Acupoint compatibility prediction method, system and equipment and medium

Abstract

The invention provides an acupoint compatibility prediction method and device, equipment and a medium, and relates to the technical field of traditional Chinese medicine acupoint compatibility prediction.The method comprises the steps that acupoint compatibility information and acupoint attribute information of to-be-predicted acupoint compatibility are obtained, and an acupoint feature composite graph is constructed based on the acupoint compatibility information and the acupoint attribute information; constructing a de-noised acupoint feature matrix on the basis of a deep non-negative matrix factorization method according to the acupoint feature composite graph, training the de-noised acupoint feature matrix through a residual image attention network model to obtain a target acupoint feature matrix, and obtaining an association relationship of compatibility of the to-be-predicted acupoints according to the target acupoint feature matrix so as to predict the compatibility of the to-be-predicted acupoints. A deep non-negative matrix factorization method is used for carrying out layer-by-layer factorization and fine adjustment on an adjacent matrix of the acupoint feature composite graph to construct a de-noised acupoint feature matrix, random noise in the network can be relieved, comprehensive and effective acupoint feature representation is obtained through a residual image attention network model, and the prediction accuracy is effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of acupoint compatibility prediction in traditional Chinese medicine, and in particular to an acupoint compatibility prediction method, device, equipment and medium. Background Art

[0002] Acupuncture is an important treatment method in the traditional Chinese medicine system. Due to the unique therapeutic effects and mechanisms of different acupoints, coordinated stimulation of multiple acupoints is often used to regulate the balance of qi and blood, maintain the patency of meridians, and thus improve the treatment effect. However, due to the reliance on traditional experience, Chinese medicine practitioners may rarely consider innovation and optimization when formulating treatment plans, so that some potential acupoint combinations and treatment plans may not be able to fully realize their potential.

[0003] Many real-world systems can be represented as complex networks, such as networks in traditional Chinese medicine, biology, and social systems. In these networks, link prediction is a well-known method for finding missing links and identifying false links. For example, in bioinformatics networks, it is very time-consuming and costly to experimentally verify all possible interaction relationships in networks such as protein-protein interaction networks and metabolic networks. Link prediction can help researchers predict the most likely interactions, so that they can focus on experimental verification of these predictions and significantly reduce research costs. Accurately predicting acupoint combinations based on advanced link prediction methods can maximize the efficacy of acupuncture and expand its scope of application. However, existing link prediction methods usually focus on mapping acupoints and their compatibility relationships into low-dimensional space and capturing the relationships between acupoints. These methods may not fully capture the complex relationships between acupoints, and the predictions are easily interfered by noise, which seriously affects the accuracy of acupoint combination prediction results. Summary of the invention

[0004] The purpose of this application is to address the above problems and provide an acupoint combination prediction method, system, device and medium to alleviate random noise in the network, obtain a comprehensive and effective acupoint feature representation, effectively improve prediction accuracy, and provide valuable reference and guidance for acupuncture clinical practice.

[0005] In a first aspect, the present application provides a method for predicting acupoint combinations, comprising:

[0006] Obtaining acupoint combination information and acupoint attribute information of the acupoint combination to be predicted;

[0007] Constructing an acupoint characteristic synthesis map based on the acupoint combination information and the acupoint attribute information;

[0008] According to the acupoint feature synthesis graph, a denoised acupoint feature matrix is ​​constructed based on a deep non-negative matrix decomposition method;

[0009] The denoised acupoint feature matrix is ​​trained through a residual graph attention network model to obtain a target acupoint feature matrix;

[0010] The correlation relationship of the acupoint combination to be predicted is obtained according to the target acupoint feature matrix.

[0011] According to the technical solution provided by the present application, the construction of acupoint feature synthesis map based on the acupoint compatibility information and the acupoint attribute information includes:

[0012] Based on the feature-aware connection fusion strategy, an acupoint compatibility map is constructed according to the acupoint compatibility information;

[0013] Based on the cosine function, constructing an acupoint attribute similarity matrix according to the acupoint attribute information;

[0014] The acupoint feature synthesis map is constructed according to the acupoint compatibility map and the acupoint attribute similarity matrix.

[0015] According to the technical solution provided by the present application, the method of constructing a denoised acupoint feature matrix based on the acupoint feature synthesis map based on a deep non-negative matrix decomposition method includes:

[0016] Based on the deep non-negative matrix factorization method, the adjacency matrix of the acupoint feature synthesis graph is decomposed at different network layers to obtain the factor matrix.

[0017] The factor matrix is ​​updated to obtain a correction factor matrix, and the matrix factors of each network layer are fine-tuned based on the correction factor matrix to obtain a denoised acupoint feature matrix.

