Intelligent syndrome differentiation information processing system of traditional Chinese medicine based on big data

Abstract

The present application relates to the field of big data information processing, and provides a TCM intelligent syndrome differentiation information processing system based on big data, which obtains a set of symptom descriptions and a set of syndrome element descriptions of the diagnosed patient object, and uses a text description analysis technology based on big data and artificial intelligence to perform semantic coding and feature association of the set of symptom descriptions and the set of syndrome element descriptions of the diagnosed patient object, thereby intelligently obtaining the inference result of the syndrome name type label based on the implicit semantic association between the contexts of the various symptom descriptions and the fusion features between the semantic features of the various syndrome element descriptions. In this way, the system can intelligently infer the syndrome name type label that conforms to individual differences based on the patient's specific symptom description and syndrome element description, reducing the differences caused by human factors, thereby achieving personalized diagnosis.

Description

Technical Field

[0001] The present application relates to the field of big data information processing, and more specifically, to a big data-based intelligent Chinese medicine syndrome differentiation information processing system. Background Art

[0002] Syndrome differentiation is the core link in the diagnosis and treatment system of traditional Chinese medicine. It provides patients with a comprehensive and personalized health management and disease treatment method by comprehensively analyzing the patient's symptoms and signs to determine the cause, nature and location of the disease.

[0003] However, in the current research on TCM syndrome differentiation, most of the methods are based on fixed quantitative values ​​of symptoms and syndrome elements, and then combined with a general algorithm of quantitative values ​​to obtain syndrome names. However, due to the large number of TCM inheritance schools, each school has its own characteristics, and TCM syndrome differentiation relies heavily on the doctor's subjective experience and intuition, and the existing quantitative methods often cannot fully reflect these personalized experiences and knowledge.

[0004] Therefore, an intelligent TCM syndrome differentiation information processing system based on big data is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a TCM intelligent syndrome differentiation information processing system based on big data, which obtains a set of symptom descriptions and a set of syndrome element descriptions of the diagnosed patient objects, and uses text description analysis technology based on big data and artificial intelligence to perform semantic encoding and feature association of the set of symptom descriptions and the set of syndrome element descriptions of the diagnosed patient objects, thereby intelligently obtaining the inference result of the syndrome type label based on the implicit semantic association between the contexts of the various symptom descriptions and the fusion features between the semantic features of the various syndrome element descriptions. In this way, the system can intelligently infer the syndrome type label that conforms to individual differences based on the patient's specific symptom descriptions and syndrome element descriptions, reducing the differences caused by human factors, thereby achieving personalized diagnosis.

[0006] According to one aspect of the present application, a TCM intelligent syndrome differentiation information processing system based on big data is provided, which comprises:

[0007] A patient object symptom description acquisition module is used to acquire a set of symptom descriptions of diagnosed patient objects;

[0008] A patient object syndrome factor description acquisition module, used to acquire a set of syndrome factor descriptions of the diagnosed patient object;

[0009] A symptom description semantic coding module, for performing semantic embedding coding on each symptom description in the set of symptom descriptions to obtain a set of symptom description semantic embedding coding vectors, and then performing contextual semantic coding on the set of symptom description semantic embedding coding vectors to obtain a set of implicit semantic association feature vectors between contextual symptoms;

[0010] A contextual semantics enhancement module, used for inputting the set of implicit semantic association feature vectors between the contextual symptoms into the contextual semantics enhancement module based on significance-globality to obtain a set of enhanced implicit semantic association feature vectors between the contextual symptoms;

[0011] A certificate element description semantic coding module, used for performing semantic embedding coding on each certificate element description in the set of certificate element descriptions to obtain a set of certificate element description semantic embedding coding vectors;

[0012] A symptom-syndrome description semantic fusion module is used to integrate the set of implicit semantic association feature vectors between the enhanced contextual symptoms and the set of semantic embedding coding vectors of the syndrome description into a sequence of symptom-syndrome semantic fusion feature vectors, and then pass them through a context encoder based on the Transformer layer to obtain a semantic fusion coding feature vector of symptom information and syndrome information as a semantic fusion coding feature of symptom information and syndrome information;

[0013] The inference result generation module is used to obtain the inference result based on the semantic fusion coding features of the symptom information and the syndrome element information, and the inference result is used to represent the type label of the syndrome name.

[0014] In the above-mentioned big data-based intelligent Chinese medicine syndrome differentiation information processing system, the symptom description semantic encoding module includes: a semantic embedding encoding unit, which is used to pass each symptom description in the set of symptom descriptions through a semantic embedding encoder based on the Bert model to obtain a set of symptom description semantic embedding encoding vectors; and an implicit semantic association unit, which is used to input the set of symptom description semantic embedding encoding vectors into a context semantic encoder based on Bi-LSTM to obtain a set of implicit semantic association feature vectors between the context symptoms.

[0015] In the above-mentioned intelligent Chinese medicine syndrome differentiation information processing system based on big data, the context semantic enhancement module includes: a prominent feature calculation unit, which is used to extract the maximum value of the implicit semantic association feature vectors between each context symptom in the set of implicit semantic association feature vectors between context symptoms to obtain a prominent context symptom implicit semantic association feature vector; a global feature calculation unit, which is used to extract the average value of the implicit semantic association feature vectors between each context symptom in the set of implicit semantic association feature vectors between context symptoms to obtain a global context symptom implicit semantic association feature vector; a feature processing unit, which is used to perform convolution coding, nonlinear activation and linear transformation on the prominent context symptom implicit semantic association feature vector and the global context symptom implicit semantic association feature vector respectively to obtain a prominent feature context symptom implicit semantic association activation feature vector and a global context symptom implicit semantic association feature vector. A global feature context-symptom implicit semantic association activation feature vector; a feature fusion unit, used to fuse the salient feature context-symptom implicit semantic association activation feature vector and the global feature context-symptom implicit semantic association activation feature vector to obtain a global-salient feature context-symptom implicit semantic association feature vector, and then perform nonlinear activation on the global-salient feature context-symptom implicit semantic association feature vector to obtain a global-salient feature context-symptom implicit semantic association weight feature vector; a feature weighting unit, used to use the global-salient feature context-symptom implicit semantic association weight feature vector as a weight, perform positional dot multiplication on the set of context-symptom implicit semantic association feature vectors, and add the set of context-symptom implicit semantic association feature vectors to obtain the set of enhanced context-symptom implicit semantic association feature vectors.

