Rheumatism and immunology drug intelligent recommendation system and method

Abstract

The present application provides a rheumatism and immunology drug intelligent recommendation system and method, which relates to the field of drug intelligent recommendation. It first obtains the current medical record information of the rheumatic patient and the drug description information of the drug to be recommended, and then uses a text information processing and analysis algorithm based on deep learning to perform word segmentation and semantic encoding on each description information in the current medical record information and the drug description information, so as to automatically determine whether to recommend the drug to be recommended to the rheumatic patient based on the semantic fusion features between the various description information and the interactive matching information between the current medical record information. In this way, personalized rheumatic and immunology drug recommendations can be achieved according to the disease characteristics and individual differences of different patients.

Description

Technical Field

[0001] The present application relates to the field of intelligent drug recommendation, and more specifically, to an intelligent recommendation system and method for rheumatism and immunology drugs. Background Art

[0002] Rheumatic and immune diseases belong to the category of internal medicine and are diseases that involve the whole body. Their clinical manifestations are complex and varied, and are often confused with the symptoms of other diseases such as skin diseases, eye inflammation, and oral ulcers. In order to ensure the smooth progress of the diagnosis and treatment process and avoid misdiagnosis, doctors need to comprehensively consider the patient's medical history, symptoms, clinical manifestations, and auxiliary examination results. Based on this comprehensive information, doctors can accurately diagnose and prescribe appropriate treatment drugs.

[0003] Based on this, the invention with publication number CN112116978A proposes a method, system and device for recommending rheumatism and immunology drugs. The existing patent scheme uses fuzzy rules to establish corresponding rules between diseases and their severity and symptoms based on existing medical records and professional guidance of rheumatologists. Then, in the stage of recommending rheumatism and immunology drugs, the matching similarity is obtained by comparing the keyword information in the patient's medical record with the existing symptoms in the fuzzy rule library. If the similarity reaches the activation threshold, the corresponding rules are activated to obtain the matching disease type and its severity, and the initial drug recommendation result is output by matching.

[0004] The above scheme uses keyword matching, ignoring the contextual association of the full text of the medical record. In addition, it matches and searches for rheumatic and immunological drugs based on the type of disease and its severity, and does not take into account the implicit association patterns and interactions between the individual differences in the disease semantics and the drug effects of rheumatic patients. It only relies on the search and matching method of the medical knowledge base, which reduces the accuracy and effectiveness of rheumatic and immunological drug recommendations.

[0005] Therefore, an optimized intelligent recommendation system and method for rheumatism and immunology drugs is desired. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present application provides an intelligent recommendation system and method for rheumatism and immunology drugs.

[0007] According to one aspect of the present application, a rheumatism and immunology drug intelligent recommendation system is provided, which includes:

[0008] A module for obtaining the current medical records of rheumatic patients, used to obtain the current medical records of rheumatic patients;

[0009] A module for acquiring description information of the drug to be recommended, used to acquire description information of the drug to be recommended, wherein the description information includes description of drug action, description of drug side effects and description of drug interaction;

[0010] A current medical record word granularity semantic encoding module is used to perform word segmentation processing on the current medical record information of the rheumatic patient and then pass it through a current medical record semantic encoder to obtain a current medical record word granularity global semantic association feature vector;

[0011] A drug description information semantic encoding module, used for performing word segmentation processing on the drug description information and then passing it through a drug description semantic encoder to obtain a semantic encoding feature vector of drug action description word granularity, a semantic encoding feature vector of drug side effect description word granularity, and a semantic encoding feature vector of drug interaction word granularity;

[0012] A drug description semantic feature multimodal fusion module, which is used to use the drug action description word granularity semantic encoding feature vector as a query vector, the drug side effect description word granularity semantic encoding feature vector as a key vector and the drug interaction word granularity semantic encoding feature vector as a value vector, and input the drug action description word granularity semantic encoding feature vector, the drug side effect description word granularity semantic encoding feature vector and the drug interaction word granularity semantic encoding feature vector into a drug description multimodal semantic joint encoder based on the Transformer mechanism to obtain a drug description multimodal semantic fusion feature vector;

[0013] A current medical record-drug description semantic interaction matching module, used to obtain a current medical record-drug description semantic interaction matching feature vector as a current medical record-drug description semantic interaction matching feature by passing the current medical record word granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector through a feature interaction matching module based on a responsive distinguishable network;

[0014] The recommendation result generation module is used to obtain a recommendation result based on the semantic interaction matching feature of the current medical record-drug description, and the recommendation result is used to indicate whether the drug to be recommended is recommended to the rheumatic patient.

[0015] In the above-mentioned intelligent recommendation system for rheumatism and immunology drugs, the current medical record word granularity semantic encoding module is used to: perform word segmentation processing on the current medical record information of the rheumatic patient, and then pass it through the current medical record semantic encoder containing a word embedding layer to obtain the current medical record word granularity global semantic association feature vector.

[0016] In the above-mentioned intelligent recommendation system for rheumatism and immunology drugs, the drug description information semantic encoding module is used to: perform word segmentation on the drug description information and then pass it through a drug description semantic encoder including a word embedding layer to obtain the drug action description word granularity semantic encoding feature vector, the drug side effect description word granularity semantic encoding feature vector and the drug interaction word granularity semantic encoding feature vector.

[0017] In the above-mentioned intelligent recommendation system for rheumatism and immunology drugs, the drug description semantic feature multimodal fusion module includes: a drug action description word granular semantic coding weight calculation unit, which is used to multiply the weight matrix of the query vector with the drug action description word granular semantic coding feature vector to obtain the drug action description word granular semantic coding weight vector; a drug side effect description word granular semantic coding weight calculation unit, which is used to multiply the weight matrix of the key vector with the drug side effect description word granular semantic coding feature vector to obtain the drug side effect description word granular semantic coding weight vector; a drug interaction word granular semantic coding weight calculation unit, which is used to multiply the weight matrix of the value vector with the drug interaction word granular semantic coding feature vector to obtain the drug interaction word granular semantic coding weight vector; a drug action-drug side effect semantic association weight calculation unit, which is used to calculate the drug action description word granular semantic coding weight vector. After obtaining the drug action-drug side effect description word granularity semantic association value by multiplying the transpose vector of the weight vector with the drug side effect description word granularity semantic encoding weight vector, the drug action-drug side effect description word granularity semantic association value is divided by the square root of the length of the drug side effect description word granularity semantic encoding weight vector to obtain the drug action-drug side effect description word granularity semantic association weight value; a drug action-drug side effect semantic association weight activation unit is used to input the drug action-drug side effect description word granularity semantic association weight value into a softmax function to obtain the drug action-drug side effect description word granularity semantic association activation weight value; a drug description semantic weighting unit is used to use the drug action-drug side effect description word granularity semantic association activation weight value as a weight value to weight the transpose vector of the drug interaction word granularity semantic encoding weight vector to obtain the drug description multimodal semantic fusion feature vector.