[0018] According to the technical solution provided by the present application, the factor matrix includes a base matrix and a coefficient matrix; accordingly, the adjacency matrix of the acupoint feature synthesis graph is decomposed at different network layers based on the deep non-negative matrix decomposition method to obtain the factor matrix, including:

[0019] Based on the deep non-negative matrix factorization method, the adjacency matrix of the acupoint feature synthesis graph is decomposed in the first preset network layer by minimizing the first preset objective function to obtain a basis matrix and a coefficient matrix;

[0020] Based on a deep non-negative matrix decomposition method, the coefficient matrix is ​​decomposed at a second preset network layer by minimizing a second preset objective function.

[0021] According to the technical solution provided by the present application, the factor matrix is ​​updated to obtain a correction factor matrix, and the matrix factors of each network layer are fine-tuned based on the correction factor matrix to obtain a denoised acupoint feature matrix, including:

[0022] Adding non-negative constraints to the basis matrix and the coefficient matrix by removing irrelevant variables and introducing a Lagrange multiplier matrix to obtain a modified basis matrix and a modified coefficient matrix;

[0023] According to the modified base matrix and the modified coefficient matrix, the base matrix or coefficient matrix corresponding to each layer of the network is fine-tuned by alternately minimizing the third preset objective function to obtain the denoised acupoint feature matrix.

[0024] According to the technical solution provided in the present application, the denoised acupoint feature matrix is ​​trained through a residual graph attention network model to obtain a target acupoint feature matrix, including:

[0025] The denoised acupoint feature matrix is ​​taken as input, mapped to the feature space through a fully connected layer, and the attention weights between acupoints are dynamically learned using the attention mechanism of the graph attention network;

[0026] A multi-head attention mechanism is used to capture the features of acupoints and obtain the intermediate acupoint feature matrix;

[0027] The intermediate acupoint feature matrix is ​​updated through the residual structure to obtain the target acupoint feature matrix of the acupoint.

[0028] According to the technical solution provided in the present application, the step of obtaining the correlation relationship of the acupoint combinations to be predicted according to the target acupoint feature matrix includes:

[0029] According to the formula: Calculating the association relationship of obtaining the acupoint combination to be predicted, wherein: is the predicted association score between acupoints i and j, sigmoid is the activation function, is the transpose of the feature representation vector of acupoint i, k j is the feature representation vector of acupoint j.

[0030] In a second aspect, the present application provides an acupoint combination prediction device, comprising:

[0031] An acquisition module is used to acquire acupoint combination information and acupoint attribute information of acupoint combination to be predicted;

[0032] A first construction module is used to construct an acupoint feature synthesis map based on the acupoint compatibility information and the acupoint attribute information;

[0033] A second construction module is used to construct a denoised acupoint feature matrix based on the acupoint feature synthesis map and a deep non-negative matrix decomposition method;

[0034] A training module, used for training the denoised acupoint feature matrix through a residual graph attention network model to obtain a target acupoint feature matrix;

[0035] The prediction module is used to obtain the correlation relationship of the acupoint combination to be predicted according to the target acupoint feature matrix.

[0036] In a third aspect, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described above when executing the computer program.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and the computer program implements the method described above when executed by a processor.

[0038] Compared with the prior art, the beneficial effects of the present application are as follows: the acupoint compatibility prediction method provided by the present application obtains the acupoint compatibility information and acupoint attribute information of the acupoint compatibility to be predicted, constructs an acupoint feature synthesis graph based on the acupoint compatibility information and the acupoint attribute information, constructs a denoised acupoint feature matrix based on the acupoint feature synthesis graph based on a deep non-negative matrix decomposition method, obtains a target acupoint feature matrix by training the denoised acupoint feature matrix through a residual graph attention network model, obtains the correlation relationship of the acupoint compatibility to be predicted based on the target acupoint feature matrix, and uses a deep non-negative matrix decomposition method to decompose and fine-tune the adjacency matrix of the acupoint feature synthesis graph layer by layer to construct a denoised acupoint feature matrix, which can effectively suppress the influence of noise. Reduce the interference of redundant information on the prediction results; use residual learning to retain the key information in the acupoint feature synthesis graph structure, reduce information loss in the multi-layer network transmission process, and enable the network to more deeply explore the potential relationship between acupoints; use the graph attention mechanism to automatically learn and identify which acupoints have a closer or more important relationship, so as to give higher weights in the prediction process, which helps to capture the nonlinear relationship and complex interaction between acupoints; gradually extract the characteristic information of acupoints through a multi-layer network structure, and finally output the prediction results of the association relationship of the acupoints to be predicted. These prediction results can reflect the difference in the effects of different acupoint combinations in the treatment of specific diseases or symptoms, and provide valuable reference and guidance for acupuncture clinical practice.

