Intelligent matching method and device based on convolutional network, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses an intelligent matching method and device based on a convolutional network, equipment and a medium, and the method comprises the steps: determining a final feature vector of a patient node; splicing the feature vector of the static attribute information, the dynamic feature vector and the knowledge feature vector to obtain a self feature vector of the doctor node; the feature vectors of the doctor nodes and the cooperative feature vectors of the doctor nodes are fused to obtain final feature vectors of the doctor nodes, and the final feature vectors of the patient nodes and feature vectors of neighbor nodes of the patient nodes are fused to obtain comprehensive feature vectors of the patient nodes; fusing the final feature vector of the doctor node and the feature vector of the neighbor node of the doctor node to obtain a comprehensive feature vector of the doctor node; and when the matching score meets a preset condition, adding the doctor node into a recommendation list. The matching efficiency of the patient node and the doctor node can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to an intelligent matching method and device based on a convolution network, an intelligent matching equipment and medium. BACKGROUND

[0002] In a medical digital scenario, a patient node and a doctor node are different virtual units. Matching the patient node and the doctor node is a core link for improving the efficiency of medical services.

[0003] However, the existing matching method only relies on static historical data. Since the static historical data can only reflect the status of the patient node and the doctor node at a certain time in the past, the information of the patient node and the information of the doctor node in the medical digital scenario are always in dynamic change. The information of the patient node is continuously updated, and the information of the doctor node also changes over time. Since the existing matching method fails to fully consider these dynamic change factors and only matches based on static historical data, the existing matching method cannot effectively capture these changes, which may result in inaccurate matching scores of the patient node and the doctor node, and inaccurate recommendation lists determined based on the matching scores. Therefore, how to match the patient node and the doctor node is a problem that needs to be solved. SUMMARY

[0004] The embodiments of the present application provide an intelligent matching method and device based on a convolution network, and an intelligent matching equipment and medium, to solve the technical problem of how to match the patient node and the doctor node.

[0005] In a first aspect, the embodiments of the present application provide an intelligent matching method based on a convolution network, applied to an electronic device, and the intelligent matching method comprises: obtaining basic information, illness records, and medical records of a patient node, determining an ultimate feature vector of the patient node based on the basic information, the illness records, and the medical records of the patient node; obtaining static attribute information, dynamic attribute information, and knowledge information of a doctor node, processing the dynamic attribute information and the knowledge information based on a predefined manner to obtain a dynamic feature vector and a knowledge feature vector, and splicing the feature vector of the static attribute information, the dynamic feature vector, and the knowledge feature vector to obtain an own feature vector of the doctor node; obtaining a collaboration feature vector of the doctor node from a collaboration relationship graph, and fusing the own feature vector of the doctor node and the collaboration feature vector of the doctor node to obtain an ultimate feature vector of the doctor node; The final feature vector of the patient node and the feature vectors of the neighbor nodes of the patient node are fused through a gated graph convolution network to obtain a comprehensive feature vector of the patient node, and the final feature vector of the doctor node and the feature vectors of the neighbor nodes of the doctor node are fused to obtain a comprehensive feature vector of the doctor node; The time trajectory attention between the patient comprehensive feature vector and the doctor comprehensive feature vector is calculated, and the semantic association attention between the patient comprehensive feature vector and the doctor comprehensive feature vector is calculated. The time trajectory attention and the semantic association attention are processed through the trained full connection layer to obtain a matching score of the patient node and the doctor node, and when the matching score meets a preset condition, the doctor node is added to a recommendation list.

[0006] In a possible implementation manner of the first aspect, the basic information of the patient node, the illness record and the medical record are obtained, the basic information is feature-encoded to obtain a basic feature vector of the patient node, the illness record is feature-encoded to obtain a first feature vector, and the medical record is feature-encoded to obtain a second feature vector; The first feature vector and the second feature vector are spliced to obtain a third feature vector, the third feature vector is input into an LSTM unit, the third feature vector is processed through the LSTM unit to obtain a fourth feature vector, the fourth feature vector is processed using a Transformer model to obtain a fifth feature vector, the fourth feature vector and the fifth feature vector are spliced to obtain a sixth feature vector, and the sixth feature vector and the basic feature vector of the patient node are spliced to obtain a final feature vector of the patient node.

