomposition, a syndrome classification device based on syndrome element decomposition, a computer device, and a storage medium, which are used to achieve matching between syndromes with the same symptoms but different naming standards, thereby improving the classification accuracy of TCM syndromes based on different standards.

[0029] like Figure 1 As shown, Figure 1 An embodiment of this application provides a syndrome type classification system based on syndrome element decomposition. The system includes a terminal and a server, which are communicatively connected. The server is communicatively connected to a database, which includes a syndrome type database and a symptom corpus database.

[0030] The terminals include electronic devices such as mobile phones, tablets, laptops, desktop computers, personal digital assistants, and wearable devices.

[0031] The server may be a single independent server or a server cluster.

[0032] The following will provide a detailed description of the syndrome classification method based on syndrome element decomposition provided in the embodiments of this application, based on the syndrome type classification system based on syndrome element decomposition.

[0033] Please refer to Figure 2 , Figure 2 This is a flowchart illustrating a first embodiment of a syndrome type classification method based on syndrome element decomposition provided in this application. This syndrome type classification method based on syndrome element decomposition can be used in the terminal of a syndrome type classification system based on syndrome element decomposition.

[0034] like Figure 2 As shown, the syndrome type classification method based on syndrome element decomposition includes steps S101 to S103.

[0035] Step S101: Obtain the certificate type name and certificate type description of the certificate type to be classified. Based on the certificate element prediction model, the certificate type name and the certificate type description, perform certificate element prediction on the certificate type to be classified to obtain the main certificate element and secondary certificate element of the certificate type to be classified.

[0036] In this embodiment, the syndrome type name and syndrome type description of a to-be-classified syndrome type with unknown classification are input into the syndrome element prediction model. The syndrome element prediction model performs syndrome element decomposition on the syndrome type name and syndrome type description of the to-be-classified syndrome type, and sequentially extracts the main syndrome elements and secondary syndrome elements therein as the symptom features of the to-be-classified syndrome type.

[0037] In an exemplary embodiment, the syndrome type name and syndrome type description of the to-be-classified syndrome type can be repeatedly input into the syndrome element prediction model. First, the main syndrome elements are extracted, and then the secondary syndrome elements are extracted. For example, the syndrome type name and syndrome type description are input into the syndrome element prediction model for the first decomposition prediction, and the result of this prediction is used as the main syndrome element; then the main syndrome element, syndrome type name and syndrome type description are input into the syndrome element prediction model again. Based on the main syndrome element as the decomposition basis, the syndrome type name and syndrome type description are split, and the elements having an associated relationship with the main syndrome element are extracted as the secondary syndrome elements; then this can be used as the decomposition basis again and input into the syndrome element prediction model for the next syndrome element prediction until no result is output or the output result is invalid.

[0038] Among them, syndrome elements are the disease locations and disease natures determined through the identification of symptoms (such as pathological information of symptoms and signs), and are the basic elements constituting the syndrome name. The "syndrome elements" referred to in the traditional Chinese medicine syndrome differentiation system have the following basic characteristics: Syndrome elements are the essential natures of diseases identified according to syndromes; Syndrome elements mainly refer to the disease locations and disease natures determined by syndrome differentiation; The content of syndrome elements is determined according to traditional Chinese medicine theory; Syndrome elements are the elements constituting the syndrome name; Disease nature syndrome elements are the essential generalizations of the struggle between pathogenic factors and healthy qi; Syndrome elements are specific diagnostic units rather than classification outlines; Syndrome elements have certain combination rules; There may be an overlapping covering relationship between some syndrome elements. Syndrome elements can include disease location syndrome elements and disease nature syndrome elements, etc. For example, disease location syndrome elements can include heart, spirit (brain), lung, spleen, liver, meridians, skin, muscles and bones, etc., and disease nature syndrome elements can include wind, cold, summer heat, heat, dampness, qi stagnation, blood stasis, yin deficiency, yang hyperactivity, etc.

[0039] Step S102: Based on the syndrome type name, the syndrome type description, the main syndrome elements, and the secondary syndrome elements, obtain the target feature vector of the to-be-classified syndrome type.

[0040] In this embodiment, the syndrome type name, syndrome type description, and the main and secondary syndrome elements predicted by the syndrome element prediction model are vectorized by features, so as to facilitate feature matching between the to-be-classified syndrome type and known syndrome types.

[0041] In an exemplary embodiment, the text such as the syndrome type name, syndrome type description, main syndrome elements, and secondary syndrome elements can be vectorized by text feature vectorization methods. Text vectorization is to represent text as a real number vector recognizable by a computer. According to different granularity sizes, text features can be represented at levels such as characters, words, sentences or chapters. The methods of text vectorization are mainly divided into discrete representation and distributed representation.