[0018] In the above-mentioned intelligent recommendation system for rheumatism and immunology drugs, the drug description semantic weighting unit is used to: calculate the product between the activation weight value of the drug action-drug side effect description word granularity semantic association and the feature value of each position in the transposed vector of the drug interaction word granularity semantic encoding weight vector to obtain the multimodal semantic fusion feature vector of the drug description.

[0019] In the above-mentioned intelligent recommendation system for rheumatism and immunology drugs, the current medical record-drug description semantic interactive matching module includes: a current medical record-drug description point-by-point semantic association unit, which is used to calculate the position-by-position response between the current medical record word granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector to obtain the current medical record-drug description semantic position-by-position response feature vector; a current medical record-drug description point-by-point semantic association feature normalization unit, which is used to use the Softmax function to normalize the current medical record-drug description semantic position-by-position response feature vector to obtain a normalized current medical record-drug description semantic position-by-position response feature vector; a current medical record-drug description semantic weight value calculation unit, which is used to normalize the normalized current medical record-drug description semantic position-by-position response feature vector. The position response feature vector is input into a learnable gating function to obtain a current medical record-drug description semantic response screening weight mask vector; a current medical record-drug description semantic response weight mask calculation unit is used to calculate the position point multiplication between the current medical record-drug description semantic response screening weight mask vector and the normalized current medical record-drug description semantic position response feature vector to obtain a current medical record-drug description semantic position response distinguishable weight mask vector; a current medical record-drug description semantic interaction matching feature generation unit is used to calculate the position point multiplication between the current medical record-drug description semantic position response distinguishable weight mask vector and the current medical record-drug description semantic position response feature vector to obtain the current medical record-drug description semantic interaction matching feature vector.

[0020] In the above-mentioned intelligent recommendation system for rheumatism and immunology drugs, the current medical record-drug description semantic weight value calculation unit is used to: input the normalized current medical record-drug description semantic position-response feature vector into the sigmoid function to calculate the weight mask value to obtain the current medical record-drug description semantic response screening weight mask vector.

[0021] In the above-mentioned intelligent recommendation system for rheumatism and immunology drugs, the recommendation result generation module is used to: pass the current medical record-drug description semantic interaction matching feature vector through a classifier-based rheumatism and immunology drug recommender to obtain the recommendation result, and the recommendation result is used to indicate whether the recommended drug is recommended to the rheumatic patient.

[0022] According to another aspect of the present application, a method for intelligently recommending rheumatism and immunology drugs is provided, comprising:

[0023] Obtain current medical record information of rheumatology patients;

[0024] Obtaining drug description information of the drug to be recommended, wherein the drug description information includes a description of drug action, a description of drug side effects, and a description of drug interactions;

[0025] The current medical record information of the rheumatic patient is segmented and passed through a current medical record semantic encoder to obtain a current medical record word granularity global semantic association feature vector;

[0026] The drug description information is segmented and passed through a drug description semantic encoder to obtain a drug action description word granularity semantic encoding feature vector, a drug side effect description word granularity semantic encoding feature vector, and a drug interaction word granularity semantic encoding feature vector;

[0027] Taking the drug action description word granularity semantic encoding feature vector as the query vector, the drug side effect description word granularity semantic encoding feature vector as the key vector and the drug interaction word granularity semantic encoding feature vector as the value vector, the drug action description word granularity semantic encoding feature vector, the drug side effect description word granularity semantic encoding feature vector and the drug interaction word granularity semantic encoding feature vector are input into a drug description multimodal semantic joint encoder based on the Transformer mechanism to obtain a drug description multimodal semantic fusion feature vector;

[0028] The current medical record word granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector are passed through a feature interaction matching module based on a responsive distinguishable network to obtain a current medical record-drug description semantic interaction matching feature vector as a current medical record-drug description semantic interaction matching feature;

[0029] Based on the semantic interaction matching feature of the current medical record-drug description, a recommendation result is obtained, and the recommendation result is used to indicate whether the drug to be recommended is recommended to the rheumatic patient.

[0030] In the above-mentioned intelligent recommendation method for rheumatic immunology drugs, the current medical record information of the rheumatic patient is processed by word segmentation and then passed through a current medical record semantic encoder to obtain a current medical record word-granularity global semantic association feature vector, including: the current medical record information of the rheumatic patient is processed by word segmentation and then passed through a current medical record semantic encoder including a word embedding layer to obtain a current medical record word-granularity global semantic association feature vector.

[0031] This application has significant technical effects due to the adoption of the above technical solutions:

[0032] The intelligent recommendation system and method for rheumatism and immunology drugs provided by the present application first obtains the current medical record information of the rheumatic patient and the drug description information of the drug to be recommended, and then uses a text information processing and analysis algorithm based on deep learning to perform word segmentation and semantic encoding on each description information in the current medical record information and the drug description information, so as to automatically determine whether to recommend the drug to be recommended to the rheumatic patient based on the semantic fusion features between the various description information and the interactive matching information between the current medical record information. In this way, personalized rheumatic and immunology drug recommendations can be achieved according to the disease characteristics and individual differences of different patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art description are briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0034] Figure 1 This is a system block diagram of the intelligent recommendation system for rheumatism and immunology drugs according to an embodiment of the present application.

[0035] Figure 2 Schematic diagram of the architecture of the intelligent recommendation system for rheumatism and immunology drugs according to an embodiment of the present application.

[0036] Figure 3 This is a block diagram of a multimodal fusion module of semantic features of drug descriptions in an intelligent recommendation system for rheumatism and immunology drugs according to an embodiment of the present application.