[0007] In a possible implementation manner of the first aspect, the static attribute information, the dynamic attribute information and the knowledge information of the doctor node are obtained, the dynamic attribute information includes the number of cases treated at past time points and patient evaluation data, and the knowledge information includes rare disease consultation data, rare disease paper data and rare disease clinical trial data. The current case number of the doctor node is obtained through a time decay model, a decay coefficient and the number of cases treated at past time points, each evaluation text in the patient evaluation data is obtained, each evaluation text is processed using a pre-trained sentiment analysis model to obtain a sentiment score of each evaluation text, a feature vector of the patient evaluation data is obtained by using a multi-head attention mechanism to fuse the weighted vectors of each evaluation text through a weighting model, the sentiment score of each evaluation text, the feature vector of each evaluation text and a time decay weight, the feature vector of the current case number and the feature vector of the patient evaluation data are spliced to obtain a combined feature vector, the combined feature vector is dimension-reduced to obtain a dimension-reduced combined feature vector, and the dimension-reduced combined feature vector is selected as the dynamic feature vector of the doctor node. access the medical knowledge graph, obtain the feature vectors of the first-order neighbors and the second-order neighbors of the doctor node from the medical knowledge graph, aggregate the feature vectors of the first-order neighbors and the second-order neighbors of the doctor node by using a graph neural network algorithm, and obtain the knowledge feature vector of the doctor node; The feature vector of the static attribute information, the dynamic feature vector, and the knowledge feature vector are spliced to obtain the self feature vector of the doctor node.

[0008] In a possible implementation manner of the first aspect, the time trajectory attention between the patient comprehensive feature vector and the doctor comprehensive feature vector is calculated by using a time trajectory attention model, and the semantic association attention between the patient comprehensive feature vector and the doctor comprehensive feature vector is calculated by using a semantic association attention model.

[0009] In a possible implementation manner of the first aspect, the time trajectory attention and the semantic association attention are processed by using the trained full connection layer to obtain the matching score of the patient node and the doctor node. The dynamic threshold is obtained by using a dynamic threshold function, and when the matching score of the patient node and the doctor node is greater than the dynamic threshold, the doctor node is added to the recommendation list.

[0010] In a possible implementation manner of the first aspect, a sample set is obtained, a plurality of training samples in the sample set are used to train the full connection layer, in the training process, the total loss value of the full connection layer is calculated by using a target loss function, when the total loss value is less than a preset loss value, the training of the full connection layer is stopped, and the trained full connection layer is saved.

[0011] In a possible implementation manner of the first aspect, the static attribute information includes education background information, position information, and title information of the doctor, the dynamic attribute information includes the number of consultation cases at a past time point and patient evaluation data, and the knowledge information includes rare disease consultation data, rare disease paper data, and rare disease clinical trial data.

[0012] In a second aspect, an embodiment of the present application provides an intelligent matching device based on a convolutional network, applied to an electronic device, and including: The first obtaining module is configured to obtain basic information, illness records, and treatment records of a patient node, and determine a final feature vector of the patient node based on the basic information, the illness records, and the treatment records of the patient node. The second obtaining module is configured to obtain static attribute information, dynamic attribute information, and knowledge information of a doctor node, process the dynamic attribute information and the knowledge information based on a predefined manner to obtain a dynamic feature vector and a knowledge feature vector, and splice the feature vector of the static attribute information, the dynamic feature vector, and the knowledge feature vector to obtain a self feature vector of the doctor node. The third acquisition module is configured to acquire a collaboration feature vector of the doctor node from the collaboration graph, and fuse the self feature vector of the doctor node and the collaboration feature vector of the doctor node to obtain a final feature vector of the doctor node. The first fusion module is configured to fuse the final feature vector of the patient node and the feature vectors of the neighbor nodes of the patient node by using a gated graph convolution network to obtain a comprehensive feature vector of the patient node, and fuse the final feature vector of the doctor node and the feature vectors of the neighbor nodes of the doctor node to obtain a comprehensive feature vector of the doctor node. The second fusion module is configured to calculate time trajectory attention between the patient comprehensive feature vector and the doctor comprehensive feature vector, and calculate semantic association attention between the patient comprehensive feature vector and the doctor comprehensive feature vector. The matching module is configured to process the time trajectory attention and the semantic association attention by using the trained full connection layer to obtain a matching score of the patient node and the doctor node, and add the doctor node to a recommendation list when the matching score meets a preset condition.

[0013] In a third aspect, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the intelligent matching method of the first aspect when executing the computer program.

[0014] In a fourth aspect, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the intelligent matching method of the first aspect.

[0015] In a fifth aspect, a computer program product is provided, and when the computer program product is executed on an electronic device, the electronic device executes the intelligent matching method of the first aspect.