[0042] In specific embodiments, discrete representation is a rule-based and statistical vectorization method, including word set models (such as one-hot encoded text vectorization methods) and bag-of-words models (such as TF-IDF text vectorization methods). These methods are based on the premise that words are independent and unrelated, forming a dictionary of all words in the text, and then counting the frequency of each word based on the dictionary. Its fundamental characteristic is that it ignores word order and contextual information in the text, reflecting it only as several independent concepts.

[0043] In specific embodiments, distributed representation refers to mapping each word from a high dimension to a low-dimensional, dense vector based on its context. The dimension of the vector needs to be specified. In the constructed vector space, the meaning of each word can be represented by its surrounding words. The advantage is that it takes into account the similarity relationships between words and reduces the dimension of the word vectors. Examples include matrix-based distributed representation, clustering-based distributed representation, and neural network-based distributed representation. For instance, the Word2vec method can vectorize all words, making the relationships between words quantitatively measurable. Word2vec is based on word embeddings and uses deep learning to predict words appearing in the context. The word vectors trained by Word2vec can effectively measure the pixel-like similarity between words. By projecting all words into a k-dimensional vector space, each word can be represented by a k-dimensional vector, thus simplifying the processing of text content into vector operations in a k-dimensional vector space.

[0044] Step S103: Based on the known feature vectors corresponding to known certificate types in the preset certificate type library, perform similarity matching on the target feature vector to determine the certificate type category of the certificate type to be classified.

[0045] In this embodiment, for any one or more new certificate types (including certificate type name and certificate type description), if there is a certificate type that has a high degree of overlap in name characters and certificate element category, they can be considered to be the same certificate type.

[0046] In one exemplary implementation, the similarity between the target feature vector of the certificate type to be classified and the known feature vector of the known certificate type can be calculated and sorted to find the top N candidate known certificate types that are closest to the certificate type to be classified in the known certificate type library. These candidate known certificate types are used as the classification basis to classify the certificate type to be classified and obtain the certificate type category of the certificate type to be classified.

[0047] Furthermore, in this embodiment, step S103 specifically includes:

[0048] Based on the similarity matching results between the known feature vector and the target feature vector, at least one candidate matching certificate type is determined from the known certificate types;

[0049] Based on the similarity ranking results between the certificate type to be classified and the candidate matching certificate type, and the certificate type category of the candidate matching certificate type, the certificate type category of the certificate type to be classified is determined.

[0050] In this embodiment, for the same syndrome type, different classification results or different syndrome types may exist depending on different classification methods, and there may also be multiple identical symptoms. In this case, multiple classification results can be set when predicting the syndrome type to be classified.

[0051] In one exemplary implementation, the evidence prediction model can be configured to output multiple classification results, and these results can be sorted according to their similarity ranking. The category of the known evidence type with the highest similarity to the evidence type to be classified can be chosen as the classification category for the evidence type to be classified; alternatively, a specific category of a known evidence type can be selected and assigned as the classification category for the evidence type to be classified, based on user needs.

[0052] This embodiment provides a syndrome classification method based on syndrome element decomposition. This method analyzes the syndrome name and description of the syndrome to be classified using a syndrome element prediction model, decomposing and predicting the primary and secondary syndrome elements of the syndrome to be classified as symptom features. By retrieving known feature vectors of known syndromes, similarity matching is performed on the target feature vector of the syndrome to be classified, thereby finding known syndromes with similar features to the syndrome to be classified in the known syndrome database, thus completing the classification of the syndrome to be classified. Through the above method, by decomposing the syndrome elements to obtain the feature vector of the syndrome to be classified, and using syndrome elements as the essence of the syndrome for syndrome matching, it is possible to match syndromes with the same symptoms but different naming standards, improving the accuracy of syndrome matching and solving the technical problem of low accuracy in syndrome classification under different standards.

[0053] Please refer to Figure 3 , Figure 3 This is a flowchart illustrating a second embodiment of a syndrome type classification method based on syndrome element decomposition provided in this application.

[0054] In this embodiment, based on the above... Figure 2 In the illustrated embodiment, prior to step S101, the method further includes:

[0055] Step S201: Obtain symptom corpus data, and pre-train the language model based on the symptom corpus data to obtain a basic pre-trained model.

[0056] In this embodiment, large-scale texts from the field of TCM diagnosis can be used as symptom corpus data to train the language model, thereby generating a basic pre-trained model.

[0057] The symptom corpus data can be a collection of textual materials such as books, papers, and guidelines in the field of Traditional Chinese Medicine. Model learning and training refers to the process of estimating model parameters based on a given sample set / corpus database. Through this corpus data, the model can learn better language representations and improve the performance of downstream tasks. Language pre-training models can include models like Word2Vec, CBOW, and GloVe, which train word vectors from unlabeled corpora; models like ELMO, which extract context-dependent word vectors; or models like ULMFiT, which use fine-tuned pre-trained models for text classification.