[0037] Figure 4 This is a block diagram of the current medical record-drug description semantic interaction matching module in the rheumatism and immunology drug intelligent recommendation system according to an embodiment of the present application.

[0038] Figure 5 The present invention is a flowchart of a method for intelligently recommending rheumatism and immunology drugs according to an embodiment of the present application. DETAILED DESCRIPTION

[0039] 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.

[0040] Rheumatology is a complex systemic disease that is usually classified as a field of internal medicine. The diagnosis of this type of disease is challenging because its symptoms may be confused with a variety of other diseases such as skin diseases, eye inflammation, or oral ulcers. In order to improve the accuracy of diagnosis and avoid misdiagnosis, doctors need to comprehensively evaluate the patient's medical history, symptoms, clinical manifestations, and auxiliary examination results. Only through this comprehensive analysis can doctors make an accurate diagnosis and recommend appropriate treatment options.

[0041] Based on this, the invention with publication number CN112116978A introduces a method, system and device for recommending rheumatology and immunology drugs. The core technical solution of this existing patent is to use fuzzy rules, combined with the professional guidance of rheumatologists, to construct corresponding rules between diseases, symptom severity and symptoms. In the drug recommendation stage, the degree of match is calculated by comparing the keywords in the patient's medical record with the symptoms in the rule library. Once the degree of match reaches the preset threshold, the system will trigger the corresponding rules to identify the patient's disease type and severity, and provide preliminary drug recommendations accordingly.

[0042] However, the technical solution of this existing patent has some limitations. It mainly relies on keyword matching, without making full use of the contextual information of the medical record text. In addition, the recommendation process is mainly based on the type and severity of the disease, without fully considering the individual differences of rheumatic patients and the complex interaction between the disease and the drug effect. This method limits the accuracy of rheumatic immunology drug recommendations to a certain extent.

[0043] Therefore, in view of the above technical problems, the technical concept of the present application is to obtain the current medical record information of rheumatic patients, and obtain the drug description information (drug action description, drug side effect description and drug interaction description) of the drug to be recommended, and use a text information processing and analysis algorithm based on deep learning to perform word segmentation and semantic encoding on the current medical record information and each description information in the drug description information, so as to automatically judge whether to recommend the drug to be recommended to the rheumatic patient according to the semantic fusion features between the various description information and the interactive matching information between the current medical record information. In this way, by performing word segmentation and semantic encoding on the medical record text and drug information of rheumatic patients through a deep learning algorithm, the contextual relationship and semantic information in the medical record and drug description can be more deeply understood. At the same time, it can identify the disease characteristics and individual differences of different patients, and mine the implicit semantic association patterns between the medical record information and drug action semantics of different patients, so as to improve the matching degree between the recommended drugs and the actual condition of the patients, and provide patients with more personalized rheumatic immune drug recommendations.

[0044] Figure 1 This is a system block diagram of the intelligent recommendation system for rheumatism and immunology drugs according to an embodiment of the present application. Figure 2FIG. 1 is a schematic diagram of the architecture of the intelligent recommendation system for rheumatism and immunology drugs according to an embodiment of the present application. Figure 1 and Figure 2 As shown, in the rheumatism and immunology drug intelligent recommendation system 100, it includes: a rheumatic patient's current medical record information acquisition module 110, which is used to obtain the current medical record information of the rheumatic patient; a to-be-recommended drug description information acquisition module 120, which is used to obtain the drug description information of the to-be-recommended drug, and the drug description information includes a drug action description, a drug side effect description and a drug interaction description; a current medical record word granularity semantic encoding module 130, which is used to perform word segmentation processing on the current medical record information of the rheumatic patient and then pass it through the current medical record semantic encoder to obtain the current medical record word granularity global semantic association feature vector; a drug description information semantic encoding module 140, which is used to perform word segmentation processing on the drug description information and then pass it through the drug description semantic encoder to obtain a drug action description word granularity semantic encoding feature vector, a drug side effect description word granularity semantic encoding feature vector and a drug interaction word granularity semantic encoding feature vector; a drug description semantic feature multimodal fusion module 150, which is used to use the drug action description word granularity semantic encoding feature vector as a query vector, the drug The word-granular semantic encoding feature vector of the drug side effect description is used as a key vector and the word-granular semantic encoding feature vector of the drug interaction is used as a value vector, and the word-granular semantic encoding feature vector of the drug action description, the word-granular semantic encoding feature vector of the drug side effect description and the word-granular semantic encoding feature vector of the drug interaction are input into a drug description multimodal semantic joint encoder based on the Transformer mechanism to obtain a drug description multimodal semantic fusion feature vector; a current medical record-drug description semantic interaction matching module 160 is used to pass the current medical record word-granular global semantic association feature vector and the drug description multimodal semantic fusion feature vector through a feature interaction matching module based on a responsive distinguishable network to obtain a current medical record-drug description semantic interaction matching feature vector as a current medical record-drug description semantic interaction matching feature; a recommendation result generation module 170 is used to obtain a recommendation result based on the current medical record-drug description semantic interaction matching feature, and the recommendation result is used to indicate whether the recommended drug is recommended to rheumatic patients.

[0045] In the embodiment of the present application, the current medical record information acquisition module 110 of the rheumatic patient is used to obtain the current medical record information of the rheumatic patient. The to-be-recommended drug description information acquisition module 120 is used to obtain the drug description information of the to-be-recommended drug, and the drug description information includes a description of the drug action, a description of the drug side effects, and a description of drug interactions. It should be understood that the current medical record information of the rheumatic patient mainly includes the patient's reason for the visit, a specific symptom description, the onset time, duration, severity of the symptoms, etc., which helps to fully understand the individual health conditions of the rheumatic patient, thereby providing a basis for drug recommendation. The drug action description in the drug description information of the to-be-recommended drug mainly describes the main therapeutic effects and indications of the drug, the drug side effect description mainly describes the common side effects and serious side effects of the drug, and the drug interaction description mainly describes the possible reduced efficacy and side effects caused by the interaction between the drug and other drugs. These detailed drug description information provides the necessary background for recommending drugs to patients and helps understand the applicability and safety of the drug. In general, the current medical history information of rheumatic patients and the drug description information of the drugs to be recommended are key elements in the drug recommendation process. Through the comprehensive analysis of these two aspects of data, individualized and accurate drug recommendations can be achieved. In particular, in one embodiment of the present application, the current medical history information data of rheumatic patients can be obtained through the hospital information system, and the drug description information of the drugs to be recommended can be obtained through the hospital's drug database.