[0016] The embodiments of the present application have two advantages. On the one hand, the target feature vector is processed by the trained full connection layer to obtain a matching score of the patient node and the doctor node, and the recommendation information is sent to the client corresponding to the patient node when the matching score meets a preset condition, thereby solving the technical problem of how to match the patient node and the doctor node. Since automatic matching is used, there is no need to manually input a matching instruction, and therefore the matching time of the patient node and the doctor node is reduced, and the matching efficiency of the patient node and the doctor node is improved. On the other hand, the recommendation list is sent to the client corresponding to the patient node, so that the recommendation list can be directly sent to the client, the circulation time of the recommendation list is greatly shortened, and the client can obtain the recommendation list meeting the own demand more timely. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments description. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0018] Figure 1 The application scenario diagram of the intelligent matching method provided by the embodiments of the present application is shown in the following figure. Figure 2 The flowchart of the intelligent matching method provided by the embodiments of the present application is shown in the following figure. Figure 3 The flowchart of S201 provided by the embodiments of the present application is shown in the following figure. Figure 4 The schematic block diagram of the intelligent matching device provided by the embodiments of the present application is shown in the following figure. Figure 5 The structural schematic diagram of the electronic device provided by the embodiments of the present application is shown in the following figure. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present application more clear, the following will further describe the present application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] In the following description, specific details such as specific system structures, techniques, etc. are presented in order to thoroughly understand the embodiments of the present application, but those skilled in the art should understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits and methods are omitted to avoid unnecessary details that hinder the description of the present application.

[0021] It should be understood that when used in the specification and the appended claims of the present application, the term "comprising" indicates the presence of the described features, whole, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, whole, steps, operations, elements, components and / or sets thereof.

[0022] It should also be understood that the term "and / or" used in the specification and the appended claims of the present application means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0023] As used in the specification and the appended claims, the term "if' can be interpreted as meaning "when" or "upon" or "in response to determining" or "in response to detecting" depending on the context. Similarly, the phrase "if it is determined" or "if [the described condition or event] is detected" can be interpreted to mean "upon determining" or "in response to determining" or "upon detecting [the described condition or event]" or "in response to detecting [the described condition or event]" depending on the context.

[0024] In addition, the terms "first", "second", "third", etc. as used in the description of the application and the appended claims are only used to distinguish descriptions, and cannot be understood as indicating or implying relative importance.

[0025] The description of "one embodiment" or "some embodiments" or the like in the present application specification means that the specific features, structures or characteristics described in connection with the embodiment are included in one or more embodiments of the present application. Therefore, the statements "in one embodiment", "in some embodiments", "in other some embodiments", "in further some embodiments" and the like appearing in different places in the specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized. The terms "include", "contain", "have" and their variants mean "including but not limited to", unless otherwise specifically emphasized.

[0026] In addition, the technical solutions of various embodiments can be combined with each other, but it must be based on the fact that a person skilled in the art can realize it, and when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist and is not within the protection scope required by the present application.

[0027] The flowchart shown in the accompanying drawings is only an example and does not necessarily include all contents and operations / steps, nor does it necessarily execute in the order described. For example, some operations / steps can be further divided, combined or partially merged, so the actual execution order may be changed according to the actual situation.

[0028] The intelligent matching method provided in the embodiments of the present application can be applied to electronic devices such as mobile phones, tablet computers, wearable devices, vehicle-mounted devices, augmented reality (AR) / virtual reality (VR) devices, notebook computers, ultra-mobile personal computers (UMPCs), netbooks, personal digital assistants (PDAs), and the like. The embodiments of the present application do not limit the specific type of electronic device.

[0029] Referring to Figure 1 , Figure 1 The application scenario of the intelligent matching method provided in the embodiments of the present application is described as follows: The electronic device accesses the hospital management system and obtains the basic information, medical records, and medical records of the patient node from the hospital management system.

[0030] In the embodiments of the present application, the electronic device can quickly obtain the basic information, medical records, and medical records of the patient node by accessing the database.

[0031] Referring to Figure 2 , Figure 2 The flowchart of the intelligent matching method provided in the embodiments of the present application is shown in the figure, and the method can be applied to an electronic device.

[0032] As Figure 2 shown, the intelligent matching method provided in the embodiments of the present application includes the following steps, which are described in detail as follows: S201, obtaining the basic information, medical records, and medical records of the patient node, and determining the final feature vector of the patient node based on the basic information, medical records, and medical records of the patient node; The patient node refers to a virtual unit constructed with a patient as the core, and the essence of the patient node is a digital carrier of the patient's medical information.

[0033] The doctor node refers to a virtual business unit constructed with a medical professional as the core, and the essence of the doctor node is a digital carrier of the doctor's professional ability.