[0058] Step S202: Obtain consultation business data, classify the consultation business data into syndrome types based on the preset recognition algorithm and expert prior knowledge, obtain symptom entity set and syndrome element dataset, and obtain the correlation classification interval based on the symptom entity set and the syndrome element dataset.

[0059] In this embodiment, symptom entities and syndrome elements are extracted by text recognition of consultation business data, thereby generating symptom entity datasets and syndrome element datasets, and thus mining the correlation between symptom entities and syndrome elements.

[0060] In one exemplary implementation, a named entity recognition algorithm can be used to extract symptom entities from the text of business data. Each consultation record can yield a list of symptom entities: e i Each medical record includes the doctor's diagnosis type: s i ;A list of syndrome elements can be obtained based on the syndrome element breakdown results of TCM experts: f i E and F are considered as the pairwise co-occurrence of symptoms and syndrome elements. The entire business consultation dataset is traversed, and the point mutual information method is used to calculate the correlation between each symptom and each syndrome element, which is described as an E×F real number matrix, where E is all symptoms and F is all syndrome elements.

[0061] Co-occurrence refers to the phenomenon where information described by feature terms appears together. Co-occurrence analysis is a quantitative study of co-occurrence phenomena to reveal the content association of information and the knowledge implied by feature terms.

[0062] Point mutual information refers to the statistical method of calculating the probability of two words appearing simultaneously in a text; the higher the probability, the stronger the correlation between the two. Its expression can be understood as:

[0063]

[0064] Where P refers to word frequency and PMI refers to co-occurrence logs. When PMI > 0, the two words are related, and the larger the PMI value, the stronger the correlation; when PMI = 0, the two words are statistically independent, neither related nor mutually exclusive; when PMI < 0, the two words are unrelated and mutually exclusive.

[0065] In one exemplary implementation, the classification intervals can be divided according to different degrees. For example, six classification intervals can be classified into six different degrees based on their correlation: definite symptoms, characteristic symptoms, common symptoms, primary correlated symptoms, secondary correlated symptoms, and occasional symptoms. This allows for a detailed breakdown of the syndrome elements and symptoms, improving classification accuracy and providing more basis for classification. Syndromes can be classified according to different needs.

[0066] Step S203: Based on the correlation classification interval, train the basic pre-trained model to obtain the ranking model, and fine-tune the parameters of the ranking model based on the backpropagation mechanism to obtain the evidence prediction model.

[0067] In this embodiment, the model can be further trained using the ranking learning algorithm based on the obtained pre-trained model to obtain the ranking model. During the training process of the ranking model, the parameters of the basic model are fine-tuned through the backpropagation mechanism to obtain the fine-tuned pre-trained model, which serves as the syndrome element prediction model. The syndrome element prediction model has captured the correlation features between symptoms and syndrome elements through parameter fine-tuning.

[0068] Backpropagation (BP) is a learning algorithm applicable to multi-layer neural networks. The BP algorithm's learning process consists of forward propagation and backward propagation. During forward propagation, input information is processed layer by layer through the input layer and hidden layers before being passed to the output layer. If the expected output value is not obtained at the output layer, the sum of squared errors between the output and the expected value is taken as the objective function. Backpropagation then begins, calculating the partial derivatives of the objective function with respect to the weights of each neuron layer. This forms the gradient of the objective function with respect to the weight vector, which serves as the basis for modifying the weights. The network's learning is completed during this weight modification process. The network learning ends when the error reaches the expected value.

[0069] The backpropagation process mainly consists of two iterative cycles: incentive propagation and weight update, until the network's response to the input reaches the predetermined target range.

[0070] The stimulus propagation process in each iteration consists of two steps: forward propagation and backward propagation. The forward propagation phase involves feeding the training input into the network to obtain the stimulus response; the backward propagation phase involves subtracting the stimulus response from the target output corresponding to the training input to obtain the response errors of the hidden and output layers.

[0071] Weight update involves multiplying the input stimulus and the corresponding error to obtain the gradient of the weights. This gradient is then multiplied by a certain ratio, inverted, and added to the weights. This ratio affects the speed and effectiveness of the training process and is therefore called the "training factor." The direction of the gradient indicates the direction of error amplification; therefore, it needs to be inverted when updating the weights to reduce the error caused by the weights.

[0072] Furthermore, based on the above Figure 3 In the illustrated embodiment, step S202 specifically includes:

[0073] Based on the preset recognition algorithm, identify the syndrome type dataset and the symptom entity set in the consultation business data;

[0074] Based on the expert prior knowledge, the certificate types in the certificate type dataset are split into certificate elements to obtain the certificate element dataset.

