Traditional Chinese medicine recommendation system based on multi-scale herbal medicine-symptom relation fusion framework

Abstract

The invention discloses a traditional Chinese medicine recommendation system based on a multi-scale herbal medicine-symptom relation fusion framework, and belongs to the field of traditional Chinese medicine intelligent recommendation. Preprocessing the clinical symptom data of the patient to obtain coded symptom data; the feature extraction and modeling module is used for converting the molecular structure data of the herbal medicine into a high-dimensional embedded vector; the recommendation generation module is used for generating a traditional Chinese medicine recommendation scheme by adopting a deep learning algorithm based on the coded symptom data and the high-dimensional embedded vector; and the user interaction module is used for providing a user interface and displaying the recommended traditional Chinese medicine scheme and explanatory information thereof.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent traditional Chinese medicine recommendation, and particularly relates to a traditional Chinese medicine recommendation system based on a multi-scale herbal-symptom relationship fusion framework. Background Art

[0002] In the field of traditional Chinese medicine recommendation, traditional methods mainly rely on the experience of traditional Chinese medicine experts and the compatibility rules of classic prescriptions. Although these methods can meet clinical needs to a certain extent, there are obvious deficiencies. First, it is difficult for traditional methods to quantify the complex relationship between herbs and symptoms, resulting in the lack of scientific basis and accuracy of the recommendation results. Second, traditional methods cannot make full use of the research results of modern pharmacology and molecular biology, limiting the practicality and universality of traditional Chinese medicine recommendation.

[0003] In view of the above problems, patent technologies in recent years have tried to use various methods to improve the accuracy and practicality of traditional Chinese medicine recommendation. For example, the prior art proposes an intelligent traditional Chinese medicine prescription recommendation system, which uses the feature vectors of symptoms and medicinal materials for intelligent recommendation through a data input module, a data processing module, and a traditional Chinese medicine prescription generation module. However, this system mainly focuses on single-scale features and fails to fully consider the multi-scale relationship between the molecular-level chemical features of herbs and clinical symptoms.

[0004] Another prior art proposes a graph convolutional herbal recommendation method based on multi-graph fusion, which formulates high-order correlations through a multi-graph structure and introduces a multi-hypergraph neural network structure to learn its joint representation. However, this method still has deficiencies in dealing with complex interaction relationships and long-range dependencies.

[0005] Although the existing patent solutions have been improved in some aspects, they still fail to comprehensively solve the problems of multi-scale feature fusion and complex interaction relationship modeling in traditional Chinese medicine recommendation. The present invention proposes a traditional Chinese medicine recommendation system based on multi-scale herbal-symptom relationships. Summary of the Invention

[0006] To solve the above technical problems, the present invention proposes a traditional Chinese medicine recommendation system based on a multi-scale herbal-symptom relationship fusion framework to solve the problems existing in the above prior art.

[0007] To achieve the above object, the present invention provides a traditional Chinese medicine recommendation system based on a multi-scale herbal-symptom relationship fusion framework, including:

[0008] A data input module, configured to collect the molecular structure data of herbs and the clinical symptom data of patients, and preprocess the clinical symptom data of the patients to obtain encoded symptom data;

[0009] A feature extraction and modeling module, configured to convert the molecular structure data of the herbs into high-dimensional embedding vectors;

[0010] A recommendation generation module, which is used to generate a traditional Chinese medicine recommendation plan by using a deep learning algorithm based on encoded symptom data and high-dimensional embedding vectors;

[0011] A user interaction module, which is used to provide a user interface to display the recommended traditional Chinese medicine plan and its explanatory information.

[0012] Optionally, the data input module includes: a symptom preprocessing unit and a symptom vectorization unit;

[0013] Among them, the symptom preprocessing unit is used to perform text cleaning and analysis on the clinical symptom data of the patient to obtain preprocessed symptom data;

[0014] The symptom vectorization unit is used to convert the preprocessed symptom data into a numerical vector through a word embedding model to obtain encoded symptom data.

[0015] Optionally, the feature extraction and modeling module includes: a heterogeneous attention relationship network unit, a multi-scale molecular graph embedding unit, and a fusion unit;

[0016] Among them, the heterogeneous attention relationship network unit is used to model the interaction relationship between herbs and symptoms through a graph attention mechanism to construct a heterogeneous graph;

[0017] The multi-scale molecular graph embedding unit is used to convert the molecular structure data of herbs into high-dimensional embedding vectors;

[0018] The fusion unit is used to obtain the global and long-range dependence relationships between herbs and symptoms based on a Transformer encoder.