[0034] The basic information includes age, gender, and genetic history, providing a fundamental basis for understanding the patient's physiological characteristics and potential risks. The medical record details the disease type, symptoms, and severity, directly reflecting the patient's current health status. The consultation history includes past treatment records, medication responses, and treatment effects, demonstrating the disease progression and intervention outcomes. Based on the patient node's basic information, medical record, and consultation history, the final feature vector of the patient node is determined, comprehensively and accurately characterizing individual patient traits and providing solid data support for medical decision-making.

[0035] The intelligent matching method, prior to acquiring the patient node's basic information, medical records, and consultation records, and determining the final feature vector of the patient node based on these records, includes: Obtain a sample set and train the fully connected layer using multiple training samples from the sample set. During the training process, calculate the total loss value of the fully connected layer using the target loss function. When the total loss value is less than the preset loss value, stop training the fully connected layer and save the trained fully connected layer.

[0036] The objective loss function is: ; ; ; This is the total loss value. This is the mean squared error loss value. It is the divergence loss value; It is an adjustment factor. It is the fully connected layer in the 1st... Predicted scores on each training sample; It is the first The actual scores corresponding to each training sample; For the fully connected layer in the 1st Predicted probability distribution on training samples; It is the first The true probability distribution corresponding to each training sample; It represents the natural logarithm or the logarithm to the base 2.

[0037] In S202, static attribute information, dynamic attribute information, and knowledge information of the doctor node are obtained, the dynamic attribute information and the knowledge information are processed based on a predefined manner to obtain a dynamic feature vector and a knowledge feature vector, a feature vector of the static attribute information, the dynamic feature vector, and the knowledge feature vector are spliced to obtain a self feature vector of the doctor node; wherein the obtaining of the static attribute information, the dynamic attribute information, and the knowledge information of the doctor node, the processing of the dynamic attribute information and the knowledge information based on the predefined manner to obtain the dynamic feature vector and the knowledge feature vector, and the splicing of the feature vector of the static attribute information, the dynamic feature vector, and the knowledge feature vector to obtain the self feature vector of the doctor node comprises: The static attribute information, the dynamic attribute information, and the knowledge information of the doctor node are obtained, the dynamic attribute information comprises the number of cases treated at a past time point and patient evaluation data, and the knowledge information comprises rare disease consultation data, rare disease paper data, and rare disease clinical trial data. The rare disease consultation data is consultation data of a rare disease, the rare disease paper data is paper data of a rare disease, and the rare disease clinical trial data is clinical trial data of a rare disease.

[0038] The current number of cases of the doctor node is obtained through a time decay model, a decay coefficient, and the number of cases treated at the past time point, each piece of evaluation text in the patient evaluation data is obtained, a pre-trained sentiment analysis model is used to process each piece of evaluation text to obtain a sentiment score of each piece of evaluation text, a weighted model, the sentiment score of each piece of evaluation text, a feature vector of each piece of evaluation text, and a time decay weight are used to fuse the weighted vectors of each piece of evaluation text using a multi-head attention mechanism to obtain a feature vector of the patient evaluation data, the feature vector of the current number of cases and the feature vector of the patient evaluation data are spliced to obtain a combined feature vector, the combined feature vector is dimensionally reduced to obtain a reduced combined feature vector, and the reduced combined feature vector is selected as the dynamic feature vector of the doctor node. The time decay model is as follows: ; The current number of cases of the doctor node is represented by N t. The decay coefficient is represented by a. The current time point is represented by t. The past time point is represented by t i. The number of cases treated at the past time point is represented by N i.

[0039] Optionally, .

[0040] The greater the interval between the past time point and the current time point is, the The value will decrease rapidly, meaning that the number of cases treated in the past has less impact on the current state.

[0041] The smaller the value, the smaller the interval between the past and the present time. The larger the value, the greater the impact of the number of cases treated in the past on the current state.

[0042] This allows for the quantification of the impact of the number of cases received at different points in time on the current number of cases.

[0043] The weighted model is as follows: ; ; It is a weighted vector for each evaluation text. It is the sentiment score for each evaluation text. To evaluate time weighting, For each evaluation text, a feature vector is generated. This indicates the time difference between the current time and the publication time of each review text. It is the decay weight. Optionally, , Access the medical knowledge graph, obtain the feature vectors of the first-order and second-order neighbors of the doctor node from the medical knowledge graph, and use the graph neural network algorithm to aggregate the feature vectors of the first-order and second-order neighbors of the doctor node to obtain the knowledge feature vector of the doctor node. The feature vectors of static attribute information, dynamic feature vectors, and knowledge feature vectors are concatenated to obtain the doctor node's own feature vector.

[0044] The medical knowledge graph clearly distinguishes between node types and edge types. Node types cover key entities in the medical field, including doctors, diseases, treatment plans, academic papers, and clinical trials. Edge types clearly express the important relationships between doctors and academic papers. This basic structure can intuitively present some core relationships in medical knowledge, providing a basic framework for subsequent knowledge storage, retrieval, and analysis.