[0016] In the above-mentioned big data-based intelligent Chinese medicine syndrome differentiation information processing system, the feature processing unit is used to: input the prominent context-symptom implicit semantic association convolution coding vector obtained by performing one-dimensional convolution coding on the prominent context-symptom implicit semantic association feature vector into the ReLU function to obtain the prominent context-symptom implicit semantic association activation coding feature vector; perform matrix multiplication of the prominent point convolution feature vector obtained by performing point convolution coding on the prominent context-symptom implicit semantic association activation coding feature vector with the first weight matrix to obtain the prominent feature context-symptom implicit semantic association activation feature vector; input the global context-symptom implicit semantic association convolution coding vector obtained by performing one-dimensional convolution coding on the global context-symptom implicit semantic association feature vector into the ReLU function to obtain the global context-symptom implicit semantic association activation coding feature vector; perform matrix multiplication of the global point convolution feature vector obtained by performing point convolution coding on the global context-symptom implicit semantic association activation coding feature vector with the second weight matrix to obtain the global feature context-symptom implicit semantic association activation feature vector.

[0017] In the above-mentioned big data-based intelligent Chinese medicine syndrome differentiation information processing system, the feature fusion unit is used to: perform feature fusion on the prominent feature context symptom implicit semantic association activation feature vector and the global feature context symptom implicit semantic association activation feature vector to obtain the global-prominent feature context symptom implicit semantic association feature vector; input the global-prominent feature context symptom implicit semantic association feature vector into the tanh function to obtain the first activated global-prominent feature context symptom implicit semantic association feature vector; input the global-prominent feature context symptom implicit semantic association feature vector into the sigmoid function to obtain the second activated global-prominent feature context symptom implicit semantic association feature vector; perform position point multiplication on the first activated global-prominent feature context symptom implicit semantic association feature vector and the second activated global-prominent feature context symptom implicit semantic association feature vector to obtain the global-prominent feature context symptom implicit semantic association weight feature vector.

[0018] In the above-mentioned big data-based intelligent Chinese medicine syndrome differentiation information processing system, the syndrome description semantic encoding module is used to: pass each syndrome description in the set of syndrome descriptions through the semantic embedding encoder based on the Bert model to obtain a set of syndrome description semantic embedding coding vectors.

[0019] In the above-mentioned TCM intelligent syndrome differentiation information processing system based on big data, the inference result generation module is used to: input the semantic fusion encoding feature vector of the symptom information and syndrome factor information into the classifier-based syndrome name dialectical reasoner to obtain the inference result.

[0020] The above-mentioned intelligent TCM syndrome differentiation information processing system based on big data also includes a training module for training the semantic embedding encoder based on the Bert model, the contextual semantic encoder based on Bi-LSTM, the contextual semantic reinforcement module based on significance-globality, the contextual encoder based on the Transformer layer and the syndrome name dialectical reasoner based on the classifier.

[0021] In the above-mentioned TCM intelligent syndrome differentiation information processing system based on big data, the training module includes: a training data acquisition unit, which is used to acquire training data, wherein the training data includes a set of training symptom descriptions of the diagnosed patient objects, a set of training syndrome descriptions of the diagnosed patient objects, and a true reasoning result, wherein the true reasoning result is a true type label of the syndrome name; a training symptom description semantic encoding unit, which is used to pass each training symptom description in the set of training symptom descriptions through the semantic embedding encoder based on the Bert model to obtain a set of training symptom description semantic embedding encoding vectors; a training symptom description context semantic association unit, which is used to input the set of training symptom description semantic embedding encoding vectors into the semantic embedding encoder based on the Bi-LSTM model. A context semantic encoder is used to obtain a set of implicit semantic association feature vectors between training context symptoms; a training symptom description context semantic reinforcement unit is used to input the set of implicit semantic association feature vectors between training context symptoms into the context semantic reinforcement module based on significance-globality to obtain a set of implicit semantic association feature vectors between training reinforcement context symptoms; a training syndrome description semantic encoding unit is used to pass each training syndrome description in the set of training syndrome descriptions through the semantic embedding encoder based on the Bert model to obtain a set of training syndrome description semantic embedding encoding vectors; a training information fusion semantic encoding unit is used to integrate the set of implicit semantic association feature vectors between training reinforcement context symptoms and the training syndrome description semantic embedding encoder. The set of encoding vectors is integrated into a sequence of training symptom-syndrome semantic fusion feature vectors and then passed through the Transformer layer-based context encoder to obtain the training symptom information and training syndrome information semantic fusion encoding feature vectors; a training reasoning result generating unit is used to input the training symptom information and training syndrome information semantic fusion encoding feature vectors into the classifier-based syndrome name dialectical reasoning device to obtain the training reasoning result; a classification loss unit is used to calculate the cross entropy loss function value between the training reasoning result and the true reasoning result to obtain the classification loss function value; a model training unit is used to train the Bert model-based semantic embedding encoder, the Bi-LSTM-based context semantic encoder, the saliency-global context semantic reinforcement module, the Transformer layer-based context encoder and the classifier-based syndrome name dialectical reasoning device based on the classification loss function value and through gradient descent back propagation.

[0022] Compared with the prior art, the TCM intelligent syndrome differentiation information processing system based on big data provided by the present application obtains a set of symptom descriptions and a set of syndrome element descriptions of the diagnosed patient object, and uses text description analysis technology based on big data and artificial intelligence to perform semantic coding and feature association of the set of symptom descriptions and the set of syndrome element descriptions of the diagnosed patient object, thereby intelligently obtaining the inference result of the syndrome type label based on the implicit semantic association between the contexts of the various symptom descriptions and the fusion features between the semantic features of the various syndrome element descriptions. In this way, the system can intelligently infer the syndrome type label that conforms to individual differences based on the patient's specific symptom description and syndrome element description, reducing the differences caused by human factors, thereby achieving personalized diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0024] Figure 1 This is a system block diagram of a big data-based TCM intelligent syndrome differentiation information processing system according to an embodiment of the present application.

[0025] Figure 2 This is a schematic diagram of the architecture of a big data-based TCM intelligent syndrome differentiation information processing system according to an embodiment of the present application.

[0026] Figure 3 It is a block diagram of the training module in the big data-based TCM intelligent syndrome differentiation information processing system according to an embodiment of the present application.

[0027] Figure 4 Schematic diagram of the algorithm derivation process according to an embodiment of the present application.

[0028] Figure 5 This is an example diagram of a symptom name library according to an embodiment of the present application.

[0029] Figure 6 This is an example diagram of a certificate element name library according to an embodiment of the present application.

[0030] Figure 7 This is an example diagram of a certificate name library according to an embodiment of the present application.

[0031] Figure 8 This is an example diagram of a weight relationship library according to an embodiment of the present application.