[0019] Optionally, the multi-scale molecular graph embedding unit includes: a molecular graph generation subunit, a molecular embedding subunit, and an analysis structure information supplement subunit;

[0020] Among them, the molecular graph generation subunit is used to decompose the molecular structure data of herbs into molecular graphs;

[0021] The molecular embedding subunit is used to learn the embedding representation of the molecular graph based on a graph neural network and aggregate multiple chemical components of one herb to obtain an embedding vector;

[0022] The analysis structure information supplement subunit is used to supplement the missing information of the embedding vector to obtain high-dimensional embedding vectors.

[0023] Optionally, the fusion unit includes: a self-attention subunit and a global enhancement subunit;

[0024] Among them, the self-attention subunit is used to respectively obtain the global information of the high-dimensional feature vector of herbs and the encoded symptom data;

[0025] The global enhancer unit is used to aggregate the global information to obtain enhanced global information.

[0026] Optionally, the recommendation generation module includes: an intelligent recommendation module and a result output module;

[0027] Among them, the intelligent recommendation module is used to generate a traditional Chinese medicine recommendation plan;

[0028] The result output module is used to output the traditional Chinese medicine recommendation plan.

[0029] The present invention also provides a computer terminal device, including:

[0030] One or more processors;

[0031] A memory, coupled to the processor, for storing one or more programs;

[0032] When the one or more programs are executed by the one or more processors, the one or more processors implement a traditional Chinese medicine recommendation system based on a multi-scale herbal-symptom relationship fusion framework.

[0033] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, a traditional Chinese medicine recommendation system based on a multi-scale herbal-symptom relationship fusion framework is implemented.

[0034] Compared with the prior art, the present invention has the following advantages and technical effects:

[0035] The present invention proposes a Chinese medicine recommendation system based on a multi-scale herbal medicine-symptom relationship fusion framework, which has significant technical effects. First, the system collects the molecular structure data of herbs and the clinical symptom data of patients through a data input module, and pre-processes and encodes the symptom data, which can accurately capture the patient's condition characteristics and provide accurate input information for subsequent recommendations. Secondly, the feature extraction and modeling module converts the molecular structure data of herbs into high-dimensional embedding vectors. This high-dimensional embedding vector can fully retain the complex characteristics of herbs, so that the intrinsic properties of herbs are more comprehensively expressed. Furthermore, the recommendation generation module uses a deep learning algorithm to generate a Chinese medicine recommendation scheme based on the encoded symptom data and high-dimensional embedding vectors. The powerful fitting ability and generalization ability of the deep learning algorithm can dig out the complex nonlinear relationship between herbs and symptoms, thereby generating a more accurate and personalized Chinese medicine recommendation scheme. Finally, the user interaction module provides a user interface to display the recommended Chinese medicine scheme and its explanatory information, so that users can intuitively understand the recommendation results and their basis, and enhance the interpretability and user experience of the system. The traditional Chinese medicine recommendation system of the present invention can realize accurate and personalized traditional Chinese medicine recommendation, improve the pertinence and effectiveness of traditional Chinese medicine treatment, and provide strong technical support for the modernization and intelligent development of traditional Chinese medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0037] Figure 1 This is a flow chart of a Chinese medicine recommendation system based on a multi-scale herb-symptom relationship fusion framework according to an embodiment of the present invention;

[0038] Figure 2 This is a structural diagram of a traditional Chinese medicine recommendation system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0040] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0041] Embodiment 1

[0042] The "Multi-scale Herbal-Symptom Relationship Fusion Framework" proposed by the present invention comprehensively fuses the molecular-level chemical characteristics of herbs with the multi-scale relationships of clinical symptoms through key technologies such as heterogeneous attention relationship networks, multi-scale molecular graph embedding, and global enhanced long-range attention mechanisms, significantly improving the accuracy and robustness of traditional Chinese medicine recommendation. This not only provides strong support for clinical decision-making in traditional Chinese medicine but also brings higher recommendation efficiency and broader application prospects.