[0039] It should be understood that the description of technical features, technical solutions, beneficial effects or similar language in this application does not imply that all features and advantages can be realized in any single embodiment. On the contrary, it is understood that the description of features or beneficial effects means that specific technical features, technical solutions or beneficial effects are included in at least one embodiment. Therefore, the description of technical features, technical solutions or beneficial effects in this specification does not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions and beneficial effects described in the present embodiment can also be combined in any appropriate manner. Those skilled in the art will understand that the embodiment can be realized without one or more specific technical features, technical solutions or beneficial effects of a specific embodiment. In other embodiments, additional technical features and beneficial effects can also be identified in a specific embodiment that does not embody all embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solution in this embodiment, the drawings required for use in the description of the embodiment will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A flowchart of an acupoint combination prediction method provided in an embodiment of the present application;

[0042] Figure 2 A schematic diagram of an acupoint combination prediction device provided in an embodiment of the present application;

[0043] Figure 3 A schematic diagram of a computer system of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to better understand the technical solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings. The description in this section is only exemplary and explanatory and should not have any limiting effect on the protection scope of the present application. Specifically, the described embodiments are only embodiments of a part of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work should fall within the scope of protection of the present application.

[0045] It should be noted that similar reference numerals and letters represent similar items in the following figures, so once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0046] In order to facilitate the understanding of this application, a brief introduction to the related technology is first given:

[0047] Acupoint combination is an important concept in TCM acupuncture, which refers to the combination of two or more acupoints based on TCM theory and clinical experience to achieve better therapeutic effects. Correct acupoint combination can enhance the therapeutic effect of acupuncture and regulate the balance of qi, blood, yin and yang in the human body. When performing acupoint combination, the following principles are usually considered: 1. Synergy: select acupoints that can enhance each other; 2. Antagonism: for complex conditions, select acupoints that can restrain and balance each other; 3. Near and far acupoint combination: combine acupoints near the lesion site and distal related acupoints; 4. Upper and lower acupoint combination: acupoints in the upper and lower parts cooperate with each other, such as diseases of the head and face can be combined with acupoints in the lower limbs; 5. Syndrome differentiation and acupoint combination: according to the patient's specific symptoms and constitution, syndrome differentiation and treatment are carried out, and the corresponding acupoints are selected. Due to the reliance on traditional experience, TCM practitioners may rarely consider innovation and optimization when formulating treatment plans, so that some potential acupoint combinations and treatment plans may not be able to fully realize their potential.

[0048] In view of this, an embodiment of the present application provides an acupoint compatibility prediction method, which obtains acupoint compatibility information and acupoint attribute information of the acupoint compatibility to be predicted, constructs an acupoint feature synthesis map based on the acupoint compatibility information and the acupoint attribute information, constructs a denoised acupoint feature matrix based on the acupoint feature synthesis map based on a deep non-negative matrix decomposition method, and uses the denoised acupoint feature matrix to train a residual graph attention network model to obtain a target acupoint feature matrix, obtains the correlation relationship of the acupoint compatibility to be predicted based on the target acupoint feature matrix, and uses the deep non-negative matrix decomposition method to decompose and fine-tune the adjacency matrix of the acupoint feature synthesis map layer by layer to construct a denoised acupoint feature matrix, which can effectively suppress the influence of noise and reduce the impact of redundant information on prediction. The results are disturbed; residual learning is used to retain the key information in the acupoint feature synthesis graph structure, and the information loss is reduced during the multi-layer network transmission process, so that the network can more deeply explore the potential relationship between acupoints; the graph attention mechanism is used to automatically learn and identify which acupoints have a closer or more important relationship, so as to give higher weights in the prediction process, which helps to capture the nonlinear relationship and complex interaction between acupoints; the characteristic information of acupoints is gradually extracted through a multi-layer network structure, and finally the correlation relationship prediction results of the acupoints to be predicted are output, which effectively improves the prediction accuracy. These prediction results can reflect the difference in the effects of different acupoint combinations in the treatment of specific diseases or symptoms, and provide valuable reference and guidance for acupuncture clinical practice.

[0049] In order to make the technical solution of the present application clearer and easier to understand, the acupoint combination prediction method provided in the embodiment of the present application is introduced below.

[0050] like Figure 1 As shown in FIG. 1 , the figure is a flow chart of a method for predicting acupoint combinations provided in this embodiment, and the method comprises the following steps:

[0051] S101, obtaining acupoint combination information and acupoint attribute information of acupoint combination to be predicted;

[0052] Specifically, the acupoint compatibility information and acupoint attribute information of the acupoint compatibility to be predicted are obtained; wherein the acupoint compatibility information includes the acupoint compatibility information including the co-occurrence relationship between acupoints and acupoints, for example, the acupoint compatibility information for acupuncture treatment of migraine: Yanglingquan-Waiguan, Yanglingquan-Touwei, Yanglingquan-Fengchi, etc.; the acupoint attribute information includes: meridians, related viscera, distribution site, nerve distribution, source, acupuncture operation, and contraindications, for example, the attribute information of Yanglingquan includes: foot Shaoyang gallbladder meridian point, gallbladder, lower limb, sciatic nerve, "Lingshu", straight puncture 1-1.5 inches, no contraindications. Of course, the acupoint compatibility information and acupoint attribute information can also include other parameter information, which can be set and adjusted according to actual conditions, and are not specifically limited here.