[0032] Fig. 9This is an example diagram of the relationship between certificate elements and certificate names according to an embodiment of the present application. DETAILED DESCRIPTION

[0033] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0034] It should be understood that big data usually refers to a large-scale and highly complex data set. With the development of technology, big data has become an important part of the digital economy era. In the intelligent syndrome differentiation information processing of traditional Chinese medicine, big data technology can process and analyze a large amount of clinical data of traditional Chinese medicine, including symptom description, syndrome description, tongue image, pulse image, etc., to realize the digitization and dataization of traditional Chinese medicine syndrome differentiation, so as to reveal disease patterns and optimize treatment plans.

[0035] Based on this, the present application proposes a TCM intelligent syndrome differentiation information processing system based on big data, which obtains a set of symptom descriptions and a set of syndrome element descriptions of the diagnosed patient object, and uses text description analysis technology based on big data and artificial intelligence to perform semantic coding and feature association of the set of symptom descriptions and the set of syndrome element descriptions of the diagnosed patient object, so as to intelligently obtain the inference result of the syndrome type label based on the implicit semantic association between the contexts of the various symptom descriptions and the fusion features between the semantic features of the various syndrome element descriptions. In this way, the system can intelligently infer the syndrome type label that conforms to individual differences based on the patient's specific symptom description and syndrome element description, reduce the differences caused by human factors, and thus achieve personalized diagnosis.

[0036] Figure 1 This is a system block diagram of a big data-based TCM intelligent syndrome differentiation information processing system according to an embodiment of the present application. Figure 2 FIG. 1 is a schematic diagram of the architecture of a TCM intelligent syndrome differentiation information processing system based on big data according to an embodiment of the present application. Figure 1 and Figure 2As shown, in the TCM intelligent syndrome differentiation information processing system 100 based on big data, it includes: a patient object symptom description acquisition module 110, which is used to obtain a set of symptom descriptions of diagnosed patient objects; a patient object syndrome element description acquisition module 120, which is used to obtain a set of syndrome element descriptions of the diagnosed patient objects; a symptom description semantic encoding module 130, which is used to perform semantic embedding encoding on each symptom description in the set of symptom descriptions to obtain a set of symptom description semantic embedding encoding vectors, and then perform context semantic encoding on the set of symptom description semantic embedding encoding vectors to obtain a set of implicit semantic association feature vectors between context symptoms; a context semantic enhancement module 140, which is used to input the set of implicit semantic association feature vectors between context symptoms into a context semantic enhancement module based on significance-globality to obtain an enhanced context semantic enhancement module. A set of implicit semantic association feature vectors between symptoms; a syndrome description semantic coding module 150, used to perform semantic embedding coding on each syndrome description in the set of syndrome descriptions to obtain a set of syndrome description semantic embedding coding vectors; a symptom syndrome description semantic fusion module 160, used to integrate the set of implicit semantic association feature vectors between the enhanced context symptoms and the set of syndrome description semantic embedding coding vectors into a sequence of symptom syndrome semantic fusion feature vectors and then pass them through a context encoder based on the Transformer layer to obtain a symptom information and syndrome information semantic fusion coding feature vector as a symptom information and syndrome information semantic fusion coding feature; an inference result generation module 170, used to obtain an inference result based on the symptom information and syndrome information semantic fusion coding features, and the inference result is used to represent the type label of the syndrome name.

[0037] In an embodiment of the present application, the patient object symptom description acquisition module 110 is used to acquire a set of symptom descriptions of the diagnosed patient object and the patient object syndrome element description acquisition module 120 is used to acquire a set of syndrome element descriptions of the diagnosed patient object. It should be understood that the patient's symptom description is the basis for doctors to carry out syndrome differentiation and treatment. Specifically, symptoms refer to abnormal manifestations or discomforts subjectively felt by patients, such as fever, cough, headache, etc., and objective physical signs observed by doctors through four diagnostic methods such as observation, auscultation, olfaction, questioning, and palpation, such as tongue image, pulse image, etc. The symptom description is the first-hand data of TCM diagnosis, which directly reflects the nature, location, degree of the disease and the patient's physical condition. Through the analysis of the symptom description, the doctor can preliminarily judge the cold and heat deficiency and excess of the pathogenesis, and the internal and external upper and lower parts of the disease, which provides a basis for subsequent syndrome differentiation and treatment. Syndrome element is the basic unit of TCM syndrome differentiation, which is a summary of the essence of the pathogenesis of the disease at the current stage. Syndrome description refers to summarizing and organizing the patient's symptoms, signs and other information to extract syndromes that reflect the essential characteristics of the disease, such as Qi deficiency, blood stasis, dampness and heat, etc. It is a deepening and refinement of TCM syndrome differentiation, and can more accurately grasp the nature and changing patterns of the disease. Therefore, in the technical solution of the present application, a set of symptom descriptions of the diagnosed patient objects is obtained, and a set of syndrome descriptions of the diagnosed patient objects is obtained. In other words, by analyzing the symptom descriptions and syndrome descriptions, relevant syndrome names can be obtained, which provides an important basis for doctors to formulate personalized treatment plans. Specifically, Figure 4 A specific example of an algorithm derivation process is provided, which first constructs a standard symptom name library, a syndrome name library, and a syndrome name library, wherein the symptom name library includes: dizziness, headache, irritability, flushed face, red wood, constipation, heavy head, nausea, backache, night sweats, tinnitus, chills, etc., the syndrome name library includes: liver, spleen, stomach, fire, heat, phlegm and dampness, yang deficiency, yin deficiency, yang hyperactivity, etc., and the syndrome name library includes: liver yang hyperactivity, liver fire hyperactivity, phlegm and dampness congestion, yin deficiency and yang hyperactivity, yin and yang deficiency, etc.; then constructs a symptom and syndrome weight relationship, and a syndrome and syndrome name composition relationship, wherein the relationship is composed by standard specifications, and the composed relationship is used to train the model of the present application; finally, an AI algorithm is used to deduce the syndrome based on the determined symptoms, and the syndrome name is determined according to the syndrome and syndrome type composition relationship.

[0038] It is worth mentioning that ordinary technicians in this field should know that before applying AI algorithms for syndrome name reasoning, the model needs to be trained through big data. Therefore, in a more specific application example, the "Standardization of Common Clinical Symptoms in Traditional Chinese Medicine" is used as the basic standard for the symptom name library to build a symptom name database. Taking into account the variability of clinical symptom terms, the symptom terms used in the "Differential Diagnosis of Symptoms in Traditional Chinese Medicine" and 5,000 real clinical medical records are combined to expand and map aliases. For specific examples, see Figure 5 .

[0039] Then, based on the syndrome theory, the 53 syndromes described in the syndrome theory are directly used as the standard of the syndrome name database to build a syndrome name database. For specific examples, see Figure 6 .