[0043] As Figure 1 - Figure 2 shown, in this embodiment, a traditional Chinese medicine recommendation system based on the multi-scale herbal-symptom relationship fusion framework is provided, including:

[0044] A data input module for collecting the molecular structure data of herbs and the clinical symptom data of patients, and preprocessing the clinical symptom data of the patients to obtain encoded symptom data.

[0045] Furthermore, the data input module includes: a symptom preprocessing unit and a symptom vectorization unit; wherein, the symptom preprocessing unit is used to perform text cleaning and analysis processing on the clinical symptom data of patients to obtain preprocessed symptom data; the symptom vectorization unit is used to convert the preprocessed symptom data into a numerical vector through a word embedding model to obtain encoded symptom data.

[0046] Furthermore, collect the molecular structure data of herbs (such as in SMILES format) and clinical symptom data. The process of preprocessing and standardizing the data includes: encoding of symptoms.

[0047] Encoding of symptoms is the process of converting the clinical symptoms of patients into a numerical representation that can be processed by a computer. This step is crucial for subsequent model training and recommendation. The following is the specific implementation process:

[0048] First, based on the symptom preprocessing unit, preprocess the symptom description and convert it into a unified format. As a specific implementation manner of this embodiment, the present invention uses text cleaning to remove irrelevant characters, punctuation marks, and stop words in the symptom description, and then uses a word segmentation method to split the symptom description into separate lexical units. For example, split "headache and fever" into two independent symptoms, "headache" and "fever".

[0049] Then, use the symptom vectorization unit to convert the preprocessed symptom data into a numerical vector. As a specific implementation manner of this embodiment, a pre-trained word embedding model (such as Word2Vec or BERT) is used to convert each symptom word into a high-dimensional vector. These models can capture the semantic relationships between words. If a symptom description contains multiple words, the embedding vectors of these words can be aggregated (such as taking the average) to obtain the overall embedding vector of the symptom.

[0050] A feature extraction and modeling module for converting the molecular structure data of the herbal medicine into a high-dimensional embedding vector.

[0051] Furthermore, the feature extraction and modeling module includes: a heterogeneous attention relationship network unit, a multi-scale molecular graph embedding unit, and a fusion unit; wherein, the heterogeneous attention relationship network unit is used to model the interaction relationship between the herbal medicine and the symptoms through a graph attention mechanism to construct a heterogeneous graph; the multi-scale molecular graph embedding unit is used to convert the molecular structure data of the herbal medicine into a high-dimensional embedding vector; the fusion unit is used to obtain the global and long-range dependence relationship between the herbal medicine and the symptoms based on a Transformer encoder.

[0052] Even further, a graph attention mechanism is adopted to uniformly model the complex interaction relationship between the herbal medicine and the symptoms. The nodes in the network include herbal medicine nodes and symptom nodes, and the edges represent the interaction relationship between the herbal medicine and the symptoms. The weights of the edges are dynamically calculated through the attention mechanism. As a specific implementation manner of this embodiment, the following steps are included:

[0053] 1) Extraction of feature vectors: Each node (whether it is a herbal medicine or a symptom) has a feature vector. These feature vectors can be pre-defined attributes (such as the chemical components of the herbal medicine, the clinical characteristics of the symptoms, etc.), or low-dimensional representations obtained through a certain embedding method.

[0054] 2) Calculation of attention scores: For each edge connecting two nodes, an attention score needs to be calculated to represent the correlation between these two nodes. The specific steps are as follows: Assume there is an edge between node A and node B, and their feature vectors are h_A and h_B respectively. Concatenate these two feature vectors to form a new vector [h_A|h_B]. Use a learnable weight matrix W to transform this concatenated vector to obtain an attention score: a AB = LeakyReLU(W·[h A ||h B ) where LeakyReLU is an activation function used to introduce non-linearity.

[0055] 3) Normalization of attention scores: To ensure that the sum of the attention scores of each node is 1 (i.e., the sum of the weights of the neighbor nodes of each node is 1), the attention scores need to be normalized. The specific method is as follows: For node A, assume it has multiple neighbor nodes (such as nodes B, C, and D), and calculate the attention scores a_AB, a_AC, and a_AD between them and A respectively. Use the softmax function to normalize these scores:

[0056]

[0057] Among them, a AB represents the normalized attention weight, indicating the importance of node B to node A.