[0046] In an embodiment of the present application, the current medical record word granularity semantic encoding module 130 is used to perform word segmentation processing on the current medical record information of the rheumatic patient and then pass it through a current medical record semantic encoder to obtain a current medical record word granularity global semantic association feature vector. Specifically, in an embodiment of the present application, the current medical record word granularity semantic encoding module 130 is used to: perform word segmentation processing on the current medical record information of the rheumatic patient and then pass it through a current medical record semantic encoder containing a word embedding layer to obtain a current medical record word granularity global semantic association feature vector. It should be understandable that there is a semantic association between the multiple words about the current medical record information contained in the current medical record information of the rheumatic patient based on the context of the local medical record information, and word embedding is a technology that maps vocabulary to a high-dimensional vector space, which makes semantically similar words closer in the vector space, which helps to capture the semantic relationship between words. Therefore, in the technical solution of the present application, the current medical record information of the rheumatic patient is segmented to divide the entire text information into multiple meaningful words, and then the current medical record semantic encoder containing a word embedding layer is used to capture and mine the semantic dependency relationship of the medical record information between the contexts contained in each word, so as to obtain the global semantic association feature vector of the current medical record word granularity.

[0047] In an embodiment of the present application, the drug description information semantic encoding module 140 is used to perform word segmentation processing on the drug description information and then pass it through a drug description semantic encoder to obtain a drug action description word granularity semantic encoding feature vector, a drug side effect description word granularity semantic encoding feature vector, and a drug interaction word granularity semantic encoding feature vector. Specifically, in an embodiment of the present application, the drug description information semantic encoding module 140 is used to: perform word segmentation processing on the drug description information and then pass it through a drug description semantic encoder including a word embedding layer to obtain a drug action description word granularity semantic encoding feature vector, a drug side effect description word granularity semantic encoding feature vector, and a drug interaction word granularity semantic encoding feature vector. It should be understood that each description information in the drug description information also contains semantic information about each drug information word granularity description, such as drug name, dosage, etc., and there is a semantic association between these word granularity information. Based on this, in the technical solution of the present application, the drug description information is segmented and then passed through a drug description semantic encoder including a word embedding layer to obtain a semantic coding feature vector of drug action description word granularity, a semantic coding feature vector of drug side effect description word granularity, and a semantic coding feature vector of drug interaction word granularity.

[0048] In an embodiment of the present application, the drug description semantic feature multimodal fusion module 150 is used to use the drug action description word granularity semantic coding feature vector as a query vector, the drug side effect description word granularity semantic coding feature vector as a key vector and the drug interaction word granularity semantic coding feature vector as a value vector, and input the drug action description word granularity semantic coding feature vector, the drug side effect description word granularity semantic coding feature vector and the drug interaction word granularity semantic coding feature vector into a drug description multimodal semantic joint encoder based on the Transformer mechanism to obtain a drug description multimodal semantic fusion feature vector. Accordingly, it is considered that there is a correlation between different drug description features based on the entire drug information between the drug action description word granularity semantic coding feature vector, the drug side effect description word granularity semantic coding feature vector and the drug interaction word granularity semantic coding feature vector. Therefore, in order to effectively process and fuse drug information from different descriptions to obtain a more comprehensive overall semantic understanding of the drug, in the technical solution of the present application, the drug action description word granularity semantic encoding feature vector is used as a query vector, the drug side effect description word granularity semantic encoding feature vector is used as a key vector, and the drug interaction word granularity semantic encoding feature vector is used as a value vector. The drug action description word granularity semantic encoding feature vector, the drug side effect description word granularity semantic encoding feature vector, and the drug interaction word granularity semantic encoding feature vector are input into the drug description multimodal semantic joint encoder based on the Transformer mechanism to obtain the drug description multimodal semantic fusion feature vector. It should be understood by ordinary technicians in this field that Transformer is a deep learning model based on the self-attention mechanism. Its core lies in the self-attention mechanism, which allows the model to simultaneously pay attention to all other elements in the sequence when processing each element in the sequence, thereby capturing the dependencies between elements. This parallel processing capability greatly improves the computational efficiency and overcomes the gradient vanishing or explosion problems that traditional recurrent neural networks and convolutional neural networks may face when processing long sequences.

[0049] It is worth mentioning that the Transformer-based drug description multimodal semantic joint encoder calculates the semantic association relationship between the drug action description word granularity semantic encoding feature vector and the drug side effect description word granularity semantic encoding feature vector respectively, and converts it into a weight value to highlight the characteristics that are more relevant to the drug action and side effect. The weight value is used to weight the drug interaction word granularity semantic encoding feature vector to more effectively capture the interaction and correlation between different drug description features, and generate a drug description multimodal semantic fusion feature vector with high expressiveness and accuracy, providing strong support for drug recommendation.

[0050] Specifically, Figure 3 FIG. 4 is a block diagram of a multimodal fusion module of drug description semantic features in an intelligent recommendation system for rheumatism and immunology drugs according to an embodiment of the present application. Figure 3 As shown, the drug description semantic feature multimodal fusion module 150 includes: a drug action description word granularity semantic coding weight calculation unit 151, which is used to multiply the weight matrix of the query vector with the drug action description word granularity semantic coding feature vector to obtain the drug action description word granularity semantic coding weight vector; a drug side effect description word granularity semantic coding weight calculation unit 152, which is used to multiply the weight matrix of the key vector with the drug side effect description word granularity semantic coding feature vector to obtain the drug side effect description word granularity semantic coding weight vector; a drug interaction word granularity semantic coding weight calculation unit 153, which is used to multiply the weight matrix of the value vector with the drug interaction word granularity semantic coding feature vector to obtain the drug interaction word granularity semantic coding weight vector; a drug action-drug side effect semantic association weight calculation unit 154, which is used to calculate the drug action description word granularity semantic coding weight vector. After obtaining the drug action-drug side effect description word granularity semantic association value by multiplying the transpose vector of the semantic encoding weight vector with the drug side effect description word granularity semantic encoding weight vector, the drug action-drug side effect description word granularity semantic association value is divided by the square root of the length of the key vector to obtain the drug action-drug side effect description word granularity semantic association weight value; a drug action-drug side effect semantic association weight activation unit 155 is used to input the drug action-drug side effect description word granularity semantic association weight value into a softmax function to obtain the drug action-drug side effect description word granularity semantic association activation weight value; a drug description semantic weighting unit 156 is used to use the drug action-drug side effect description word granularity semantic association activation weight value as a weight value to weight the transpose vector of the drug interaction word granularity semantic encoding weight vector to obtain the drug description multimodal semantic fusion feature vector.