[0040] Finally, we adopted the Classification and Code of TCM Syndrome (GBT 15657-2021) as the basic standard for the syndrome name database, and constructed a syndrome name database. We also proofread the syndrome names in 5,000 real clinical medical records, and allowed for later expansion in consideration of clinical variability. For specific examples, see Figure 7 .

[0041] The weight relationship between symptoms and syndrome elements is the core of this method and the key to determining the final model. This step is the difference between our research methods and other research methods in the field of TCM syndrome differentiation. We train based on medical records. The recommended training medical record selection requirements are: 1. The more medical records, the more accurate it is, and it is recommended to exceed 2,000 cases; 2. Cover as many symptoms as possible; 3. Cover as many syndrome names as possible; 4. Cover as many patient categories as possible, involving men, women and all age groups. If the number of medical records is small and the dimensions covered are small, it is recommended to do it in a special disease way to ensure a certain degree of accuracy. We collected more than 5,000 clinical medical records, involving hundreds of diseases in multiple departments such as internal medicine, surgery, gynecology, pediatrics, and ENT, covering more than 700 common symptoms. The final syndrome differentiation accuracy rate and the similarity with manual syndrome differentiation reached more than 95%, which is completely inconsistent with the medical theory of zero syndrome differentiation.

[0042] Furthermore, in an application example of the present application, the collected medical record data is corrected, and the data of the chief complaint, current medical history, past medical history, four TCM examinations, physical examination, and auxiliary examinations are manually corrected. The text content is disassembled according to our standard symptom name library, and spliced ​​with ";" according to the standard symptom name, and the spliced ​​content is stored in the newly added field "symptom group"; the TCM syndrome name is also disassembled, and according to the rules of the syndrome element and the syndrome name, it is split into multiple syndrome elements, and spliced ​​with ";", and the spliced ​​content is stored in the newly added field "syndrome element group". All the corrected medical records are imported into the database.

[0043] Among them, weight relationship = symptom frequency x symptom weight x syndrome weight, explanation: weight relationship is the weight relationship of a symptom to a syndrome; symptom frequency is the frequency of a symptom in a syndrome; symptom weight is the posterior probability of a symptom in a syndrome; syndrome weight is the posterior probability of a syndrome in a symptom. Based on the weight relationship algorithm, calculate the relationship values ​​of all symptoms in the symptom library and syndrome in the syndrome library to form a weight relationship library, see Figure 8 .

[0044] The composition relationship between syndrome elements and syndrome names is the key to deriving the final result. Because the syndrome names we finally derived are based on the standard of "Classification and Codes of Traditional Chinese Medicine Diseases and Syndromes" (GBT 15657-2021), our composition rules are a reverse method. If the syndrome names are combined with multiple components of syndrome elements, then there will be thousands or tens of thousands of combinations of syndrome names, which is not conducive to standardization and even more unfavorable for clinical use. Therefore, we adopt a reverse method. 1. Using the standard syndrome name as a reference, extract the explicit syndrome elements from the syndrome name. 2. Based on manual verification by multiple experts, extract the invisible syndrome elements from the syndrome name. For examples of the composition relationship between syndrome elements and syndrome names, see Fig. 9 .

[0045] In an embodiment of the present application, the symptom description semantic encoding module 130 is used to perform semantic embedding encoding on each symptom description in the set of symptom descriptions to obtain a set of symptom description semantic embedding encoding vectors, and then perform contextual semantic encoding on the set of symptom description semantic embedding encoding vectors to obtain a set of implicit semantic association feature vectors between contextual symptoms. Specifically, in an embodiment of the present application, the symptom description semantic encoding module includes: a semantic embedding encoding unit, which is used to pass each symptom description in the set of symptom descriptions through a semantic embedding encoder based on the Bert model to obtain a set of symptom description semantic embedding encoding vectors; and a hidden semantic association unit, which is used to input the set of symptom description semantic embedding encoding vectors into a contextual semantic encoder based on Bi-LSTM to obtain a set of implicit semantic association feature vectors between contextual symptoms. It should be understood that the contextual semantic information about the symptoms exists between each symptom description in the set of symptom descriptions. The Bert model is a bidirectional Transformer model that can simultaneously consider the contextual information of all words in the text. Based on this, in the technical solution of the present application, each symptom description in the set of symptom descriptions is respectively passed through a semantic embedding encoder based on the Bert model to capture and extract the semantic information between each phrase or word in each symptom description, thereby obtaining a set of symptom description semantic embedding coding vectors. Furthermore, considering that the degree of semantic association of symptom descriptions implied between different semantic spans in the set of symptom description semantic embedding coding vectors is different. Therefore, in order to effectively capture and consider the semantic association dependency between different symptom description semantic embedding coding vectors, in the technical solution of the present application, the set of symptom description semantic embedding coding vectors is input into a contextual semantic encoder based on Bi-LSTM to obtain a set of implicit semantic association feature vectors between contextual symptoms.

[0046] In an embodiment of the present application, the context semantics enhancement module 140 is used to input the set of implicit semantic association feature vectors between the context symptoms into the context semantics enhancement module based on significance-globality to obtain a set of implicit semantic association feature vectors between the context symptoms. It should be understood that the set of implicit semantic association feature vectors between the context symptoms reflects the key implicit semantic association features between the context symptoms and the global information based on the implicit semantic association between the entire context symptoms. Based on this, in order to comprehensively utilize the key information, important features and semantic association information of the entire context symptoms in the context symptoms to enhance the features of the semantic association of the set of implicit semantic association feature vectors between the context symptoms, so as to strengthen the semantic connection between the symptoms, in the technical solution of the present application, the set of implicit semantic association feature vectors between the context symptoms is input into the context semantics enhancement module based on significance-globality to obtain a set of implicit semantic association feature vectors between the context symptoms. In detail, the context semantics enhancement module based on significance-globality extracts the key semantic association information and the semantic relationship of the entire context symptoms in the set of implicit semantic association feature vectors between the context symptoms to obtain prominent features and global features. Next, the salient features and the global features are convolved and nonlinearly processed respectively to improve the model's understanding and rich representation capabilities of the implicit semantic association features between symptoms. Then, the salient features and global features obtained after processing are cascaded and nonlinearly activated to comprehensively utilize the implicit semantic association features of different scales to improve the expression capabilities of the implicit semantic association features between symptoms to obtain a weight vector. Finally, based on the weight vector, the set of implicit semantic association feature vectors between contextual symptoms is weighted to strengthen the key implicit semantic association information, thereby obtaining a set of enhanced implicit semantic association feature vectors between contextual symptoms.