[0075] Based on the symptom entity set and the syndrome element dataset, the correlation ranking between each symptom and each syndrome element is obtained, and based on the correlation ranking, the correlation between the symptom and syndrome element is classified into levels to obtain the correlation classification interval.

[0076] In this embodiment, the preset recognition algorithm can be a named entity recognition algorithm. The named entity recognition algorithm extracts symptom entities from the text of the consultation business data. Each consultation record can extract a list of symptom entities. Each consultation record has a syndrome type given by the physician. A list of syndrome elements can be obtained based on the syndrome element decomposition results of the syndrome type by TCM experts. The output results of symptom entities, syndrome types and syndrome elements can also be in the form of groups, tables, sets, etc.

[0077] In one exemplary implementation, the point mutual information method can be used to calculate the correlation between each feature in the symptom entity set and the syndrome element set, and to statistically determine the correlation between each symptom and each syndrome element. After completing the syndrome element correlation statistics, the symptoms of each syndrome element can be sorted according to the correlation magnitude, and the entire numerical interval can be divided into N quantiles. For example, if N=6, the numerical interval can be divided into 6 classification intervals.

[0078] Furthermore, based on the above Figure 3 In the illustrated embodiment, step S203 specifically includes:

[0079] Based on the level division of the correlation, the correlation classification intervals are scored and assigned to obtain the correlation score corresponding to each correlation classification interval.

[0080] Based on the correlation classification intervals corresponding to the correlation between each symptom and each syndrome element, and the correlation scores, the correlation scores between each symptom and each syndrome element are determined.

[0081] The ranking model is obtained by training the basic pre-trained model based on the symptom entity set, the evidence element dataset, and the correlation score.

[0082] In this embodiment, the correlation values ​​can be classified into levels based on the correlation calculation results. According to the ranking of the correlation magnitude, different correlation classification intervals can be scored and assigned. Then, the pre-trained model can be trained to rank the intervals, so that the pre-trained model can learn the corresponding interval ranking and obtain the ranking model. This allows the ranking model to output a set of correlation scores between symptoms and syndrome elements for any set of input symptoms and syndrome elements.

[0083] In one exemplary implementation, for the relevance classification intervals, corresponding values ​​can be assigned to intervals of different degrees. For example, if the intervals are divided into 6 categories, scores of 1-6 can be assigned sequentially according to the relevance from smallest to largest. For instance, "anxiety" is a "primary relevance symptom" for the syndrome element "mental spirit." Therefore, the combination {anxiety, mental spirit} = 3, meaning that when "anxiety" and "mental spirit" are input, the ranking model can output a relevance score of 3. However, "anxiety" and "wind" are not closely related, so {anxiety, wind} = 0.

[0084] Please refer to Figure 4 , Figure 4 This is a flowchart illustrating a third embodiment of a syndrome type classification method based on syndrome element decomposition provided in this application.

[0085] Based on the above Figure 2 In the illustrated embodiment, step S101 specifically includes:

[0086] Step S301: Based on the evidence element prediction model, the evidence type name, and the evidence type description, perform an initial prediction on the evidence type to be classified to obtain the main evidence elements of the evidence type to be classified.

[0087] Step S302: Based on the evidence element prediction model, the evidence type name, the evidence type description, and the main evidence elements of the evidence type to be classified, the evidence type to be classified is re-predicted to obtain the secondary evidence elements of the evidence type to be classified.

[0088] In this embodiment, in order to enable the AI ​​model to more accurately decompose the evidence type, a waterfall-style training and prediction method can be adopted to deeply decompose the evidence elements of the evidence type.

[0089] In one exemplary implementation, the evidence element prediction model, whose parameters have been fine-tuned through backpropagation, can be used as the base model and trained in two stages. In the first stage, the training input includes the evidence name, evidence type description, and the main evidence elements of the evidence type. In the second stage, the input from the first stage (evidence name and evidence type description), the secondary evidence elements of the evidence type, and the evidence element prediction results from the first stage are input into the model for training.

[0090] The first stage of the model aims to learn the main evidence elements, while the second stage aims to deduce auxiliary evidence elements based on their relationship with the main evidence elements.

[0091] Optionally, both stages of the model can be optimized using multi-label classification loss, with the goal of making the types of evidence elements output by the model as close as possible to the types of evidence elements split by the experts.

[0092] Please refer to Figure 5 , Figure 5 This is a flowchart illustrating the fourth embodiment of a syndrome type classification method based on syndrome element decomposition provided in this application.

[0093] Based on the above Figure 2 In the embodiment shown, step S101 further includes:

[0094] Step S401: Based on the evidence element prediction model, the evidence type name, and the evidence type description, extract the target evidence element of the evidence type to be classified.