[0051] More specifically, in an embodiment of the present application, the drug description semantic weighting unit 156 is used to: calculate the product of the activation weight value of the drug action-drug side effect description word granularity semantic association and the feature value of each position in the transposed vector of the drug interaction word granularity semantic encoding weight vector to obtain the drug description multimodal semantic fusion feature vector.

[0052] In the embodiment of the present application, specifically, the drug description semantic feature multimodal fusion module 150 is used to: use the drug action description word granularity semantic coding feature vector as the query vector, the drug side effect description word granularity semantic coding feature vector as the key vector and the drug interaction word granularity semantic coding feature vector as the value vector, input the drug action description word granularity semantic coding feature vector, the drug side effect description word granularity semantic coding feature vector and the drug interaction word granularity semantic coding feature vector into the drug description multimodal semantic joint encoder based on the Transformer mechanism, and process them with the following joint coding formula to obtain the drug description multimodal semantic fusion feature vector; wherein, the joint coding formula is:

[0053] ;

[0054] in, , and They are the granular semantic coding feature vector of drug action description words, the granular semantic coding feature vector of drug side effect description words and the granular semantic coding feature vector of drug interaction words, , and a weight matrix representing the query vector, the key vector and the value vector, is the length of the key vector, yes function, It is the multimodal semantic fusion feature vector of the drug description.

[0055] In an embodiment of the present application, the current medical record-drug description semantic interaction matching module 160 is used to pass the current medical record word-granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector through a feature interaction matching module based on a responsive distinguishable network to obtain the current medical record-drug description semantic interaction matching feature vector as the current medical record-drug description semantic interaction matching feature. Accordingly, in order to more finely and significantly enhance the word-granularity semantic feature interaction between the current medical record word-granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector, so as to better reflect the semantic interactivity between the current medical record and drug description features. In the technical solution of the present application, the current medical record word-granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector are passed through a feature interaction matching module based on a responsive distinguishable network to obtain the current medical record-drug description semantic interaction matching feature vector.

[0056] In particular, the feature interaction matching module based on the responsive distinguishable network achieves fine position-level processing and intelligent weight optimization of the feature vector by accurately adjusting the importance of each word-granular semantic feature element, thereby significantly improving the ability to represent the semantic interaction between the two features and enabling the model to understand the data more deeply. In detail, by performing granular feature interaction between the word-granular global semantic association feature vector of the current medical record and the multimodal semantic fusion feature vector of the drug description, the local semantic association relationship between the features can be revealed, and the current medical record-drug description semantic position-by-position response feature vector is obtained. Subsequently, the element-by-element response is normalized using the Softmax function to generate a probability distribution to ensure the standardized processing of the response value and prepare for the weight selection of the gating mechanism. Next, the normalized current medical record-drug description semantic position-by-position response feature vector is input into the gating function to adaptively select features through the gating mechanism and output a screening weight mask vector, which dynamically adjusts the importance of the two at the semantic interaction feature position. Then, the element-wise product of the screening weight mask and the normalized response feature is calculated to further refine the weight assignment, ensure that the feature information at each position is adapted to the corresponding weight, and generate a distinguishable weight mask vector. Finally, the distinguishable weight mask is element-wise multiplied with the position-by-position response feature vector to more effectively identify and utilize the key semantic interaction matching information in the data, and obtain a comprehensive representation that can fully reflect the interaction of the semantic features of the current medical records and drug descriptions.

[0057] Specifically, Figure 4 FIG. 1 is a block diagram of a current medical record-drug description semantic interaction matching module in a rheumatism and immunology drug intelligent recommendation system according to an embodiment of the present application. Figure 4As shown, the current medical record-drug description semantic interaction matching module 160 includes: a current medical record-drug description point-by-point semantic association unit 161, which is used to calculate the position-by-position response between the current medical record word granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector to obtain the current medical record-drug description semantic position-by-position response feature vector; a current medical record-drug description point-by-point semantic association feature normalization unit 162, which is used to use the Softmax function to normalize the current medical record-drug description semantic position-by-position response feature vector to obtain a normalized current medical record-drug description semantic position-by-position response feature vector; a current medical record-drug description semantic weight value calculation unit 163, which is used to calculate the normalized current medical record-drug description semantic position-by-position response feature vector. The feature vector is input into a learnable gating function to obtain a current medical record-drug description semantic response screening weight mask vector; a current medical record-drug description semantic response weight mask calculation unit 164 is used to calculate the position point multiplication between the current medical record-drug description semantic response screening weight mask vector and the normalized current medical record-drug description semantic position-by-position response feature vector to obtain a current medical record-drug description semantic position-by-position response distinguishable weight mask vector; a current medical record-drug description semantic interaction matching feature generation unit 165 is used to calculate the position point multiplication between the current medical record-drug description semantic position-by-position response distinguishable weight mask vector and the current medical record-drug description semantic position-by-position response feature vector to obtain the current medical record-drug description semantic interaction matching feature vector.

[0058] More specifically, in an embodiment of the present application, the current medical record-drug description semantic weight value calculation unit 163 is used to: input the normalized current medical record-drug description semantic position-by-position response feature vector into a sigmoid function to perform weight mask value calculation to obtain the current medical record-drug description semantic response screening weight mask vector.