[0045] For ease of explanation, the following example is provided: Node type: Doctor: An individual doctor.

[0046] Disease: Rare disease entity, such as pulmonary hypertension, ICD-10 code.

[0047] Treatment options: Clinical treatment methods, such as targeted drugs.

[0048] Academic papers: papers published by doctors, MeSH term coding).

[0049] Clinical trials: trials in which doctors participate.

[0050] Edge type: Doctor - Specializes in - Disease; Doctor - Publishes - Academic paper; Doctor - Participates in - Clinical trial; Academic paper - Mentions - Disease; Clinical trial - Targets - Disease.

[0051] GraphSAGE aggregation: aggregating first-order neighbors and second-order neighbors of the doctor node, the first-order neighbors being the diseases specialized in, the papers published, and the second-order neighbors being the features of the diseases mentioned in the papers, so as to organically integrate these scattered but related information. This aggregation method not only enriches the feature representation of the doctor node, but also captures the information of the doctor node in multiple aspects such as professional skill, academic research and knowledge extension, thereby constructing a more complete and accurate feature vector for the doctor node in the medical knowledge graph and providing a solid foundation for medical recommendation.

[0052] Static attribute information includes the educational background information, position information and title information of the doctor, and the static attribute information is the basic information of the professional ability of the doctor, which can directly reflect the professional ability of the doctor; Dynamic attribute information includes the number of cases treated at past time points and patient evaluation data, which reflect the work ability of the doctor; Knowledge information includes rare disease consultation data, rare disease paper data and rare disease clinical trial data, which reflect the learning ability of the doctor.

[0053] The feature vector of the static attribute information, the dynamic feature vector and the knowledge feature vector are spliced to obtain the self-feature vector of the doctor node, which can comprehensively and stereoscopically present the professional ability, work ability and learning ability of the doctor corresponding to the doctor node.

[0054] S203, obtaining the collaboration feature vector of the doctor node from the collaboration relationship graph, and fusing the self-feature vector of the doctor node and the collaboration feature vector of the doctor node to obtain the final feature vector of the doctor node; The fusion of the self-feature vector of the doctor node and the collaboration feature vector of the doctor node to obtain the final feature vector of the doctor node includes: generate a gating value according to the gating formula, the self feature vector of the doctor node, and the collaboration feature of the doctor node, and generate a final feature vector of the doctor node by weighting and summing the self feature vector of the doctor node and the collaboration feature vector of the doctor node through a fusion formula; wherein the gating formula is: ; denotes the gating value, denotes a Sigmoid function; denotes a bias term for adjusting the output of the gating mechanism model; denotes the self feature vector of the doctor node; denotes the collaboration feature vector of the doctor node; denotes a feature vector formed by splicing the self feature vector of the doctor node and the collaboration feature vector of the doctor node.

[0055] wherein the fusion formula is: ; denotes the final feature vector of the doctor node; denotes the gating value; denotes the self feature vector of the doctor node; the symbol denotes element-wise multiplication; denotes the collaboration feature vector of the doctor node.

[0056] The self feature vector of the doctor node reflects the self ability of the doctor, and the collaboration feature vector of the doctor node reflects the collaboration ability of the doctor in team collaboration. Through the fusion formula and the gating value, the self feature vector of the doctor node and the collaboration feature vector of the doctor node can be dynamically fused to obtain the final feature vector of the doctor node, which retains the information of the self ability of the doctor and the information of the collaboration ability of the doctor. S204, by the gating graph convolution network, the final feature vector of the patient node and the feature vector of the neighbor node of the patient node are fused to obtain a comprehensive feature vector of the patient node, and the final feature vector of the doctor node and the feature vector of the neighbor node of the doctor node are fused to obtain a comprehensive feature vector of the doctor node. The gating graph convolution network is a neural network model combining a graph neural network and a gating mechanism.

[0057] In the medical knowledge graph, the patient node and the doctor node are not isolated, the neighbor node of the patient node refers to other nodes having a direct association relationship with the patient node. The neighbor node of the doctor node refers to other nodes having a direct association relationship with the doctor node.

[0058] The final feature vector of the patient node and the feature vector of the neighbor node of the patient node are fused to obtain a comprehensive feature vector of the patient node, and a comprehensive and accurate patient portrait can be constructed.

[0059] The final feature vector of the doctor node and the feature vector of the neighbor node of the doctor node are fused to construct a comprehensive and accurate doctor portrait.