[0095] Step S402: Based on the syndrome element prediction model, score and predict the correlation between the symptoms corresponding to the syndrome to be classified and the target syndrome element to obtain the correlation score of each target syndrome element.

[0096] Step S403: Based on the correlation scores of each target evidence element, determine the main evidence element of the evidence type to be classified, and perform autoregressive training on the evidence element prediction model based on the evidence type name, the evidence type description and the main evidence element of the evidence type to be classified, to obtain the secondary evidence element of the evidence type to be classified.

[0097] In this embodiment, the target syndrome elements in the syndrome name and syndrome description can be extracted by the syndrome element prediction model. Based on the symptoms of the syndrome to be classified, the relevance of each target syndrome element is scored. Based on the relevance score, the extracted target syndrome elements are ranked by similarity, thereby extracting the main syndrome elements from the target syndrome elements, and then extracting the related secondary syndrome elements based on the main syndrome elements.

[0098] In one exemplary implementation, the top-ranking target evidence elements with the highest similarity can be selected as primary evidence elements, such as the top three target evidence elements with the highest similarity, and the rest can be selected as secondary evidence elements. Then, based on the primary evidence elements, other relevant evidence elements can be extracted from the evidence type name and evidence type description as secondary evidence elements.

[0099] In one exemplary implementation, target evidence elements that reach a preset similarity standard can be used as primary evidence elements, such as target evidence elements with a similarity of 85% or higher. Those below this similarity standard are used as secondary evidence elements. Then, based on the primary evidence elements, other relevant evidence elements are extracted from the evidence type name and evidence type description as secondary evidence elements.

[0100] Please refer to Figure 6 , Figure 6 This is a schematic block diagram of a pattern classification device based on pattern element decomposition provided in an embodiment of this application. This pattern classification device is used to execute the aforementioned pattern classification method based on pattern element decomposition. The pattern classification device can be configured in a server and a terminal.

[0101] like Figure 6 As shown, the syndrome type classification device 500 based on syndrome element decomposition includes: syndrome element prediction module 501, feature vector acquisition module 502, and category prediction module 503.

[0102] The evidence element prediction module 501 is used to obtain the evidence type name and evidence type description of the evidence type to be classified, and to perform evidence element prediction on the evidence type to be classified based on the evidence element prediction model, the evidence type name and the evidence type description, so as to obtain the main evidence element and secondary evidence element of the evidence type to be classified.

[0103] The feature vector acquisition module 502 is used to obtain the target feature vector of the certificate type to be classified based on the certificate type name, the certificate type description, the main certificate element and the secondary certificate element;

[0104] The category prediction module 503 is used to perform similarity matching on the target feature vector based on the known feature vectors corresponding to known certificate types in the preset certificate type library, and to determine the certificate type category of the certificate type to be classified.

[0105] In one embodiment, the syndrome type classification device 500 based on syndrome element decomposition further includes a syndrome element prediction model training module, used to acquire symptom corpus data, pre-train a language model based on the symptom corpus data to obtain a basic pre-trained model; acquire consultation business data, classify the consultation business data into syndrome types based on a preset recognition algorithm and expert prior knowledge to obtain a symptom entity set and a syndrome element dataset, and obtain a correlation classification interval based on the symptom entity set and the syndrome element dataset; train the basic pre-trained model based on the correlation classification interval to obtain a ranking model, and fine-tune the parameters of the ranking model based on a backpropagation mechanism to obtain the syndrome element prediction model.

[0106] In one embodiment, the syndrome element prediction model training module is further configured to: identify the syndrome type dataset and the symptom entity set in the consultation business data based on the preset recognition algorithm; perform syndrome element splitting on the syndrome types in the syndrome type dataset based on the expert prior knowledge to obtain the syndrome element dataset; obtain the correlation ranking between each symptom and each syndrome element based on the symptom entity set and the syndrome element dataset; and classify the correlation between symptoms and syndrome elements according to the correlation ranking to obtain the correlation classification interval.

[0107] In one embodiment, the syndrome element prediction model training module is further configured to: assign scores to the correlation classification intervals based on the correlation level division to obtain the correlation score corresponding to each correlation classification interval; determine the correlation score between each symptom and each syndrome element based on the correlation classification intervals corresponding to the correlation between each symptom and each syndrome element and the correlation score; and train the basic pre-trained model based on the symptom entity set, the syndrome element dataset, and the correlation score to obtain the ranking model.

[0108] In one embodiment, the category prediction module 503 is further configured to determine at least one candidate matching certificate type among the known certificate types based on the similarity matching result of the known feature vector and the target feature vector; and to determine the certificate type category of the certificate type to be classified based on the similarity ranking result between the certificate type to be classified and the candidate matching certificate type and the certificate type category of the candidate matching certificate type.