[0047] Specifically, in an embodiment of the present application, the contextual semantic enhancement module includes: a prominent feature calculation unit, which is used to extract the maximum value of each implicit semantic association feature vector between contextual symptoms in the set of implicit semantic association feature vectors between contextual symptoms to obtain a prominent contextual symptom implicit semantic association feature vector; a global feature calculation unit, which is used to extract the average value of each implicit semantic association feature vector between contextual symptoms in the set of implicit semantic association feature vectors between contextual symptoms to obtain a global contextual symptom implicit semantic association feature vector; a feature processing unit, which is used to perform convolution coding, nonlinear activation and linear transformation on the prominent contextual symptom implicit semantic association feature vector and the global contextual symptom implicit semantic association feature vector respectively to obtain a prominent feature contextual symptom implicit semantic association activation feature vector and a global feature vector. The following is an activation feature vector of implicit semantic associations between symptoms; a feature fusion unit, used to fuse the activation feature vector of implicit semantic associations between prominent feature context symptoms and the activation feature vector of implicit semantic associations between global feature context symptoms to obtain a global-prominent feature context symptom implicit semantic association feature vector, and then perform nonlinear activation on the global-prominent feature context symptom implicit semantic association feature vector to obtain a global-prominent feature context symptom implicit semantic association weight feature vector; a feature weighting unit, used to use the global-prominent feature context symptom implicit semantic association weight feature vector as a weight, perform positional dot multiplication on the set of context symptom implicit semantic association feature vectors and add the set of context symptom implicit semantic association feature vectors to obtain the set of enhanced context symptom implicit semantic association feature vectors.

[0048] More specifically, in an embodiment of the present application, the feature processing unit is used to: input the convolution coding vector of the implicit semantic association between prominent context symptoms obtained after performing one-dimensional convolution coding on the implicit semantic association feature vector between prominent context symptoms into the ReLU function to obtain the activation coding feature vector of the implicit semantic association between prominent context symptoms; perform matrix multiplication of the prominent point convolution feature vector obtained after performing point convolution coding on the activation coding feature vector of the implicit semantic association between prominent context symptoms with the first weight matrix to obtain the activation feature vector of the implicit semantic association between prominent feature context symptoms; input the convolution coding vector of the global contextual semantic association between implicit context symptoms obtained after performing one-dimensional convolution coding on the implicit semantic association feature vector between global context symptoms into the ReLU function to obtain the activation coding feature vector of the global contextual semantic association between implicit context symptoms; perform matrix multiplication of the global point convolution feature vector obtained after performing point convolution coding on the activation coding feature vector of the implicit semantic association between global context symptoms with the second weight matrix to obtain the activation feature vector of the implicit semantic association between global feature context symptoms.

[0049] More specifically, in an embodiment of the present application, the feature fusion unit is used to: perform feature fusion on the activation feature vector of implicit semantic association between the prominent feature context symptoms and the activation feature vector of implicit semantic association between the global feature context symptoms to obtain the global-prominent feature context symptom implicit semantic association feature vector; input the global-prominent feature context symptom implicit semantic association feature vector into the tanh function to obtain a first activated global-prominent feature context symptom implicit semantic association feature vector; input the global-prominent feature context symptom implicit semantic association feature vector into the sigmoid function to obtain a second activated global-prominent feature context symptom implicit semantic association feature vector; perform positional point multiplication on the first activated global-prominent feature context symptom implicit semantic association feature vector and the second activated global-prominent feature context symptom implicit semantic association feature vector to obtain the global-prominent feature context symptom implicit semantic association weight feature vector.

[0050] In the embodiment of the present application, specifically, the contextual semantic enhancement module is used to: use the contextual semantic enhancement module based on significance-globality to process the set of implicit semantic association feature vectors between the contextual symptoms with the following semantic enhancement formula to obtain the set of implicit semantic association feature vectors between the enhanced contextual symptoms; wherein the semantic enhancement formula is: ;in, is the set of implicit semantic association feature vectors between the contextual symptoms, and are the maximum value and the average value of each eigenvector in the set of eigenvectors, and are the implicit semantic association feature vectors between the prominent contextual symptoms and the implicit semantic association feature vectors between the global contextual symptoms, is a one-dimensional convolutional code, yes function, is point convolutional coding, and are the first weight matrix and the second weight matrix respectively, and are respectively the implicit semantic association activation feature vector between the prominent feature context symptoms and the implicit semantic association activation feature vector between the global feature context symptoms, is the function, yes function, It means point multiplication by position. is the implicit semantic association weight feature vector between the global-salient feature context symptoms, It is a set of implicit semantic association feature vectors between the enhanced context symptoms.

[0051] In the embodiment of the present application, the certificate description semantic coding module 150 is used to perform semantic embedding coding on each certificate description in the set of certificate descriptions to obtain a set of certificate description semantic embedding coding vectors. Specifically, in the embodiment of the present application, the certificate description semantic coding module is used to: pass each certificate description in the set of certificate descriptions through the semantic embedding encoder based on the Bert model to obtain a set of certificate description semantic embedding coding vectors. It should be understood that, considering that each certificate description in the set of certificate descriptions contains semantic information about the context of each description, therefore, in the technical solution of the present application, each certificate description in the set of certificate descriptions is passed through the semantic embedding encoder based on the Bert model to obtain a set of certificate description semantic embedding coding vectors, which can better capture the semantic information and semantic associations in each certificate description, thereby improving the semantic understanding and representation capabilities of the certificate descriptions.

[0052] In an embodiment of the present application, the symptom-syndrome description semantic fusion module 160 is used to integrate the set of implicit semantic association feature vectors between the enhanced contextual symptoms and the set of semantic embedding coding vectors of syndrome descriptions into a sequence of symptom-syndrome semantic fusion feature vectors, and then pass them through a context encoder based on a Transformer layer to obtain a semantic fusion coding feature vector of symptom information and syndrome information as a semantic fusion coding feature of symptom information and syndrome information. Accordingly, in order to better capture the semantic association and interaction between the set of implicit semantic association feature vectors between the enhanced contextual symptoms and the set of semantic embedding coding vectors of syndrome descriptions, in the technical solution of the present application, the set of implicit semantic association feature vectors between the enhanced contextual symptoms and the set of semantic embedding coding vectors of syndrome descriptions are integrated into a sequence of symptom-syndrome semantic fusion feature vectors, so that the comprehensive semantic understanding ability of the overall information of the diagnosed patient object can be improved, thereby better performing the association analysis between symptom descriptions and syndrome descriptions. Then, the sequence of symptom-syndrome semantic fusion feature vectors is passed through a context encoder based on the Transformer layer to capture the dependencies and context information between different semantic distances of symptom-syndrome semantic fusion features, so as to obtain a more representative symptom information and syndrome information semantic fusion encoding feature vector.