[0058] 4) Aggregate the information of neighboring nodes: Each node aggregates the information of its neighboring nodes according to the normalized attention weights. Specifically: For node A, its new feature vector can be calculated in the following way: This means that the new feature vector of node A is the weighted sum of the feature vectors of all its neighboring nodes, and the weights are the normalized attention scores.

[0059] This embodiment can effectively capture the complex interaction relationship between herbs and symptoms, improve the representation ability of the model. At the same time, the dynamic weight calculation enables the model to adaptively adjust the attention to different relationships.

[0060] Furthermore, the multi-scale molecular graph embedding unit, based on the principle of network pharmacology, transforms the molecular structure of herbs into high-dimensional embedding vectors, and the embedding vectors contain molecular features at multiple scales (such as atomic level, molecular level, etc.).

[0061] The multi-scale molecular graph embedding unit includes: a molecular graph generation subunit, a molecular embedding subunit, and an analysis structure information supplement subunit; among them, the molecular graph generation subunit is used to decompose the molecular structure data of herbs into molecular graphs; the molecular embedding subunit is used to learn the embedding representation of the molecular graph based on a graph neural network and aggregate multiple chemical components of one herb to obtain an embedding vector; the analysis structure information supplement subunit is used to supplement the missing information of the embedding vector to obtain a high-dimensional embedding vector.

[0062] As a specific implementation manner of this embodiment, it includes the following steps: 1) SMILES representation: The chemical components of herbs can be described by their molecular structures, and the molecular structures can be represented by SMILES (Simplified Molecular Input LineEntry System) strings. SMILES is a coding method for describing molecular structures in text form, which can concisely represent complex molecular structures. The SMILES string is a compact representation of the molecular structure and can be easily parsed by computer programs.

[0063] 2) From SMILES to molecular graph: In order to transform the SMILES string into a high-dimensional embedding vector, first, the molecular graph generation subunit needs to parse it into a molecular graph. A molecular graph is a graph structure, where: nodes represent atoms (such as carbon, oxygen, hydrogen, etc.). Edges represent chemical bonds (such as single bonds, double bonds, etc.). This process can be completed by chemical informatics tools (such as RDKit or OpenBabel), and these tools can parse the SMILES string into a molecular graph.

[0064] 3) The molecular embedding subunit uses a GNN to generate molecular embeddings: After obtaining the molecular graph, the next step is to use a graph neural network (GNN) to generate a high-dimensional embedding vector of the molecule. A GNN is a neural network specifically designed to process graph-structured data, capable of learning the embedding representation of nodes while preserving the structural information of the graph. The specific steps are as follows:

[0065] Initialize node features: The initial features of each atomic node can include the atomic type (such as carbon, oxygen, etc.) and the chemical properties of the atom (such as electronegativity, atomicity, etc.).

[0066] Message passing: Through the message passing mechanism of the GNN, each atomic node aggregates the information of its neighbor nodes and updates its own feature representation. This process can be expressed as:

[0067]

[0068] where is the feature of node v at the l+1 layer, and N(v) is the set of neighbor nodes of node v.

[0069] Readout function: After multiple layers of message passing, the GNN generates the final embedding vector for each node. For the entire molecule, all atomic node embeddings can be aggregated into a molecular embedding vector through a readout function (such as average pooling or max pooling):

[0070]

[0071] 4) Aggregate the molecular embeddings of herbs: An herb may contain multiple chemical components, so it is necessary to aggregate the embedding vectors of these components. For each herb, add and average the embedding vectors of all its chemical components:

[0072]

[0073] where N is the number of chemical components of the herb, and h molecule,i is the embedding vector of the i-th component.

[0074] 5) The structural information supplementation subunit supplements the missing molecular structural information: In practical applications, the molecular structural information of some herbs may be missing. To fill in this missing information, a variational autoencoder (VAE) is trained using the known herb molecular embedding vectors and their corresponding attributes (such as the nature and flavor of the herb, meridian tropism, etc.). For herbs with missing molecular structures, their corresponding molecular embedding vectors are generated through the VAE model using their known attribute vectors.

[0075] Furthermore, the fusion unit uses a Transformer encoder to capture the global and long-range dependencies between the herbal medicine and symptom features.

[0076] The fusion unit includes: a self-attention sub-unit and a global enhancement sub-unit; wherein, the self-attention sub-unit is used to respectively obtain the high-dimensional feature vector of the herbal medicine and the global information of the encoded symptom data; the global enhancement sub-unit is used to aggregate the global information to obtain enhanced global information.