[0060] S205, calculating the time trajectory attention between the patient comprehensive feature vector and the doctor comprehensive feature vector, and calculating the semantic association attention between the patient comprehensive feature vector and the doctor comprehensive feature vector; The calculation of the time trajectory attention between the patient comprehensive feature vector and the doctor comprehensive feature vector, and the calculation of the semantic association attention between the patient comprehensive feature vector and the doctor comprehensive feature vector, include: The time trajectory attention between the patient comprehensive feature vector and the doctor comprehensive feature vector is calculated by a time trajectory attention model, and the semantic association attention between the patient comprehensive feature vector and the doctor comprehensive feature vector is calculated by a semantic association attention model.

[0061] The time trajectory attention model is: ; The time trajectory attention is: The patient comprehensive feature vector is: The doctor comprehensive feature vector is: The softmax function represents a function for converting a vector into a probability distribution.

[0062] The semantic association attention model is: ; The semantic association attention between the patient comprehensive feature vector and the doctor comprehensive feature vector is: The combination operator symbol is: The patient comprehensive feature vector is: The doctor comprehensive feature vector is: The one-dimensional convolution calculation is performed on the patient comprehensive feature vector and the doctor comprehensive feature vector. ​ representing a one-dimensional convolution calculation.

[0063] The timing attention is calculated by a scaled dot-product, and the semantic attention is calculated by a one-dimensional convolution.

[0064] The timing attention is calculated by a scaled dot-product, which can efficiently capture the global correlation between different time steps in the timing data. The similarity of the features at each time step is quickly measured by the scaled dot-product calculation, and the numerical range is stabilized by the scaling factor, so that the dependence relationship of the distant time steps is clearly preserved in the long sequence, avoiding the information decay caused by the lengthening of the distance. By capturing the correlation between the patient node and the doctor node in the time dimension through the scaled dot-product attention, it is suitable for long-term record information of the patient node. When processing the long-term record information of the patient node and the data of the doctor node, the matching efficiency is crucial. The scaled dot-product attention mechanism can quickly filter out potential matching doctor nodes by calculating the similarity of the features at different time steps in parallel, greatly reducing the computational amount and time cost in the matching process.

[0065] The semantic attention is calculated by a one-dimensional convolution, which focuses on the extraction of semantic patterns within a local timing segment. When the convolution kernel slides on the time axis, it can automatically focus on the local combination features of adjacent time steps, capture the coherence changes of short-term semantics using local receptive fields, and strengthen the perception ability of dynamic semantic details. In the process of matching the patient node and the doctor node, the one-dimensional convolution can focus on the current symptom information of the patient node and the doctor expertise information of the doctor node. In this way, the system can deeply mine the subtle association between the current symptom information and the doctor expertise information, thereby enhancing the fine-grained matching of the current symptom information of the patient node and the doctor expertise information, and improving the matching effect.

[0066] 206, the matching score of the patient node and the doctor node is obtained by processing the timing trajectory attention and the semantic association attention through the trained full connection layer, and when the matching score meets the preset condition, the doctor node is added to the recommendation list.

[0067] The matching score of the patient node and the doctor node is obtained by processing the timing trajectory attention and the semantic association attention through the trained full connection layer, and when the matching score meets the preset condition, the doctor node is added to the recommendation list. The matching score of the patient node and the doctor node is obtained by processing the timing trajectory attention and the semantic association attention through the trained full connection layer, and when the matching score meets the preset condition, the doctor node is added to the recommendation list. ​The dynamic threshold is obtained by a dynamic threshold function, and when the matching score of the patient node and the doctor node is greater than the dynamic threshold, the doctor node is added to the recommendation list. Wherein, the matching score of the patient node and the doctor node is obtained by processing the time trajectory attention and the semantic association attention through the trained full connection layer, including: The matching score of the patient node and the doctor node is obtained by processing the time trajectory attention and the semantic association attention through the trained full connection layer and the matching score model; Wherein, the matching score model is: ; is the matching score of the patient node and the doctor node; Sigmoid is an activation function; FC represents the trained full connection layer; is the time trajectory attention; is the patient comprehensive feature vector; is the semantic association attention between the patient comprehensive feature vector and the doctor comprehensive feature vector; is the doctor comprehensive feature vector. The matching score of the patient node and the doctor node is used to describe the matching degree between the patient node and the doctor node; the higher the matching score of the patient node and the doctor node, the more suitable the doctor node is for the patient node; the lower the matching score of the patient node and the doctor node, the less suitable the doctor node is for the patient node.

[0068] Wherein, the dynamic threshold function is: ; is the dynamic threshold, is the initial threshold, is the adjustment coefficient, is the time decay factor, which ensures that the threshold increases linearly with time; is the number of days. Preferably, , , .