[0053] S102, constructing an acupoint feature synthesis map based on the acupoint compatibility information and the acupoint attribute information;

[0054] Specifically, firstly, based on the feature-aware connection fusion strategy, an acupoint compatibility map is constructed according to the acupoint compatibility information, then based on the cosine function, an acupoint attribute similarity matrix is ​​constructed according to the acupoint attribute information, and then according to the acupoint compatibility map and the acupoint attribute similarity matrix, the acupoint feature synthesis map is constructed.

[0055] S103, constructing a denoised acupoint feature matrix based on the acupoint feature synthesis graph and a deep non-negative matrix decomposition method;

[0056] Specifically, based on the deep non-negative matrix decomposition method, the adjacency matrix of the acupoint feature synthesis graph is decomposed at different network layers to obtain a factor matrix, and the factor matrix is ​​updated to obtain a correction factor matrix, and the matrix factors of each network layer are fine-tuned based on the correction factor matrix to obtain a denoised acupoint feature matrix. The factor matrix includes at least a base matrix and a coefficient matrix.

[0057] S104, training the denoised acupoint feature matrix through a residual graph attention network model to obtain a target acupoint feature matrix;

[0058] Specifically, the denoised acupoint feature matrix is ​​first taken as input, mapped to the feature space through a fully connected layer, and the attention weights between acupoints are dynamically learned using the attention mechanism of the graph attention network. Then, a multi-head attention mechanism is used to capture the features of the acupoints to obtain the intermediate acupoint feature matrix. Then, the intermediate acupoint feature matrix is ​​updated through the residual structure to obtain the target acupoint feature matrix of the acupoint.

[0059] S105, acquiring the correlation relationship of the acupoint combinations to be predicted according to the target acupoint feature matrix.

[0060] Specifically, the correlation relationship of the acupoint combinations to be predicted is calculated according to the target acupoint feature matrix, so as to predict whether there is a correlation between the acupoints.

[0061] The acupoint compatibility prediction method provided by the present application obtains the acupoint compatibility information and acupoint attribute information of the acupoint compatibility to be predicted, constructs an acupoint feature synthesis map based on the acupoint compatibility information and the acupoint attribute information, constructs a denoised acupoint feature matrix based on the acupoint feature synthesis map based on a deep non-negative matrix decomposition method, trains the denoised acupoint feature matrix through a residual graph attention network model to obtain a target acupoint feature matrix, obtains the correlation relationship of the acupoint compatibility to be predicted based on the target acupoint feature matrix, and uses a deep non-negative matrix decomposition method to decompose and fine-tune the adjacency matrix of the acupoint feature synthesis map layer by layer to construct a denoised acupoint feature matrix, which can effectively suppress the influence of noise and reduce the impact of redundant information on prediction. The results are disturbed; residual learning is used to retain the key information in the acupoint feature synthesis graph structure, and the information loss is reduced during the multi-layer network transmission process, so that the network can more deeply explore the potential relationship between acupoints; the graph attention mechanism is used to automatically learn and identify which acupoints have a closer or more important relationship, so as to give higher weights in the prediction process, which helps to capture the nonlinear relationship and complex interaction between acupoints; the characteristic information of acupoints is gradually extracted through a multi-layer network structure, and finally the correlation relationship prediction results of the acupoint compatibility to be predicted are output. These prediction results can reflect the difference in the effects of different acupoint combinations in the treatment of specific diseases or symptoms, and provide valuable reference and guidance for acupuncture clinical practice.

[0062] On the basis of the above embodiment, further, constructing the acupoint feature synthesis map based on the acupoint compatibility information and the acupoint attribute information includes:

[0063] Based on the feature-aware connection fusion strategy, an acupoint compatibility map is constructed according to the acupoint compatibility information;

[0064] Specifically, the acupoint combination information includes acupoints and co-occurrence relationships between acupoints, and the acupoint combination graph is constructed with the acupoints as nodes and the co-occurrence relationships between acupoints as edges.

[0065] Based on the cosine function, constructing an acupoint attribute similarity matrix according to the acupoint attribute information;

[0066] Specifically, according to the formula:

[0067]

[0068] Calculate the similarity between any two acupoints, where s i,j is the similarity between any two acupoints i and j, p i is the attribute feature vector of acupoint i, p j is the attribute feature vector of acupoint j; the similarity between each acupoint and the other acupoints is taken as an element in turn to construct the acupoint attribute similarity matrix.

[0069] The acupoint feature synthesis map is constructed according to the acupoint compatibility map and the acupoint attribute similarity matrix.