[0059] In the embodiment of the present application, specifically, the current medical record-drug description semantic interaction matching module 160 is used to: process the current medical record word granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector through a feature interaction matching module based on a responsive distinguishable network with the following interaction matching formula to obtain the current medical record-drug description semantic interaction matching feature vector; wherein the interaction matching formula is:

[0060] ;

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] in, and are respectively the multimodal semantic fusion feature vector of the drug description and the global semantic association feature vector of the current medical record word granularity, represents the semantic position-by-position response feature vector of the current medical record-drug description, yes function, is the normalized current medical record-drug description semantic position-by-position response feature vector, yes function, is the current medical record-drug description semantic response screening weight mask vector, It is the point multiplication by position. is the current medical history-drug description semantic position-by-position response distinguishable weight mask vector, It is the semantic interaction matching feature vector of the current medical record-drug description.

[0066] In an embodiment of the present application, the recommendation result generation module 170 is used to obtain a recommendation result based on the semantic interaction matching feature of the current medical record-drug description, and the recommendation result is used to indicate whether the drug to be recommended is recommended to rheumatic patients. Specifically, in an embodiment of the present application, the recommendation result generation module 170 is used to: pass the current medical record-drug description semantic interaction matching feature vector through a classifier-based rheumatology and immunology drug recommender to obtain the recommendation result, and the recommendation result is used to indicate whether the drug to be recommended is recommended to rheumatic patients. That is, the current medical record-drug description semantic interaction matching feature obtained by interactively matching the word granularity global semantic association feature vector of the current medical record and the multimodal semantic fusion feature vector of the drug description is used for classification processing, so as to automatically determine whether to recommend the drug to be recommended to rheumatic patients. In this way, by performing word segmentation and semantic encoding on the medical record text of rheumatic patients through a deep learning algorithm, the contextual relationship and semantic information in the medical record can be more deeply understood. At the same time, it can identify the disease characteristics and individual differences of different patients, and explore the implicit semantic association patterns between the medical records and drug descriptions of different patients, thereby improving the matching degree between drug recommendations and the patient's actual condition, and providing patients with more personalized drug recommendations.

[0067] In particular, considering that the current medical record word-granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector respectively represent the word embedding semantic coding features of the current medical record information of rheumatic patients and the word embedding coding contextual association features of the drug description information of the recommended drugs, when performing feature interaction matching based on a responsive distinguishable network, it is necessary to compensate for the imbalance in responsive interaction matching caused by the imbalance in the correspondence ratio under the fine-grained distribution of different semantic feature encoding dimension sequences, so as to improve the expression effect of the current medical record-drug description semantic interaction matching feature vector, thereby improving the accuracy of the classification results.

[0068] Therefore, preferably, the present application corrects the current medical record-drug description semantic interaction matching feature vector when passing the current medical record-drug description semantic interaction matching feature vector through a classifier-based rheumatology and immunology drug recommender to obtain a recommendation result, including: calculating the first norm and second norm of the mean vector of the current medical record word granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector; calculating the weighted sum of the square root of the second norm and the first norm, and performing a dot multiplication with the dot-sum vector of the current medical record word granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector to obtain a first current medical record-drug description semantic interaction matching correction subvector; and calculating the first norm of the current medical record word granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector to obtain a first current medical record-drug description semantic interaction matching correction subvector. The dot product vector of the global semantic association feature vector and the multimodal semantic fusion feature vector of the drug description is multiplied by the square root of the length of the current medical record-drug description semantic interaction matching feature vector to obtain a second current medical record-drug description semantic interaction matching correction subvector; the weighted sum of the first current medical record-drug description semantic interaction matching correction subvector and the second current medical record-drug description semantic interaction matching correction subvector is calculated to obtain a current medical record-drug description semantic interaction matching correction vector; the current medical record-drug description semantic interaction matching correction vector is multiplied by the current medical record-drug description semantic interaction matching feature vector to obtain a corrected current medical record-drug description semantic interaction matching feature vector.

[0069] The current medical record-drug description semantic interaction matching correction vector is expressed as:

[0070] ;

[0071] in, and are respectively the global semantic association feature vector of the current medical record word granularity and the multimodal semantic fusion feature vector of the drug description, is the mean vector of the global semantic association feature vector of the current medical record word granularity and the multimodal semantic fusion feature vector of the drug description, is the one-norm of the eigenvector, is the reciprocal of the square root of the two-norm of the eigenvector, It is added by location point. It is the point multiplication by position. is the length of the semantic interaction matching feature vector of the current medical record-drug description, and and is the weighted sum weight as a hyperparameter, is the current medical record-drug description semantic interaction matching correction vector.

[0072] Therefore, based on the mean vector norm between the current medical record word-granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector, the constrained representation of the structured foreground and background distinction of the supermanifold of the current medical record word-granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector is used to model the feature-level key correspondence between the feature vectors of the current medical record word-granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector, and the global association relationship between the corresponding features is adjusted, so as to make positive fine-grained correspondence suggestions through the unbalanced proportion control between the corresponding feature values ​​of the current medical record word-granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector, so as to avoid the imbalance of the responsive interactive matching domain shift between the current medical record word-granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector by focusing on the focus.

[0073] In this way, by correcting the current medical record-drug description semantic interaction matching feature vector through the current medical record-drug description semantic interaction matching correction vector, the expression effect of the current medical record-drug description semantic interaction matching feature vector can be improved, thereby improving the accuracy of the recommendation result obtained by the classifier-based rheumatology and immunology drug recommender. In this way, by performing word segmentation and semantic encoding on the medical record text of rheumatic patients through deep learning algorithms, the contextual relationship and semantic information in the medical record can be more deeply understood. At the same time, it can identify the disease characteristics and individual differences of different patients, and explore the implicit semantic association patterns between the medical record information and drug descriptions of different patients, thereby improving the matching degree between drug recommendations and the actual condition of patients, and providing patients with more personalized drug recommendations.

[0074] In summary, the intelligent recommendation system 100 for rheumatism and immunology drugs based on the embodiment of the present application is explained, which first obtains the current medical record information of the rheumatic patient and the drug description information of the drug to be recommended, and then uses a text information processing and analysis algorithm based on deep learning to perform word segmentation and semantic encoding on the current medical record information and each description information in the drug description information, so as to automatically determine whether to recommend the drug to be recommended to the rheumatic patient based on the semantic fusion features between the various description information and the interactive matching information between the current medical record information. In this way, personalized rheumatic and immunology drug recommendations can be achieved according to the disease characteristics and individual differences of different patients.