[0053] In an embodiment of the present application, the inference result generation module 170 is used to obtain an inference result based on the semantic fusion coding features of the symptom information and the syndrome element information, and the inference result is used to represent the type label of the syndrome name. Specifically, in an embodiment of the present application, the inference result generation module is used to: input the semantic fusion coding feature vector of the symptom information and the syndrome element information into the syndrome name dialectical reasoner based on the classifier to obtain the inference result. That is, the semantic fusion coding features of the symptom information and the syndrome element information obtained by semantic fusion of the set of implicit semantic association feature vectors between the enhanced contextual symptoms and the set of semantic embedding coding vectors of the syndrome element description are classified and processed, so as to intelligently obtain the inference result of the type label of the syndrome name. In this way, the system can intelligently infer the syndrome name type label that conforms to individual differences based on the patient's specific symptom description and syndrome element description, reduce the differences caused by human factors, and thus achieve personalized diagnosis.

[0054] It is worth mentioning that those skilled in the art should be aware that before applying a deep neural network model for inference, the deep neural network model must be trained so that the deep neural network can implement specific functional capabilities.

[0055] Specifically, in the technical solution of the present application, the big data-based TCM intelligent syndrome differentiation information processing system also includes a training module for training the Bert model-based semantic embedding encoder, the Bi-LSTM-based contextual semantic encoder, the saliency-globality-based contextual semantic reinforcement module, the Transformer layer-based context encoder, and the classifier-based syndrome dialectical reasoner.

[0056] Figure 3 FIG. 1 is a block diagram of a training module in a TCM intelligent syndrome differentiation information processing system based on big data according to an embodiment of the present application. Figure 3As shown, the training module 200 includes: a training data acquisition unit 210, which is used to acquire training data, wherein the training data includes a set of training symptom descriptions of the diagnosed patient objects, a set of training syndrome descriptions of the diagnosed patient objects, and a true reasoning result, wherein the true reasoning result is a true type label of the syndrome name; a training symptom description semantic encoding unit 220, which is used to pass each training symptom description in the set of training symptom descriptions through the semantic embedding encoder based on the Bert model to obtain a set of training symptom description semantic embedding encoding vectors; a training symptom description context semantic association unit 230, which is used to input the set of training symptom description semantic embedding encoding vectors into the semantic embedding encoder based on the Bi-LSTM model. The context semantic encoder is used to obtain a set of implicit semantic association feature vectors between training context symptoms; the training symptom description context semantic reinforcement unit 240 is used to input the set of implicit semantic association feature vectors between training context symptoms into the context semantic reinforcement module based on significance-globality to obtain a set of implicit semantic association feature vectors between training reinforcement context symptoms; the training syndrome description semantic encoding unit 250 is used to pass each training syndrome description in the set of training syndrome descriptions through the semantic embedding encoder based on Bert model to obtain a set of training syndrome description semantic embedding encoding vectors; the training information fusion semantic encoding unit 260 is used to integrate the set of implicit semantic association feature vectors between training reinforcement context symptoms and the training syndrome description semantic embedding encoder. The set of encoding vectors is integrated into a sequence of training symptom-syndrome semantic fusion feature vectors and then passed through the Transformer layer-based context encoder to obtain the training symptom information and training syndrome information semantic fusion encoding feature vectors; a training reasoning result generating unit 270 is used to input the training symptom information and training syndrome information semantic fusion encoding feature vectors into the classifier-based syndrome name dialectical reasoning device to obtain the training reasoning result; a classification loss unit 280 is used to calculate the cross entropy loss function value between the training reasoning result and the true reasoning result to obtain the classification loss function value; a model training unit 290 is a training module for training the Bert model-based semantic embedding encoder, the Bi-LSTM-based context semantic encoder, the saliency-globality-based context semantic reinforcement module, the Transformer layer-based context encoder and the classifier-based syndrome name dialectical reasoning device based on the classification loss function value and through gradient descent back propagation.

[0057] In particular, here, the set of implicit semantic association feature vectors between the training enhanced contextual symptoms expresses the multi-dimensional contextual association enhanced feature representation of the encoded text semantic features of each training symptom description in the set of training symptom descriptions, and the set of semantic embedding encoding vectors of the training certificate description expresses the source encoded text semantic features of each training certificate description in the set of training certificate descriptions. In this way, after integration and further passing through the context encoder based on the self-attention mechanism of the Transformer layer, the semantic fusion encoding feature vectors of the training symptom information and the training certificate information will also have feature aggregation class representation imbalance due to the difference in self-attention weight distribution, and make the probability density distribution convergence logic of in-class features and out-of-class features after the clustering operation inconsistent.

[0058] Based on this, in the preferred embodiment, when the semantic embedding encoder based on the Bert model, the contextual semantic encoder based on Bi-LSTM, the contextual semantic reinforcement module based on saliency-globality, the contextual encoder based on the Transformer layer and the certificate name dialectical reasoner based on the classifier are trained based on the classification loss function value, wherein, in each iteration of the training, the semantic fusion encoding feature vector of the training symptom information and the training certificate element information is iteratively optimized,

[0059] Wherein, the iterative optimization process specifically includes the following steps: performing clustering operations on the semantically fused coding feature vectors of the training symptom information and the training syndrome information, for example, performing clustering operations based on the distance between eigenvalues, and determining the number of eigenvalues ​​in the clustering set obtained by the clustering operation, that is, the number of in-class features, and subtracting the number of in-class features from the total number of eigenvalues ​​of the semantically fused coding feature vectors of the training symptom information and the training syndrome information to obtain the number of out-of-class features; dividing the total number of eigenvalues ​​by the number of in-class features and the number of out-of-class features respectively to obtain the class importance value and the out-of-class proportion value, and calculating the inverse of the class importance value to obtain the class constraint value; calculating the semantically fused coding feature vectors of the training symptom information and the training syndrome information Each eigenvalue of the quantity is a power function with the class importance value as the exponent, and is added to the exponential value with the natural function as the base and the class constraint value as the exponent, and then multiplied by the out-of-class proportion value to obtain a training symptom information and training syndrome information semantic fusion coding modulation vector; the training symptom information and training syndrome information semantic fusion coding feature vector is point multiplied with the class importance value to obtain a training symptom information and training syndrome information semantic fusion coding ontology vector; the weighted sum of the training symptom information and training syndrome information semantic fusion coding modulation vector and the training symptom information and training syndrome information semantic fusion coding ontology vector is calculated with a weight hyperparameter to obtain an optimized training symptom information and training syndrome information semantic fusion coding feature vector.