[0069] For ease of illustration, the following examples are given: For example, when t=0, it means that after 0 days, the doctor nodes with a matching score greater than the dynamic threshold are doctor node B, doctor node C and doctor node D; t=30, it means that after 30 days, the doctor nodes with a matching score greater than the dynamic threshold are doctor node B and doctor node D; t=90, it means that after 90 days, the only doctor node with a matching score greater than the dynamic threshold is doctor node B. It reflects that the higher the requirement for doctors is with the increase of time.

[0070] The dynamic threshold is obtained through a dynamic threshold function, and when the matching score of the patient node and the doctor node is greater than the dynamic threshold, it indicates that the doctor node meets the requirements of the patient node, and the doctor node is added to the recommendation list.

[0071] The recommendation list is sent to the client corresponding to the patient node. The recommendation information is directly presented on the client corresponding to the patient node, and the patient can quickly lock the target doctor, saving a lot of time and effort.

[0072] The embodiments of the present application have two advantages. On the one hand, the matching score of the patient node and the doctor node is obtained by processing the time track attention and the semantic correlation attention through the trained full connection layer, and when the matching score meets the preset condition, the doctor node is added to the recommendation list, solving the technical problem of how to match the patient node and the doctor node. Since automatic matching is used, there is no need to manually input matching instructions, thus reducing the matching time of the patient node and the doctor node, and improving the matching efficiency of the patient node and the doctor node. On the other hand, the recommendation list is sent to the client corresponding to the patient node, so that the recommendation list can be directly sent to the client, greatly shortening the circulation time of the recommendation list, and enabling the client to obtain the recommendation list meeting its own needs more timely.

[0073] Please refer to Figure 3 , Figure 3 The flowchart of S201 provided by the embodiments of the present application is described as follows: S301, the basic information, the illness record and the medical record of the patient node are obtained, the basic information is feature encoded to obtain the basic feature vector of the patient node, the illness record is feature encoded to obtain the first feature vector, and the medical record is feature encoded to obtain the second feature vector; S302, the first feature vector and the second feature vector are spliced to obtain a third feature vector, the third feature vector is input into an LSTM unit, the third feature vector is processed through the LSTM unit to obtain a fourth feature vector, the fourth feature vector is processed using a Transformer model to obtain a fifth feature vector, the fourth feature vector and the fifth feature vector are spliced to obtain a sixth feature vector, and the sixth feature vector and the basic feature vector of the patient node are spliced to obtain the final feature vector of the patient node.

[0074] The Chinese name of LSTM is Long Short-Term Memory, and the English full name of LSTM is Long Short-Term Memory. LSTM unit is a special recurrent neural network specially designed to solve the gradient disappearance or explosion problem of traditional RNN when processing long sequence data.

[0075] In this embodiment, the final feature vector of the patient node integrates the static attribute information and dynamic change information of the patient node. Therefore, the final feature vector of the patient node is a more comprehensive and richer feature representation, enabling the fully connected layer to capture more subtle patterns and relationships in the final feature vector of the patient node.

[0076] For the intelligent matching method described in the above embodiments, please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic block diagram of the intelligent matching device provided in the embodiments of this application. Figure 4 The intelligent matching device 400 shown can be applied to, for example... Figure 1 The application scenario diagram shows electronic devices. The following section uses electronic devices as an example to illustrate this. Figure 4 The intelligent matching device 400 shown will be described in detail. The intelligent matching device 400 may include a first acquisition module 401, a second acquisition module 402, a third acquisition module 403, a first fusion module 404, a second fusion module 405, and a matching module 406.

[0077] The first acquisition module 401 is used to acquire the basic information, medical records, and medical records of the patient node, and to determine the final feature vector of the patient node based on the basic information, medical records, and medical records of the patient node. The second acquisition module 402 is used to acquire the static attribute information, dynamic attribute information, and knowledge information of the doctor node. It processes the dynamic attribute information and knowledge information in a predefined manner to obtain dynamic feature vectors and knowledge feature vectors. It concatenates the feature vectors of the static attribute information, dynamic feature vectors, and knowledge feature vectors to obtain the doctor node's own feature vector. The third acquisition module 403 is used to acquire the collaborative feature vector of the doctor node from the collaborative relationship graph, and to fuse the doctor node’s own feature vector and the doctor node’s collaborative feature vector to obtain the final feature vector of the doctor node. The first fusion module 404 is used to fuse the final feature vector of the patient node and the feature vectors of the patient node's neighboring nodes through a gated graph convolutional network to obtain the comprehensive feature vector of the patient node, and to fuse the final feature vector of the doctor node and the feature vectors of the doctor node's neighboring nodes to obtain the comprehensive feature vector of the doctor node. The second fusion module 405 is used to calculate the temporal trajectory attention between the patient's comprehensive feature vector and the doctor's comprehensive feature vector, and to calculate the semantic association attention between the patient's comprehensive feature vector and the doctor's comprehensive feature vector. The matching module 406 is used to process temporal trajectory attention and semantic association attention through the trained fully connected layer to obtain the matching score of patient nodes and doctor nodes. When the matching score meets the preset conditions, the doctor node is added to the recommendation list.