[0075] As described above, the intelligent recommendation system 100 for rheumatoid and immunological drugs according to the embodiment of the present application can be implemented in various terminal devices, such as a server for intelligent recommendation of rheumatoid and immunological drugs. In one example, the intelligent recommendation system 100 for rheumatoid and immunological drugs 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 intelligent recommendation system 100 for rheumatoid and immunological drugs 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 intelligent recommendation system 100 for rheumatoid and immunological drugs can also be one of the many hardware modules of the terminal device.

[0076] Alternatively, in another example, the rheumatism and immunology drug intelligent recommendation system 100 and the terminal device may also be separate devices, and the rheumatism and immunology drug intelligent recommendation 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.

[0077] Figure 5 FIG. 1 is a flow chart of a method for intelligently recommending rheumatism and immunology drugs according to an embodiment of the present application. Figure 5As shown, in the intelligent recommendation method for rheumatic immunology drugs, it includes: S110, obtaining the current medical record information of the rheumatic patient; S120, obtaining the drug description information of the drug to be recommended, the drug description information including the drug action description, the drug side effect description and the drug interaction description; S130, performing word segmentation processing on the current medical record information of the rheumatic patient and passing it through the current medical record semantic encoder to obtain the current medical record word granularity global semantic association feature vector; S140, performing word segmentation processing on the drug description information and passing it through the drug description semantic encoder to obtain the drug action description word granularity semantic coding feature vector, the drug side effect description word granularity semantic coding feature vector and the drug interaction word granularity semantic coding feature vector; S150, using the drug action description word granularity semantic coding feature vector as the query vector, the drug side effect description word granularity semantic coding feature vector as the key vector and the The word-granular semantic encoding feature vector of drug-drug interaction is used as the value vector, and the word-granular semantic encoding feature vector of the drug action description, the word-granular semantic encoding feature vector of the drug side effect description and the word-granular semantic encoding feature vector of the drug-drug interaction are input into the drug description multimodal semantic joint encoder based on the Transformer mechanism to obtain a drug description multimodal semantic fusion feature vector; S160, the word-granular global semantic association feature vector of the current medical record and the drug description multimodal semantic fusion feature vector are passed through a feature interaction matching module based on a responsive differentiable network to obtain a current medical record-drug description semantic interaction matching feature vector as a current medical record-drug description semantic interaction matching feature; S170, based on the current medical record-drug description semantic interaction matching feature, a recommendation result is obtained, and the recommendation result is used to indicate whether the recommended drug is recommended to rheumatic patients.

[0078] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned rheumatism and immunity drug intelligent recommendation method have been described in the above reference. Figures 1 to 4 The description of the intelligent recommendation system for rheumatism and immunology drugs has been introduced in detail, and therefore, its repeated description will be omitted.

[0079] In summary, the intelligent recommendation method for rheumatism and immunology drugs based on the embodiment of the present application is explained, which first obtains the current medical record information of the rheumatic patient and the drug description information of the drug to be recommended, and then uses a text information processing and analysis algorithm based on deep learning to perform word segmentation and semantic encoding on the current medical record information and each description information in the drug description information, so as to automatically determine whether to recommend the drug to be recommended to the rheumatic patient based on the semantic fusion features between the various description information and the interactive matching information between the current medical record information. In this way, personalized rheumatic and immunology drug recommendations can be achieved according to the disease characteristics and individual differences of different patients.

[0080] The above is only a preferred embodiment of the present application and does not constitute any form of limitation to the present application. Although the present application has been disclosed as a preferred embodiment as above, it is not intended to limit the present application. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of the technical contents disclosed above without departing from the scope of the technical solution of the present application. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present application without departing from the content of the technical solution of the present application still falls within the scope of the technical solution of the present application.

Claims

1. An intelligent recommendation system for rheumatism and immunology drugs, characterized in that: include: A module for acquiring the current medical records of rheumatic patients, used to acquire the current medical records of rheumatic patients; A module for acquiring description information of the drug to be recommended, used to acquire description information of the drug to be recommended, wherein the description information includes description of drug action, description of drug side effects and description of drug interaction; A current medical record word granularity semantic encoding module is used to perform word segmentation processing on the current medical record information of the rheumatic patient and then pass it through a current medical record semantic encoder to obtain a current medical record word granularity global semantic association feature vector; A drug description information semantic encoding module, used for performing word segmentation processing on the drug description information and then passing it through a drug description semantic encoder to obtain a semantic encoding feature vector of drug action description word granularity, a semantic encoding feature vector of drug side effect description word granularity, and a semantic encoding feature vector of drug interaction word granularity; A drug description semantic feature multimodal fusion module, which is used to use the drug action description word granularity semantic encoding feature vector as a query vector, the drug side effect description word granularity semantic encoding feature vector as a key vector and the drug interaction word granularity semantic encoding feature vector as a value vector, and input the drug action description word granularity semantic encoding feature vector, the drug side effect description word granularity semantic encoding feature vector and the drug interaction word granularity semantic encoding feature vector into a drug description multimodal semantic joint encoder based on the Transformer mechanism to obtain a drug description multimodal semantic fusion feature vector; A current medical record-drug description semantic interaction matching module, used to obtain a current medical record-drug description semantic interaction matching feature vector as a current medical record-drug description semantic interaction matching feature by passing the current medical record word granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector through a feature interaction matching module based on a responsive distinguishable network; A recommendation result generating module, used for obtaining a recommendation result based on the semantic interaction matching feature of the current medical record-drug description, wherein the recommendation result is used to indicate whether to recommend the drug to be recommended to the rheumatic patient; The current medical record-drug description semantic interaction matching module includes: A current medical record-drug description point-by-point semantic association unit, used to calculate the position-by-position response between the current medical record word granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector to obtain the current medical record-drug description semantic position-by-position response feature vector; A current medical record-drug description point-by-point semantic association feature normalization unit, used to use a Softmax function to normalize the current medical record-drug description semantic position-by-position response feature vector to obtain a normalized current medical record-drug description semantic position-by-position response feature vector; A current medical record-drug description semantic weight value calculation unit, used for inputting the normalized current medical record-drug description semantic position-by-position response feature vector into a learnable gating function to obtain a current medical record-drug description semantic response screening weight mask vector; A current medical record-drug description semantic response weight mask calculation unit, used to calculate the position point multiplication between the current medical record-drug description semantic response screening weight mask vector and the normalized current medical record-drug description semantic position-by-position response feature vector to obtain a current medical record-drug description semantic position-by-position response distinguishable weight mask vector; A current medical record-drug description semantic interaction matching feature generation unit, used for calculating the position point multiplication between the current medical record-drug description semantic position-by-position response distinguishable weight mask vector and the current medical record-drug description semantic position-by-position response feature vector to obtain the current medical record-drug description semantic interaction matching feature vector; Among them, the current medical record-drug description semantic weight value calculation unit is used to: input the normalized current medical record-drug description semantic position-by-position response feature vector into the sigmoid function to calculate the weight mask value to obtain the current medical record-drug description semantic response screening weight mask vector.