[0060] Specifically, in this preferred embodiment, the semantic fusion coding feature vector of the training symptom information and the training syndrome information is optimized to obtain the optimized semantic fusion coding feature vector of the training symptom information and the training syndrome information. The process formula is expressed as follows: ;in, is the semantic fusion encoding feature vector of the training symptom information and the training syndrome information, is the total number of eigenvalues ​​of the semantic fusion encoding feature vector of the training symptom information and the training syndrome information, is the number of intra-class features of the semantic fusion encoding feature vector of the training symptom information and the training syndrome information, It is the point multiplication by position. is the weight hyperparameter, is added by position, It means to perform exponential operation on the eigenvalues ​​at each position in the eigenvector. It is the optimized semantic fusion encoding feature vector of training symptom information and training syndrome information.

[0061] Therefore, while performing clustering operations on the semantic fusion coding feature vectors of the training symptom information and the training syndrome information, the clustering importance measure of the feature value of the semantic fusion coding feature vectors of the training symptom information and the training syndrome information is combined with the class constraint measure of the clustering operation to further perform responsive modulation with the out-of-class proportion factor, and based on the clustering scaling of the feature ontology representation of the feature vectors of the training symptom information and the training syndrome information, the clustering operation is used to establish the simplicity and effectiveness of the overall feature purification of the feature set of the training symptom information and the training syndrome information semantic fusion coding feature vectors, and to suppress the logical inconsistency of the probability density distribution convergence caused by the clustering operation, so as to improve the classification iteration effect of the syndrome name dialectical reasoner based on the classifier of the feature set input of the semantic fusion coding feature vectors of the training symptom information and the training syndrome information, that is, to improve the speed of classification training and the accuracy of classification results. In this way, the system can intelligently infer the syndrome name type label that conforms to individual differences based on the patient's specific symptom description and syndrome element description, reduce the differences caused by human factors, and thus achieve personalized diagnosis.

[0062] In another example of the present application, the following algorithm can also be used to infer the appropriate certificate name, wherein the specific algorithm is:

[0063] Among them, Zj: any certificate element in the certificate element set, Ai: weight relationship value, Zi: certificate element set.

[0064] First, based on the mapping of each collected clinical symptom, find the relevant syndrome elements and weight relationship values, that is, form Zi and Ai. Then, add up the weight relationship values ​​of each syndrome element to get Zj. If it is a major symptom, Ai is multiplied by 1.5; if it is a mild symptom, Ai is multiplied by 0.7. Next, arrange Zj from large to small, screen the syndrome element combinations with Zj greater than or equal to 100, and use the relationship between syndrome elements and syndrome names to find a suitable syndrome name. Finally, if there is no suitable syndrome name, use the following rules: when the number of syndrome elements with Zj greater than or equal to 100 is greater than or equal to 5, increase the base value by 100; when it is less than or equal to 3, reduce the base value by 100; when it is equal to 4, increase it first and then decrease it. Increase or decrease by 10 each time until a suitable syndrome name is found.

[0065] In summary, the intelligent TCM syndrome differentiation information processing system 100 based on big data according to the embodiment of the present application is explained, which obtains the set of symptom descriptions and the set of syndrome element descriptions of the diagnosed patient objects, and uses the text description analysis technology based on big data and artificial intelligence to perform semantic coding and feature association of the set of symptom descriptions and the set of syndrome element descriptions of the diagnosed patient objects, so as to intelligently obtain the inference result of the syndrome type label based on the implicit semantic association between the contexts of the various symptom descriptions and the fusion features between the semantic features of the various syndrome element descriptions. In this way, the system can intelligently infer the syndrome type label that conforms to individual differences according to the patient's specific symptom descriptions and syndrome element descriptions, reduce the differences caused by human factors, and thus achieve personalized diagnosis.

[0066] As described above, the TCM intelligent syndrome differentiation information processing system 100 based on big data according to the embodiment of the present application can be implemented in various terminal devices, such as a server for TCM intelligent syndrome differentiation information processing based on big data. In one example, the TCM intelligent syndrome differentiation information processing system 100 based on big data according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the TCM intelligent syndrome differentiation information processing system 100 based on big data can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the TCM intelligent syndrome differentiation information processing system 100 based on big data can also be one of the many hardware modules of the terminal device.

[0067] Alternatively, in another example, the big data-based intelligent Chinese medicine syndrome differentiation information processing system 100 and the terminal device may also be separate devices, and the big data-based intelligent Chinese medicine syndrome differentiation information processing system 100 may be connected to the terminal device via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0068] The above is merely an example of the principles of the present disclosure, and various modifications may be made by those skilled in the art without departing from the scope of the present disclosure. The above embodiments are presented for the purpose of illustration rather than limitation. The present disclosure may also take many forms other than those explicitly described herein.

Claims

1. A TCM intelligent syndrome differentiation information processing system based on big data, characterized in that: include: A patient object symptom description acquisition module is used to acquire a set of symptom descriptions of diagnosed patient objects; A patient object syndrome factor description acquisition module, used to acquire a set of syndrome factor descriptions of the diagnosed patient object; A symptom description semantic coding module, for performing semantic embedding coding on each symptom description in the set of symptom descriptions to obtain a set of symptom description semantic embedding coding vectors, and then performing contextual semantic coding on the set of symptom description semantic embedding coding vectors to obtain a set of implicit semantic association feature vectors between contextual symptoms; A contextual semantics enhancement module, used for inputting the set of implicit semantic association feature vectors between the contextual symptoms into the contextual semantics enhancement module based on significance-globality to obtain a set of enhanced implicit semantic association feature vectors between the contextual symptoms; A certificate element description semantic coding module, used for performing semantic embedding coding on each certificate element description in the set of certificate element descriptions to obtain a set of certificate element description semantic embedding coding vectors; A symptom-syndrome description semantic fusion module is used to integrate the set of implicit semantic association feature vectors between the enhanced contextual symptoms and the set of semantic embedding coding vectors of the syndrome description into a sequence of symptom-syndrome semantic fusion feature vectors, and then pass them through a context encoder based on the Transformer layer to obtain a semantic fusion coding feature vector of symptom information and syndrome information as a semantic fusion coding feature of symptom information and syndrome information; The inference result generation module is used to obtain the inference result based on the semantic fusion coding features of the symptom information and the syndrome element information, and the inference result is used to represent the type label of the syndrome name.

2. According to the big data-based TCM intelligent syndrome differentiation information processing system of claim 1, it is characterized in that: The symptom description semantic encoding module includes: A semantic embedding coding unit, used for respectively passing each symptom description in the set of symptom descriptions through a semantic embedding encoder based on a Bert model to obtain a set of semantic embedding coding vectors of the symptom descriptions; The latent semantic association unit is used to input the set of symptom description semantic embedding coding vectors into the Bi-LSTM-based contextual semantic encoder to obtain a set of latent semantic association feature vectors between the contextual symptoms.