[0070] Specifically, the co-occurrence matrix of the acupoint combination map is obtained, the link weights of the co-occurrence matrix and the similarity matrix are respectively obtained, the weights of the co-occurrence matrix and the similarity matrix are added together, and a specified preset threshold is used to construct an acupoint feature synthesis map.

[0071] On the basis of the above embodiment, further, constructing a denoised acupoint feature matrix based on the acupoint feature synthesis map based on a deep non-negative matrix decomposition method includes:

[0072] Based on the deep non-negative matrix factorization method, the adjacency matrix of the acupoint feature synthesis graph is decomposed at different network layers to obtain the factor matrix.

[0073] Specifically, the factor matrix includes a basis matrix and a coefficient matrix; based on the deep non-negative matrix decomposition method, the adjacency matrix of the acupoint feature synthesis graph is decomposed at the first preset network layer by minimizing the first preset objective function to obtain the basis matrix and the coefficient matrix; based on the deep non-negative matrix decomposition method, the coefficient matrix is ​​decomposed at the second preset network layer by minimizing the second preset objective function.

[0074] The factor matrix is ​​updated to obtain a correction factor matrix, and the matrix factors of each network layer are fine-tuned based on the correction factor matrix to obtain a denoised acupoint feature matrix.

[0075] Specifically, irrelevant variables are removed and a Lagrange multiplier matrix is ​​introduced to add non-negative constraints to the basis matrix and the coefficient matrix, so as to obtain a modified basis matrix and a modified coefficient matrix; according to the modified basis matrix and the modified coefficient matrix, the basis matrix or coefficient matrix corresponding to each layer of the network is fine-tuned by alternately minimizing the third preset objective function to obtain the denoised acupoint feature matrix.

[0076] On the basis of the above embodiment, further, the factor matrix includes a base matrix and a coefficient matrix; accordingly, the adjacency matrix of the acupoint feature synthesis graph is decomposed at different network layers based on the deep non-negative matrix decomposition method to obtain the factor matrix, including:

[0077] Based on the deep non-negative matrix factorization method, the adjacency matrix of the acupoint feature synthesis graph is decomposed in the first preset network layer by minimizing the first preset objective function to obtain a basis matrix and a coefficient matrix;

[0078] Specifically, the adjacency matrix C of the acupoint feature synthesis graph is pre-trained in each network layer using a deep non-negative matrix decomposition method to establish factor matrices: initial approximations of the base matrix U and the coefficient matrix W. The adjacency matrix of the acupoint feature synthesis graph is decomposed in the first preset network layer by minimizing the first preset objective function to obtain the base matrix and the coefficient matrix; wherein the first preset network layer is the first layer, and the first preset objective function is:

[0079]

[0080] Among them, C is the adjacency matrix of the acupoint feature synthesis graph, U1 is the basis matrix of the first layer, W1 is the coefficient matrix of the first layer, ||·|| F is the Frobenius norm.

[0081] Based on a deep non-negative matrix decomposition method, the coefficient matrix is ​​decomposed at a second preset network layer by minimizing a second preset objective function.

[0082] Specifically, the coefficient matrix is ​​obtained by decomposing the adjacency matrix of the acupoint feature synthesis graph by minimizing the second preset objective function in the second preset network layer, wherein the second preset network layer is the second layer, the coefficient matrix W1=U2W2, and the second preset objective function is:

[0083]

[0084] Among them, W1 is the coefficient matrix of the first layer, U2 is the basis matrix of the second layer, W2 is the coefficient matrix of the second layer, ||·|| F is the Frobenius norm.

[0085] On the basis of the above embodiment, further, the factor matrix is ​​updated to obtain a correction factor matrix, and the matrix factors of each network layer are fine-tuned based on the correction factor matrix to obtain a denoised acupoint feature matrix, including:

[0086] Adding non-negative constraints to the basis matrix and the coefficient matrix by removing irrelevant variables and introducing a Lagrange multiplier matrix to obtain a modified basis matrix and a modified coefficient matrix;

[0087] According to the modified base matrix and the modified coefficient matrix, the base matrix or coefficient matrix corresponding to each layer of the network is fine-tuned by alternately minimizing the third preset objective function to obtain the denoised acupoint feature matrix.

[0088] Specifically, the basis matrix U i For example: First, delete irrelevant variables and introduce the Lagrange multiplier matrix Θ to the basis matrix U i Applying non-negative constraints, we get the simplified third preset objective function:

[0089]

[0090] in, is the basis matrix U i The third preset objective function is, C is the adjacency matrix of the acupoint feature synthesis graph, Ψ i-1 =U1U2…U i-1 , Φ i+1 =U i+1 …U l-1 U l , U i (1≤i≤l) is the basis matrix; W l is the coefficient matrix of the lth layer, ||·|| F is the Frobenius norm, Θ is the Lagrange multiplier matrix; when i = 1, Ψ0 = I; when i = l, Φ l+1 =I; I is the degree matrix of the original matrix C. Based on matrix properties:

[0091] Tr(A T )=Tr(A),Tr(AB)=Tr(BA)

[0092] Among them, Tr(A T ) is the trace of the transpose of matrix A, Tr(A) is the trace of matrix A, Tr(AB) is the trace of the product of matrix A and matrix B, and Tr(BA) is the trace of the product of matrix B and matrix A.