[0109] In one embodiment, the evidence element prediction module 501 is further configured to perform an initial prediction on the evidence type to be classified based on the evidence element prediction model, the evidence type name, and the evidence type description to obtain the main evidence element of the evidence type to be classified; and to perform a re-prediction on the evidence type to be classified based on the evidence element prediction model, the evidence type name, the evidence type description, and the main evidence element of the evidence type to be classified to obtain the secondary evidence element of the evidence type to be classified.

[0110] In one embodiment, the syndrome element prediction module 501 is further configured to: extract target syndrome elements of the syndrome to be classified based on the syndrome element prediction model, the syndrome name, and the syndrome description; predict the correlation between the symptoms corresponding to the syndrome to be classified and the target syndrome elements based on the syndrome element prediction model to obtain the correlation score of each target syndrome element; determine the main syndrome elements of the syndrome to be classified based on the correlation scores of each target syndrome element; and perform autoregressive training on the syndrome element prediction model based on the syndrome name, the syndrome description, and the main syndrome elements of the syndrome to be classified to obtain the secondary syndrome elements of the syndrome to be classified.

[0111] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the device and each module described above can be referred to the corresponding processes in the aforementioned embodiments of the syndrome type classification method based on syndrome element decomposition, and will not be repeated here.

[0112] The apparatus provided in the above embodiments can be implemented as a computer program, which can be used in, for example... Figure 7 It runs on the computer device shown.

[0113] Please see Figure 7 , Figure 7 This is a schematic block diagram illustrating the structure of a computer device according to an embodiment of this application. The computer device may be a terminal.

[0114] See Figure 7 The computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0115] Non-volatile storage media can store operating systems and computer programs. These computer programs include program instructions that, when executed, cause the processor to perform any evidence type classification method based on evidence element decomposition.

[0116] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0117] Internal memory provides an environment for the execution of computer programs stored in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any type classification method based on element decomposition.

[0118] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 7The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0119] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0120] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps:

[0121] Obtain the certificate type name and certificate type description of the certificate type to be classified. Based on the certificate element prediction model, the certificate type name and the certificate type description, perform certificate element prediction on the certificate type to be classified to obtain the main certificate element and secondary certificate element of the certificate type to be classified.

[0122] Based on the certificate type name, the certificate type description, the main certificate elements, and the secondary certificate elements, the target feature vector of the certificate type to be classified is obtained.

[0123] Based on the known feature vectors corresponding to known certificate types in the preset certificate type library, similarity matching is performed on the target feature vectors to determine the certificate type category of the certificate type to be classified.

[0124] In one embodiment, before implementing the steps of obtaining the certificate type name and certificate type description of the certificate type to be classified, and performing certificate element prediction on the certificate type to be classified based on the certificate element prediction model, the certificate type name, and the certificate type description to obtain the primary and secondary certificate elements of the certificate type to be classified, the processor is further configured to implement:

[0125] Acquire symptom corpus data, and pre-train the language model based on the symptom corpus data to obtain a basic pre-trained model;

[0126] Acquire consultation business data, classify the consultation business data into syndrome types based on preset recognition algorithms and expert prior knowledge, obtain symptom entity set and syndrome element dataset, and obtain correlation classification interval based on the symptom entity set and the syndrome element dataset;

[0127] Based on the aforementioned correlation classification interval, the basic pre-trained model is trained to obtain a ranking model, and the parameters of the ranking model are fine-tuned based on the backpropagation mechanism to obtain the evidence prediction model.

[0128] In one embodiment, when the processor acquires consultation data, performs syndrome type classification on the consultation data based on a preset recognition algorithm and expert prior knowledge to obtain a symptom entity set and a syndrome element dataset, and obtains a correlation classification interval based on the symptom entity set and the syndrome element dataset, it is configured to:

[0129] Based on the preset recognition algorithm, identify the syndrome type dataset and the symptom entity set in the consultation business data;

[0130] Based on the expert prior knowledge, the certificate types in the certificate type dataset are split into certificate elements to obtain the certificate element dataset.

[0131] Based on the symptom entity set and the syndrome element dataset, the correlation ranking between each symptom and each syndrome element is obtained, and based on the correlation ranking, the correlation between the symptom and syndrome element is classified into levels to obtain the correlation classification interval.

[0132] In one embodiment, when the processor trains the basic pre-trained model based on the relevance classification interval to obtain the ranking model, it is configured to:

[0133] Based on the level division of the correlation, the correlation classification intervals are scored and assigned to obtain the correlation score corresponding to each correlation classification interval.

[0134] Based on the correlation classification intervals corresponding to the correlation between each symptom and each syndrome element, and the correlation scores, the correlation scores between each symptom and each syndrome element are determined.

[0135] The ranking model is obtained by training the basic pre-trained model based on the symptom entity set, the evidence element dataset, and the correlation score.