3. The TCM intelligent syndrome differentiation information processing system based on big data according to claim 2 is characterized in that: The context semantics enhancement module includes: A salient feature calculation unit, used for extracting the maximum value of each contextual symptom implicit semantic association feature vector in the set of contextual symptom implicit semantic association feature vectors to obtain a salient contextual symptom implicit semantic association feature vector; A global feature calculation unit, used for extracting the average value of each contextual symptom implicit semantic association feature vector in the set of contextual symptom implicit semantic association feature vectors to obtain a global contextual symptom implicit semantic association feature vector; A feature processing unit, used for performing convolution coding, nonlinear activation and linear transformation on the salient contextual symptom implicit semantic association feature vector and the global contextual symptom implicit semantic association feature vector respectively to obtain a salient feature contextual symptom implicit semantic association activation feature vector and a global feature contextual symptom implicit semantic association activation feature vector; A feature fusion unit, used for fusing the salient feature context-symptom implicit semantic association activation feature vector and the global feature context-symptom implicit semantic association activation feature vector to obtain a global-salient feature context-symptom implicit semantic association feature vector, and then performing nonlinear activation on the global-salient feature context-symptom implicit semantic association feature vector to obtain a global-salient feature context-symptom implicit semantic association weight feature vector; A feature weighting unit is used to use the global-prominent feature context-symptom implicit semantic association weight feature vector as a weight, perform positional dot multiplication on the set of context-symptom implicit semantic association feature vectors, and add the set of context-symptom implicit semantic association feature vectors to obtain the set of enhanced context-symptom implicit semantic association feature vectors.

4. The TCM intelligent syndrome differentiation information processing system based on big data according to claim 3 is characterized in that: The feature processing unit is used to: Inputting the convolutional coding vector of the latent semantic association between prominent contextual symptoms obtained by performing one-dimensional convolution coding on the latent semantic association feature vector between prominent contextual symptoms into the ReLU function to obtain the activation coding feature vector of the latent semantic association between prominent contextual symptoms; Performing matrix multiplication of a salient point convolution feature vector obtained by performing point convolution encoding on the salient context-symptom implicit semantic association activation encoding feature vector and a first weight matrix to obtain the salient feature context-symptom implicit semantic association activation feature vector; Inputting the convolutional coding vector of the implicit semantic association between global contextual symptoms obtained by performing one-dimensional convolution coding on the implicit semantic association feature vector between global contextual symptoms into the ReLU function to obtain the activation coding feature vector of the implicit semantic association between global contextual symptoms; The global point convolution feature vector obtained by performing point convolution encoding on the global contextual inter-symptom implicit semantic association activation coding feature vector is matrix multiplied with the second weight matrix to obtain the global feature contextual inter-symptom implicit semantic association activation feature vector.

5. The TCM intelligent syndrome differentiation information processing system based on big data according to claim 4 is characterized in that: The feature fusion unit is used to: Performing feature fusion on the salient feature context-symptom implicit semantic association activation feature vector and the global feature context-symptom implicit semantic association activation feature vector to obtain the global-salient feature context-symptom implicit semantic association feature vector; Inputting the global-salient feature context-symptom implicit semantic association feature vector into a tanh function to obtain a first activated global-salient feature context-symptom implicit semantic association feature vector; Inputting the global-prominent feature context-symptom implicit semantic association feature vector into a Sigmoid function to obtain a second activated global-prominent feature context-symptom implicit semantic association feature vector; The first activated global-salient feature context-symptom implicit semantic association feature vector and the second activated global-salient feature context-symptom implicit semantic association feature vector are point-wise multiplied to obtain the global-salient feature context-symptom implicit semantic association weight feature vector.

6. The TCM intelligent syndrome differentiation information processing system based on big data according to claim 5 is characterized in that: The certificate element description semantic encoding module is used to: pass each certificate element description in the set of certificate element descriptions through the semantic embedding encoder based on the Bert model to obtain a set of the certificate element description semantic embedding encoding vectors.

7. The TCM intelligent syndrome differentiation information processing system based on big data according to claim 6 is characterized in that: The inference result generating module is used to input the semantic fusion encoding feature vector of the symptom information and syndrome factor information into the syndrome name dialectical inference device based on the classifier to obtain the inference result.

8. The TCM intelligent syndrome differentiation information processing system based on big data according to claim 7 is characterized in that: It also includes a training module for training the Bert model-based semantic embedding encoder, the Bi-LSTM-based contextual semantic encoder, the saliency-globality-based contextual semantic reinforcement module, the Transformer layer-based context encoder, and the classifier-based certificate name dialectical reasoner.

9. The TCM intelligent syndrome differentiation information processing system based on big data according to claim 8 is characterized in that: The training module comprises: A training data acquisition unit, used to acquire training data, wherein the training data includes a set of training symptom descriptions of the diagnosed patient objects, a set of training syndrome descriptions of the diagnosed patient objects, and a true reasoning result, wherein the true reasoning result is a true type label of the syndrome name; A training symptom description semantic encoding unit, used for passing each training symptom description in the set of training symptom descriptions through the semantic embedding encoder based on the Bert model to obtain a set of training symptom description semantic embedding encoding vectors; A training symptom description context semantic association unit, used for inputting the set of training symptom description semantic embedding coding vectors into the Bi-LSTM-based context semantic encoder to obtain a set of implicit semantic association feature vectors between training context symptoms; A training symptom description context semantics enhancement unit, used for inputting the set of implicit semantic association feature vectors between the training context symptoms into the saliency-global-based context semantics enhancement module to obtain a set of implicit semantic association feature vectors between the training enhanced context symptoms; A training evidence description semantic encoding unit, used for passing each training evidence description in the set of training evidence descriptions through the semantic embedding encoder based on the Bert model to obtain a set of training evidence description semantic embedding encoding vectors; A training information fusion semantic coding unit, used to integrate the set of implicit semantic association feature vectors between the training reinforcement contextual symptoms and the set of semantic embedding coding vectors describing the training syndrome into a sequence of training symptom-syndrome semantic fusion feature vectors, and then pass them through the context encoder based on the Transformer layer to obtain a semantic fusion coding feature vector of training symptom information and training syndrome information; A training reasoning result generating unit, used for inputting the semantic fusion encoding feature vector of the training symptom information and the training syndrome element information into the classifier-based syndrome name dialectical reasoning device to obtain a training reasoning result; A classification loss unit, used to calculate a cross entropy loss function value between the training reasoning result and the true reasoning result to obtain a classification loss function value; A model training unit is a training module for training the semantic embedding encoder based on the Bert model, the contextual semantic encoder based on Bi-LSTM, the contextual semantic reinforcement module based on saliency-globality, the contextual encoder based on the Transformer layer, and the classifier-based certificate name dialectical reasoner based on the classification loss function value and through back propagation of gradient descent.

Images