[0093] The third preset objective function is converted into the form of a matrix trace and further rewritten as follows:

[0094]

[0095] Then, the partial derivative is set by Relative to U i The value of is 0:

[0096]

[0097] According to the Karush-Kuhn-Tucker Conditions: ΘU i =0

[0098] The following formula can be obtained:

[0099]

[0100] Then we get the following update formula:

[0101]

[0102] The base matrix is ​​updated according to the above updating formula to obtain a modified base matrix.

[0103] Similarly, based on a similar derivation process, the coefficient matrix W can be obtained i Update formula for (1≤i≤l):

[0104]

[0105] And the coefficient matrix W l The update formula is:

[0106]

[0107] The coefficient matrix may be updated according to the above updating formula to obtain a correction coefficient matrix.

[0108] Each layer matrix is ​​fine-tuned by alternately minimizing a third preset objective function, which is:

[0109]

[0110] Where λ and L represent the regularization parameter and coefficient matrix W respectively l The Laplace matrix of They are the basis matrices U l ...U2, U1 transpose, tr() represents the rank of the matrix, tr(W l LW l T ) is a regular term used to reduce overfitting and random noise in the decomposition process. According to the above formula, all variables in the third preset objective function are updated to obtain the denoising matrix The feature matrix of the acupoint feature synthesis map is subsequently used for robustness.

[0111] On the basis of the above embodiment, further, the denoised acupoint feature matrix is ​​trained through a residual graph attention network model to obtain a target acupoint feature matrix, including:

[0112] The denoised acupoint feature matrix is ​​taken as input, mapped to the feature space through a fully connected layer, and the attention weights between acupoints are dynamically learned using the attention mechanism of the graph attention network;

[0113] A multi-head attention mechanism is used to capture the features of acupoints and obtain the intermediate acupoint feature matrix;

[0114] The intermediate acupoint feature matrix is ​​updated through the residual structure to obtain the target acupoint feature matrix of the acupoint.

[0115] Specifically, the residual graph attention network includes a fully connected layer, a graph attention layer, a residual connection, and a nonlinear activation layer. The denoised acupoint feature matrix of the acupoint feature synthesis graph is used as input, mapped to the feature space through the fully connected layer, and then the attention weights between acupoints are dynamically learned using the attention mechanism of the graph attention network. Recorded as to represent the acupoint vector; the softmax function is used to dynamically balance the attention values ​​between nodes. The attention score formula of acupoints i and j is as follows:

[0116]

[0117] in, represents the attention score of the hth attention head, W is the learnable parameter matrix, is the set of neighbor nodes of acupoint i, and || is the connection operation of the matrix.

[0118] In order to ensure the stability of the self-attention mechanism, a multi-head attention mechanism is further used to capture the features of acupoints. The output feature formula of acupoint i is as follows:

[0119]

[0120] Among them, k i is the output feature of acupoint i, || is the connection operation of the matrix, H represents the number of multi-head attention mechanisms, σ is the nonlinear activation function, represents the attention score under the h-th attention head, is the set of neighbor nodes of acupoint i, W (h) represents the learnable weight matrix, c i is the feature vector of acupoint i in the denoised acupoint feature matrix.

[0121] Then, the residual structure is used to prevent noise from appearing during model propagation, and the output characteristic formula of acupoint i is updated as follows:

[0122]

[0123] Among them, k i is the output feature of acupoint i, σ is the nonlinear activation function, W r represents the linear weight matrix that ensures dimensional alignment between input and output features, is the feature vector of acupoint i in the denoised acupoint feature matrix, is the set of neighbor nodes of acupoint i, H represents the number of multi-head attention mechanisms, represents the attention score under the h-th attention head, W (h)The output feature of each acupoint is calculated according to the above updated formula as an element to form the target acupoint feature matrix of the acupoint.

[0124] On the basis of the above embodiment, further, the obtaining the correlation relationship of the acupoint combination to be predicted according to the target acupoint feature matrix includes:

[0125] According to the formula: Calculating the association relationship of obtaining the acupoint combination to be predicted, wherein: is the predicted association score between acupoints i and j, sigmoid is the activation function, is the transpose of the feature representation vector of acupoint i, k j is the feature representation vector of acupoint j.

[0126] Specifically, according to the formula: The correlation relationship of the acupoint combination to be predicted is calculated, and if the output result is 1, there is a correlation between the acupoints included in the acupoint combination to be predicted, and if the output result is 1, there is no correlation between the acupoints included in the acupoint combination to be predicted. It can be understood that the binary cross entropy loss function can also be used to train the model, and the loss function formula is as follows:

[0127]

[0128] Among them, y i，j is the true label. If acupoint i is associated with acupoint j, then y i，j The value is 1, otherwise it is 0. N represents the total number of entity pairs. During the training process, the Adam optimizer is used to minimize the loss to learn the model parameters.