[0136] In one embodiment, when the processor performs similarity matching on the target feature vector based on the known feature vectors corresponding to known certificate types in a preset certificate type library to determine the certificate type category, it is configured to:

[0137] Based on the similarity matching results between the known feature vector and the target feature vector, at least one candidate matching certificate type is determined from the known certificate types;

[0138] Based on the similarity ranking results between the certificate type to be classified and the candidate matching certificate type, and the certificate type category of the candidate matching certificate type, the certificate type category of the certificate type to be classified is determined.

[0139] In one embodiment, when the processor implements the evidence element prediction model, the evidence type name, and the evidence type description to predict the evidence element of the evidence type to be classified and obtain the main evidence element and secondary evidence element of the evidence type to be classified, it is configured to:

[0140] Based on the evidence element prediction model, the evidence type name, and the evidence type description, a preliminary prediction is made on the evidence type to be classified, and the main evidence elements of the evidence type to be classified are obtained.

[0141] Based on the evidence element prediction model, the evidence type name, the evidence type description, and the main evidence elements of the evidence type to be classified, the evidence type to be classified is re-predicted to obtain the secondary evidence elements of the evidence type to be classified.

[0142] In one embodiment, when the processor implements the evidence element prediction model, the evidence type name, and the evidence type description to predict the evidence element of the evidence type to be classified and obtain the primary and secondary evidence elements of the evidence type to be classified, it is also configured to implement:

[0143] Based on the evidence element prediction model, the evidence type name, and the evidence type description, the target evidence element of the evidence type to be classified is extracted.

[0144] Based on the syndrome element prediction model, the correlation between the symptoms corresponding to the syndrome to be classified and the target syndrome element is scored and predicted to obtain the correlation score of each target syndrome element.

[0145] Based on the correlation scores of each target evidence element, the main evidence element of the evidence type to be classified is determined, and based on the evidence type name, the evidence type description and the main evidence element of the evidence type to be classified, the evidence element prediction model is trained by autoregression to obtain the secondary evidence element of the evidence type to be classified.

[0146] The embodiments of this application also provide a computer-readable storage medium storing a computer program, the computer program including program instructions, and the processor executing the program instructions to implement any of the evidence type classification methods based on evidence element decomposition provided in the embodiments of this application.

[0147] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0148] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A syndrome type classification method based on syndrome element decomposition, characterized in that, include: Obtain the certificate type name and certificate type description of the certificate type to be classified. Based on the certificate element prediction model, the certificate type name and the certificate type description, perform certificate element prediction on the certificate type to be classified to obtain the main certificate element and secondary certificate element of the certificate type to be classified. Based on the certificate type name, the certificate type description, the main certificate elements, and the secondary certificate elements, the target feature vector of the certificate type to be classified is obtained. Based on the known feature vectors corresponding to known certificate types in the preset certificate type library, similarity matching is performed on the target feature vectors to determine the certificate type category of the certificate type to be classified. The step of predicting the primary and secondary evidence elements of the evidence type based on the evidence element prediction model, the evidence type name, and the evidence type description, and obtaining the primary and secondary evidence elements of the evidence type to be classified, includes: The syndrome type name and description are input into the syndrome element prediction model for the first decomposition prediction, and the result of this prediction is taken as the primary syndrome element. Then, the primary syndrome element, syndrome type name, and syndrome type description are input into the syndrome element prediction model again. Using the primary syndrome element as the decomposition basis, the syndrome type name and syndrome type description are split, and the elements that are related to the primary syndrome element are extracted as secondary syndrome elements. Then, using these as the decomposition basis, they are input into the syndrome element prediction model for the next syndrome element prediction, until there is no result output or the output result is invalid.

2. The syndrome type classification method based on syndrome element decomposition according to claim 1, characterized in that, Before obtaining the certificate type name and certificate type description of the certificate type to be classified, and performing certificate element prediction on the certificate type to be classified based on the certificate element prediction model, the certificate type name, and the certificate type description to obtain the main certificate element and secondary certificate element of the certificate type to be classified, the method further includes: Acquire symptom corpus data, and pre-train the language model based on the symptom corpus data to obtain a basic pre-trained model; Acquire consultation business data, classify the consultation business data into syndrome types based on preset recognition algorithms and expert prior knowledge, obtain symptom entity set and syndrome element dataset, and obtain correlation classification interval based on the symptom entity set and the syndrome element dataset; Based on the aforementioned correlation classification interval, the basic pre-trained model is trained to obtain a ranking model, and the parameters of the ranking model are fine-tuned based on the backpropagation mechanism to obtain the evidence prediction model.