[0077] Through the self-attention mechanism, the model can focus on the long-range interactions between features and introduce a global enhancement strategy to further improve the ability to capture global information. In the model design, in order to better understand the complex relationship between herbal medicine and symptoms, it is necessary to focus on the long-range interactions between features. This means that the model not only has to consider local information but also be able to capture global information. For this reason, the self-attention mechanism is adopted in the paper and a global enhancement strategy is introduced. As a specific implementation manner of this embodiment, it includes the following content:

[0078] 1) The role of the self-attention mechanism: The core idea of the self-attention mechanism is to enable the model to dynamically focus on the important relationships between different features, rather than just local and adjacent relationships. Specifically, it can help the model capture the long-range dependencies between features, that is, those interactions that are far apart but still important. For example, suppose we have a set of features (such as the chemical properties of herbal medicine or the clinical manifestations of symptoms), the self-attention mechanism will calculate the correlation (or similarity) between each feature and all other features and assign weights according to these correlations. In this way, the model can adjust the importance of each feature based on global information.

[0079] 2) The specific process of the self-attention mechanism:

[0080] Calculate the attention scores: For each feature (the high-dimensional feature vector of the herbal medicine and the encoded symptom data), calculate its correlation (or similarity) with all other features to obtain an attention score. This score reflects the intensity of the interaction between features.

[0081] Normalize the attention scores: To ensure that the sum of the attention scores of each feature is 1, use a normalization method (such as the softmax function) to process these scores. In this way, each feature will obtain a weight according to its correlation with other features.

[0082] Weighted summation: Combine the weighted representations of each feature with other features to obtain a new feature representation. This process can be regarded as the information exchange between features, enabling each feature to "see" the information of other features.

[0083] 3) The global enhancer unit introduces a global enhancement strategy: In addition to capturing long-range interactions, the paper also introduces a global enhancement strategy to further improve the model's ability to capture global information. The core idea of the global enhancement strategy is to enable the model to better understand and utilize the overall information, rather than just local features.

[0084] Specifically, the global enhancement strategy includes the following:

[0085] Aggregation of global information: Introduce a special token (or vector) in the feature representation to aggregate global information. This token can be regarded as a "global node" that can collect and integrate the information of all features.

[0086] Enhancing the role of global information: During the model's calculation process, pay special attention to the information of this global node and combine it with the information of other features. In this way, when making decisions, the model will not only consider local information but also refer to global information.

[0087] Capturing long-range dependencies: Through the self-attention mechanism, the model can capture the long-range dependency relationships between features. Combining with the global enhancement strategy, the model can better understand the complex relationships between features, thereby improving the overall performance.

[0088] 4) Specific implementation process: The following is the specific implementation process of combining the self-attention mechanism and the global enhancement strategy.

[0089] Initialization of feature representation: Assume there is a set of feature vectors X = [x1, x2,..., xn], where each xi represents a feature (such as the chemical properties of herbs or the manifestations of symptoms).

[0090] Introduce the global node: Add a special global node g to the feature vector to aggregate global information. Initially, g can be set as the average value of all feature vectors:

[0091]

[0092] Calculate the attention scores: For each feature xi, calculate the attention scores between it and the global node g as well as other features xj:

[0093] a ij = Attention(x i , x j )

[0094] where Attention is an attention function, which can be a simple dot product or a more complex neural network.

[0095] Normalized attention scores: The attention scores are normalized using the softmax function to obtain the weights for each feature:

[0096]

[0097] Weighted summation and global information update: The features are weighted and summed to obtain the updated feature representation:

[0098]

[0099] At the same time, the information of the global node g is updated:

[0100] Fusion of global information: The information of the global node is fused into the representation of each feature to further enhance the role of global information: x″ i = x′ i + g′

[0101] Final output: The updated feature representation is used for subsequent calculation or prediction tasks.

[0102] Example 2

[0103] Collect the molecular structure data and clinical symptom data of herbs, and perform preprocessing and standardization. As a specific implementation of this example, collect the symptom data of diabetic complication patients (such as proteinuria, blurred vision, numbness in limbs, etc.), and extract the corresponding herbal prescription data. At the same time, collect the molecular structure information of relevant herbs.

[0104] Construct a heterogeneous graph, where the nodes include herb nodes and symptom nodes, and the edges represent the interactions between herbs and symptoms.