[0129] In summary, the acupoint compatibility prediction method provided by the present application obtains the acupoint compatibility information and acupoint attribute information of the acupoint compatibility to be predicted, constructs an acupoint feature synthesis map based on the acupoint compatibility information and the acupoint attribute information, constructs a denoised acupoint feature matrix based on the acupoint feature synthesis map based on the deep non-negative matrix decomposition method, and uses the denoised acupoint feature matrix to train the residual graph attention network model to obtain the target acupoint feature matrix, obtains the correlation relationship of the acupoint compatibility to be predicted based on the target acupoint feature matrix, and uses the deep non-negative matrix decomposition method to decompose and fine-tune the adjacency matrix of the acupoint feature synthesis map layer by layer to construct the denoised acupoint feature matrix, which can effectively suppress the influence of noise and reduce redundant information. Interference with the prediction results; Residual learning is used to retain the key information in the acupoint feature synthesis graph structure, and the information loss is reduced during the multi-layer network transmission process, so that the network can more deeply explore the potential relationship between acupoints; The graph attention mechanism is used to automatically learn and identify which acupoints have a closer or more important relationship, so as to give higher weights in the prediction process, which helps to capture the nonlinear relationship and complex interaction between acupoints; The characteristic information of acupoints is gradually extracted through a multi-layer network structure, and finally the prediction results of the correlation relationship of the acupoints to be predicted are output. These prediction results can reflect the difference in the effects of different acupoint combinations in the treatment of specific diseases or symptoms, and provide valuable reference and guidance for acupuncture clinical practice.

[0130] Combination of the above Figure 1 The acupoint combination prediction method provided in the embodiment of the present application is introduced in detail. The acupoint combination prediction device, electronic device and computer-readable storage medium provided in the embodiment of the present application will be introduced in conjunction with the accompanying drawings.

[0131] like Figure 2 As shown in FIG. 1 , the figure is a schematic diagram of an acupoint combination prediction device provided by the present application, and the device comprises:

[0132] An acquisition module 201 is used to acquire acupoint combination information and acupoint attribute information of acupoint combination to be predicted;

[0133] A first construction module 202 is used to construct an acupoint feature synthesis map based on the acupoint compatibility information and the acupoint attribute information;

[0134] The second construction module 203 is used to construct a denoised acupoint feature matrix based on the acupoint feature synthesis map and a deep non-negative matrix decomposition method;

[0135] A training module 204 is used to obtain a target acupoint feature matrix by training the denoised acupoint feature matrix through a residual graph attention network model;

[0136] The prediction module 205 is used to obtain the correlation relationship of the acupoint combination to be predicted according to the target acupoint feature matrix.

[0137] The acupoint combination prediction device provided in the embodiment of the present application can correspond to executing the acupoint combination prediction method described in the embodiment of the present application, and the above functions of each module of the system correspond to implementing Figure 1 For the sake of brevity, the corresponding process of the method shown will not be repeated here.

[0138] An embodiment of the present application also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the acupoint combination prediction method as described in the above embodiment when executing the computer program.

[0139] like Figure 3 As shown, the computer system 300 of the electronic device includes a CPU 301, which can perform various appropriate actions and processes according to the program stored in the ROM 302 or the program loaded from the storage part 308 to the RAM 303. In the RAM 303, various programs and data required for system operation are also stored. The CPU 301, the ROM 302 and the RAM 403 are connected to each other through the bus 304. The I / O interface 305 is also connected to the bus 304. Among them, CPU 301 represents a central processing unit, ROM 302 represents a read-only memory, RAM 403 represents a random access memory, and I / O represents input / output.

[0140] The following components are connected to the I / O interface 305: an input section 306 including a keyboard, a mouse, etc.; an output section 307 including a cathode ray tube, a liquid crystal display, etc., and a speaker, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN card, a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. A removable medium 311, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 310 as needed, so that a computer program read therefrom is installed into the storage section 308 as needed.

[0141] In particular, the process of the reverse injection current signal delay suppression method described in the above embodiment can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program contains a program code for executing the reverse injection current signal delay suppression method described in the above embodiment. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 309, and / or installed from the removable medium 311. When the computer program is executed by the CPU 301, the above-mentioned functions defined in the present computer system 300 are executed.

[0142] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for suppressing the delay of a reverse injection current signal as described in the above embodiment is implemented.

[0143] Specifically, the computer-readable storage medium may be included in the electronic device described in the above embodiment; or it may exist independently without being assembled into the electronic device. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the reverse injection current signal delay suppression method as described in the above embodiment.