3. The syndrome type classification method based on syndrome element decomposition according to claim 2, characterized in that, The process of acquiring consultation data involves classifying the consultation data into syndrome types based on a preset recognition algorithm and expert prior knowledge, obtaining a symptom entity set and a syndrome element dataset, and obtaining a correlation classification interval based on the symptom entity set and the syndrome element dataset, including: Based on the preset recognition algorithm, identify the syndrome type dataset and the symptom entity set in the consultation business data; Based on the expert prior knowledge, the certificate types in the certificate type dataset are split into certificate elements to obtain the certificate element dataset. Based on the symptom entity set and the syndrome element dataset, the correlation ranking between each symptom and each syndrome element is obtained, and based on the correlation ranking, the correlation between the symptom and syndrome element is classified into levels to obtain the correlation classification interval.

4. The syndrome type classification method based on syndrome element decomposition according to claim 3, characterized in that, The step of training the basic pre-trained model based on the relevance classification interval to obtain the ranking model includes: Based on the level division of the correlation, the correlation classification intervals are scored and assigned to obtain the correlation score corresponding to each correlation classification interval. Based on the correlation classification intervals corresponding to the correlation between each symptom and each syndrome element, and the correlation scores, the correlation scores between each symptom and each syndrome element are determined. The ranking model is obtained by training the basic pre-trained model based on the symptom entity set, the evidence element dataset, and the correlation score.

5. The syndrome type classification method based on syndrome element decomposition according to claim 1, characterized in that, The step of performing similarity matching on the target feature vector based on the known feature vectors corresponding to known certificate types in a preset certificate type library to determine the certificate type category includes: Based on the similarity matching results between the known feature vector and the target feature vector, at least one candidate matching certificate type is determined from the known certificate types; Based on the similarity ranking results between the certificate type to be classified and the candidate matching certificate type, and the certificate type category of the candidate matching certificate type, the certificate type category of the certificate type to be classified is determined.

6. The syndrome type classification method based on syndrome element decomposition according to claim 1, characterized in that, The method of predicting the primary and secondary evidence elements of the evidence type based on the evidence element prediction model, the evidence type name, and the evidence type description, and obtaining the primary and secondary evidence elements of the evidence type to be classified, includes: Based on the evidence element prediction model, the evidence type name, and the evidence type description, a preliminary prediction is made on the evidence type to be classified, and the main evidence elements of the evidence type to be classified are obtained. Based on the evidence element prediction model, the evidence type name, the evidence type description, and the main evidence elements of the evidence type to be classified, the evidence type to be classified is re-predicted to obtain the secondary evidence elements of the evidence type to be classified.

7. The syndrome type classification method based on syndrome element decomposition according to claim 1, characterized in that, The step of predicting the primary and secondary evidence elements of the evidence type based on the evidence element prediction model, the evidence type name, and the evidence type description, to obtain the primary and secondary evidence elements of the evidence type to be classified, further includes: Based on the evidence element prediction model, the evidence type name, and the evidence type description, the target evidence element of the evidence type to be classified is extracted. Based on the syndrome element prediction model, the correlation between the symptoms corresponding to the syndrome to be classified and the target syndrome element is scored and predicted to obtain the correlation score of each target syndrome element. Based on the correlation scores of each target evidence element, the main evidence element of the evidence type to be classified is determined, and based on the evidence type name, the evidence type description and the main evidence element of the evidence type to be classified, the evidence element prediction model is trained by autoregression to obtain the secondary evidence element of the evidence type to be classified.

8. A pattern classification device based on pattern element decomposition, characterized in that, include: The evidence element prediction module is used to obtain the evidence type name and evidence type description of the evidence type to be classified, and to perform evidence element prediction on the evidence type to be classified based on the evidence element prediction model, the evidence type name and the evidence type description, so as to obtain the main evidence element and secondary evidence element of the evidence type to be classified. The evidence element prediction module is further used to input the evidence type name and evidence type description into the evidence element prediction model for the first decomposition prediction, and take the result of this prediction as the primary evidence element; then input the primary evidence element, evidence type name, and evidence type description into the evidence element prediction model again, using the primary evidence element as the decomposition basis, splitting the evidence type name and evidence type description, and extracting the elements that are related to the primary evidence element as secondary evidence elements; then using this as the decomposition basis again, inputting it into the evidence element prediction model for the next evidence element prediction, until there is no result output or the output result is invalid; The feature vector acquisition module is used to obtain the target feature vector of the certificate type to be classified based on the certificate type name, the certificate type description, the main certificate element and the secondary certificate element; The category prediction module is used to perform similarity matching on the target feature vector based on the known feature vectors corresponding to known certificate types in the preset certificate type library, and to determine the certificate type category of the certificate type to be classified.

9. A computer device, characterized in that, The computer device includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, it implements the steps of the evidence type classification method based on evidence element decomposition as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, it implements the steps of the evidence type classification method based on evidence element decomposition as described in any one of claims 1 to 7.

Images