[0105] Input the preprocessed data into the multi-scale herb-symptom relationship fusion framework for model training.

[0106] Optimize the model parameters to ensure that the model can effectively capture the multi-scale relationships between herbs and symptoms.

[0107] Deploy the trained model to the recommendation system and receive the input from traditional Chinese medicine doctors through the user interface.

[0108] Generate a traditional Chinese medicine recommendation plan in real time according to the input symptom information and provide explanatory information.

[0109] Evaluate the recommendation accuracy and practicality of the system through clinical verification and user feedback.

[0110] Continuously optimize the model and system functions to improve the recommendation effect.

[0111] The above are only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A traditional Chinese medicine recommendation system based on a multi-scale herbal medicine-symptom relationship fusion framework, characterized in that, Comprising: A data input module for collecting molecular structure data of Chinese herbs and clinical symptom data of patients, and preprocessing the clinical symptom data of the patients to obtain encoded symptom data; A feature extraction and modeling module for converting the molecular structure data of the Chinese herbs into high-dimensional embedding vectors; A recommendation generation module for generating a Chinese medicine recommendation scheme based on the encoded symptom data and the high-dimensional embedding vectors using a deep learning algorithm; A user interaction module for providing a user interface to display the recommended Chinese medicine scheme and its explanatory information.

2. The traditional Chinese medicine recommendation system based on the multi-scale herbal medicine-symptom relationship fusion framework according to claim 1, wherein The data input module includes: a symptom preprocessing unit and a symptom vectorization unit; Wherein, the symptom preprocessing unit is used for text cleaning and analysis processing of the clinical symptom data of the patients to obtain preprocessed symptom data; The symptom vectorization unit is used for converting the preprocessed symptom data into a numerical vector through a word embedding model to obtain encoded symptom data.

3. The traditional Chinese medicine recommendation system based on the multi-scale herbal medicine-symptom relationship fusion framework according to claim 1, characterized in that The feature extraction and modeling module includes: a heterogeneous attention relationship network unit, a multi-scale molecular graph embedding unit, and a fusion unit; Wherein, the heterogeneous attention relationship network unit is used for modeling the interaction relationship between Chinese herbs and symptoms through a graph attention mechanism to construct a heterogeneous graph; The multi-scale molecular graph embedding unit is used for converting the molecular structure data of Chinese herbs into high-dimensional embedding vectors; The fusion unit is used for obtaining the global and long-range dependence relationships between Chinese herbs and symptoms based on a Transformer encoder.

4. The traditional Chinese medicine recommendation system based on the multi-scale herbal medicine-symptom relationship fusion framework according to claim 3, wherein The multi-scale molecular graph embedding unit includes: a molecular graph generation subunit, a molecular embedding subunit, and an analysis structure information supplement subunit; Wherein, the molecular graph generation subunit is used for decomposing the molecular structure data of Chinese herbs into molecular graphs; The molecular embedding subunit is used for learning the embedding representation of the molecular graph based on a graph neural network and aggregating multiple chemical components of one Chinese herb to obtain an embedding vector; The analysis structure information supplement subunit is used for supplementing the missing information of the embedding vector to obtain a high-dimensional embedding vector.

5. The traditional Chinese medicine recommendation system based on the multi-scale herbal medicine-symptom relationship fusion framework according to claim 3, wherein The fusion unit includes: a self-attention subunit and a global enhancement subunit; Wherein, the self-attention subunit is used for respectively obtaining the global information of the high-dimensional feature vector of Chinese herbs and the encoded symptom data; The global enhancement subunit is used for aggregating the global information to obtain enhanced global information.

6. The traditional Chinese medicine recommendation system based on the multi-scale herbal medicine-symptom relationship fusion framework according to claim 1, characterized in that The recommendation generation module includes: an intelligent recommendation module and a result output module; Wherein, the intelligent recommendation module is used for generating a Chinese medicine recommendation scheme; The result output module is used for outputting the Chinese medicine recommendation scheme.

7. A computer terminal device, characterized in that, Comprising: One or more processors; A memory coupled to the processor for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the Chinese medicine recommendation system based on the multi-scale Chinese herb-symptom relationship fusion framework as described in any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the Chinese medicine recommendation system based on the multi-scale Chinese herb-symptom relationship fusion framework as described in any one of claims 1-6.

Images