[0144] It should be noted that the computer-readable storage medium shown in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical fiber, a portable compact disk read-only memory, an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above.

[0145] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and its core ideas of this application. The above is only the preferred implementation method of this application. It should be pointed out that due to the limitations of textual expression and the objective existence of infinite specific structures, ordinary technicians in this technical field can make several improvements, modifications or changes without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes or combinations, or the direct application of the inventive concept and technical solution to other occasions without improvement, should be regarded as the scope of protection of this application.

Claims

1. A method for predicting acupoint combinations, characterized in that: include: Obtaining acupoint combination information and acupoint attribute information of the acupoint combination to be predicted; Constructing an acupoint characteristic synthesis map based on the acupoint combination information and the acupoint attribute information; According to the acupoint feature synthesis graph, a denoised acupoint feature matrix is ​​constructed based on a deep non-negative matrix decomposition method; The denoised acupoint feature matrix is ​​trained through a residual graph attention network model to obtain a target acupoint feature matrix; The correlation relationship of the acupoint combination to be predicted is obtained according to the target acupoint feature matrix.

2. The method according to claim 1, characterized in that The constructing of acupoint feature synthesis graph based on the acupoint combination information and the acupoint attribute information includes: Based on the feature-aware connection fusion strategy, an acupoint compatibility map is constructed according to the acupoint compatibility information; Based on the cosine function, constructing an acupoint attribute similarity matrix according to the acupoint attribute information; The acupoint feature synthesis map is constructed according to the acupoint compatibility map and the acupoint attribute similarity matrix.

3. The method according to claim 1, characterized in that The step of constructing a denoised acupoint feature matrix based on the acupoint feature synthesis graph and a deep non-negative matrix decomposition method comprises: Based on the deep non-negative matrix factorization method, the adjacency matrix of the acupoint feature synthesis graph is decomposed at different network layers to obtain the factor matrix. The factor matrix is ​​updated to obtain a correction factor matrix, and the matrix factors of each network layer are fine-tuned based on the correction factor matrix to obtain a denoised acupoint feature matrix.

4. The method according to claim 3, characterized in that The factor matrix includes a base matrix and a coefficient matrix; accordingly, the method based on deep non-negative matrix decomposition decomposes the adjacency matrix of the acupoint feature synthesis graph at different network layers to obtain the factor matrix, including: Based on the deep non-negative matrix factorization method, the adjacency matrix of the acupoint feature synthesis graph is decomposed in the first preset network layer by minimizing the first preset objective function to obtain a basis matrix and a coefficient matrix; Based on a deep non-negative matrix decomposition method, the coefficient matrix is ​​decomposed at a second preset network layer by minimizing a second preset objective function.

5. The method according to claim 4, characterized in that The updating of the factor matrix to obtain a correction factor matrix, and fine-tuning the matrix factors of each network layer based on the correction factor matrix to obtain a denoised acupoint feature matrix, includes: Adding non-negative constraints to the basis matrix and the coefficient matrix by removing irrelevant variables and introducing a Lagrange multiplier matrix to obtain a modified basis matrix and a modified coefficient matrix; According to the modified base matrix and the modified coefficient matrix, the base matrix or coefficient matrix corresponding to each layer of the network is fine-tuned by alternately minimizing the third preset objective function to obtain the denoised acupoint feature matrix.

6. The method according to claim 1, characterized in that The denoised acupoint feature matrix is ​​trained through a residual graph attention network model to obtain a target acupoint feature matrix, including: The denoised acupoint feature matrix is ​​taken as input, mapped to the feature space through a fully connected layer, and the attention weights between acupoints are dynamically learned using the attention mechanism of the graph attention network; A multi-head attention mechanism is used to capture the features of acupoints and obtain the intermediate acupoint feature matrix; The intermediate acupoint feature matrix is ​​updated through the residual structure to obtain the target acupoint feature matrix of the acupoint.

7. The method according to claim 6, characterized in that The step of obtaining the correlation relationship of the acupoint combinations to be predicted according to the target acupoint feature matrix includes: According to the formula: Calculating the association relationship of obtaining the acupoint combination to be predicted, wherein: is the predicted association score between acupoints i and j, sigmoid is the activation function, is the transpose of the feature representation vector of acupoint i, k j is the feature representation vector of acupoint j.

8. An acupoint combination prediction device, characterized in that: include: An acquisition module is used to acquire acupoint combination information and acupoint attribute information of acupoint combination to be predicted; A first construction module is used to construct an acupoint feature synthesis map based on the acupoint compatibility information and the acupoint attribute information; A second construction module is used to construct a denoised acupoint feature matrix based on the acupoint feature synthesis map and a deep non-negative matrix decomposition method; A training module, used for training the denoised acupoint feature matrix through a residual graph attention network model to obtain a target acupoint feature matrix; The prediction module is used to obtain the correlation relationship of the acupoint combination to be predicted according to the target acupoint feature matrix.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images