2. The intelligent recommendation system for rheumatism and immunology drugs according to claim 1, characterized in that: The current medical record word granularity semantic encoding module is used to: perform word segmentation processing on the current medical record information of the rheumatic patient and then pass it through a current medical record semantic encoder including a word embedding layer to obtain the current medical record word granularity global semantic association feature vector.

3. The intelligent recommendation system for rheumatism and immunology drugs according to claim 2, characterized in that: The drug description information semantic encoding module is used to: perform word segmentation processing on the drug description information and then pass it through a drug description semantic encoder including a word embedding layer to obtain the drug action description word granularity semantic encoding feature vector, the drug side effect description word granularity semantic encoding feature vector and the drug interaction word granularity semantic encoding feature vector.

4. The intelligent recommendation system for rheumatism and immunology drugs according to claim 3 is characterized in that: The drug description semantic feature multimodal fusion module includes: a drug action description word granularity semantic coding weight calculation unit, used for multiplying the weight matrix of the query vector by the drug action description word granularity semantic coding feature vector to obtain a drug action description word granularity semantic coding weight vector; A drug side effect description word granularity semantic coding weight calculation unit, used for multiplying the weight matrix of the key vector with the drug side effect description word granularity semantic coding feature vector to obtain a drug side effect description word granularity semantic coding weight vector; a drug interaction word granularity semantic coding weight calculation unit, used for multiplying the weight matrix of the value vector with the drug interaction word granularity semantic coding feature vector to obtain a drug interaction word granularity semantic coding weight vector; a drug effect-drug side effect semantic association weight calculation unit, configured to calculate the product of the transposed vector of the drug effect description word granularity semantic coding weight vector and the drug side effect description word granularity semantic coding weight vector to obtain a drug effect-drug side effect description word granularity semantic association value, and then divide the drug effect-drug side effect description word granularity semantic association value by the square root of the length of the drug side effect description word granularity semantic coding weight vector to obtain a drug effect-drug side effect description word granularity semantic association weight value; A drug effect-drug side effect semantic association weight activation unit, used for inputting the drug effect-drug side effect description word granularity semantic association weight value into a softmax function to obtain a drug effect-drug side effect description word granularity semantic association activation weight value; The drug description semantic weighting unit is used to use the drug action-drug side effect description word granularity semantic association activation weight value as the weight value, and weight the transposed vector of the drug interaction word granularity semantic encoding weight vector to obtain the drug description multimodal semantic fusion feature vector.

5. The intelligent recommendation system for rheumatism and immunology drugs according to claim 4 is characterized in that: The drug description semantic weighting unit is used to calculate the product between the activation weight value of the drug effect-drug side effect description word granularity semantic association and the feature value of each position in the transposed vector of the drug interaction word granularity semantic encoding weight vector to obtain the drug description multimodal semantic fusion feature vector.

6. The intelligent recommendation system for rheumatism and immunology drugs according to claim 5, characterized in that: The recommendation result generation module is used to: pass the current medical record-drug description semantic interaction matching feature vector through a classifier-based rheumatology and immunology drug recommender to obtain the recommendation result, and the recommendation result is used to indicate whether the drug to be recommended is recommended to the rheumatic patient.

7. A method for intelligently recommending rheumatism and immunity drugs, using the intelligent recommendation system for rheumatism and immunity drugs according to claim 1, characterized in that: include: Obtain current medical record information of rheumatology patients; Obtaining drug description information of the drug to be recommended, wherein the drug description information includes a description of drug action, a description of drug side effects, and a description of drug interactions; The current medical record information of the rheumatic patient is segmented and passed through a current medical record semantic encoder to obtain a current medical record word granularity global semantic association feature vector; The drug description information is segmented and passed through a drug description semantic encoder to obtain a drug action description word granularity semantic encoding feature vector, a drug side effect description word granularity semantic encoding feature vector, and a drug interaction word granularity semantic encoding feature vector; Taking the drug action description word granularity semantic encoding feature vector as the query vector, the drug side effect description word granularity semantic encoding feature vector as the key vector and the drug interaction word granularity semantic encoding feature vector as the value vector, the drug action description word granularity semantic encoding feature vector, the drug side effect description word granularity semantic encoding feature vector and the drug interaction word granularity semantic encoding feature vector are input into a drug description multimodal semantic joint encoder based on the Transformer mechanism to obtain a drug description multimodal semantic fusion feature vector; The current medical record word granularity global semantic association feature vector and the drug description multimodal semantic fusion feature vector are passed through a feature interaction matching module based on a responsive distinguishable network to obtain a current medical record-drug description semantic interaction matching feature vector as a current medical record-drug description semantic interaction matching feature; Based on the semantic interaction matching feature of the current medical record-drug description, a recommendation result is obtained, and the recommendation result is used to indicate whether the drug to be recommended is recommended to the rheumatic patient.

8. The intelligent recommendation method for rheumatism and immunology drugs according to claim 7, characterized in that: The current medical record information of the rheumatic patient is processed by word segmentation and then passed through a current medical record semantic encoder to obtain a current medical record word-granularity global semantic association feature vector, including: the current medical record information of the rheumatic patient is processed by word segmentation and then passed through a current medical record semantic encoder including a word embedding layer to obtain a current medical record word-granularity global semantic association feature vector.

Images