[0078] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0079] The beneficial effects of this application's embodiments are twofold. Firstly, by processing temporal trajectory attention and semantic association attention through a trained fully connected layer, matching scores between patient nodes and doctor nodes are obtained. When the matching score meets preset conditions, the doctor node is added to the recommendation list, solving the technical problem of how to match patient nodes and doctor nodes. Due to automatic matching, there is no need to manually input matching instructions, thus reducing the matching time between patient nodes and doctor nodes and improving matching efficiency. Secondly, by sending the recommendation list to the client corresponding to the patient node, the recommendation list can be directly delivered to the client, greatly shortening the circulation time of the recommendation list and enabling the client to obtain a recommendation list that meets its needs more promptly.

[0080] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0081] like Figure 5 As shown, Figure 5 The electronic device 2 includes: at least one processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the at least one processor 20, wherein the processor 20 executes the computer program 22 to implement the steps in any of the above method embodiments.

[0082] The electronic device 2 may include, but is not limited to, a processor 20 and a memory 21. Those skilled in the art will understand that... Figure 5 This is merely an example of electronic device 2 and does not constitute a limitation on electronic device 2. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, it may also include input / output devices, network access devices, etc.

[0083] The processor 20 is used to run a computer program 22 stored in the memory 21, and performs the following steps when executing the computer program 22: Obtain basic information, medical records, and consultation records of the patient node; and determine the final feature vector of the patient node based on these information. Obtaining static attribute information, dynamic attribute information and knowledge information of the doctor node, processing the dynamic attribute information and the knowledge information based on a predefined manner to obtain a dynamic feature vector and a knowledge feature vector, splicing the feature vector of the static attribute information, the dynamic feature vector and the knowledge feature vector to obtain a self-feature vector of the doctor node; obtaining a collaboration feature vector of the doctor node from the collaboration graph, and fusing the self-feature vector of the doctor node and the collaboration feature vector of the doctor node to obtain a final feature vector of the doctor node; Fusing the final feature vector of the patient node and the feature vector of the neighbor node of the patient node through the gated graph convolution network to obtain a comprehensive feature vector of the patient node, and fusing the final feature vector of the doctor node and the feature vector of the neighbor node of the doctor node to obtain a comprehensive feature vector of the doctor node; Calculating the time trajectory attention between the patient comprehensive feature vector and the doctor comprehensive feature vector, and calculating the semantic correlation attention between the patient comprehensive feature vector and the doctor comprehensive feature vector; Processing the time trajectory attention and the semantic correlation attention through the trained full connection layer to obtain a matching score of the patient node and the doctor node, and adding the doctor node to a recommendation list when the matching score meets a preset condition.

[0084] In some embodiments, the processor 20 is configured to implement: Obtaining basic information, disease records and treatment records of the patient node, performing feature encoding on the basic information to obtain a basic feature vector of the patient node, performing feature encoding on the disease records to obtain a first feature vector, and performing feature encoding on the treatment records to obtain a second feature vector; Splicing the first feature vector and the second feature vector to obtain a third feature vector, inputting the third feature vector into an LSTM unit, processing the third feature vector through the LSTM unit to obtain a fourth feature vector, processing the fourth feature vector using a Transformer model to obtain a fifth feature vector, splicing the fourth feature vector and the fifth feature vector to obtain a sixth feature vector, and splicing the sixth feature vector and the basic feature vector of the patient node to obtain a final feature vector of the patient node.

[0085] In some embodiments, the processor 20 is configured to implement: Obtaining static attribute information, dynamic attribute information and knowledge information of the doctor node, the dynamic attribute information including the number of past treatment cases and patient evaluation data, and the knowledge information including rare disease consultation data, rare disease paper data and rare disease clinical trial data; The current case number of the doctor node is obtained through the time decay model, the decay coefficient and the number of cases received at the past time point, each evaluation text in the patient evaluation data is obtained, each evaluation text is processed using a pre-trained sentiment analysis model to obtain a sentiment score of each evaluation text, a weighted model, a sentiment score of each evaluation text, a feature vector of each evaluation text and a time decay weight are used to fuse the weighted vector of each evaluation text using a multi-head attention mechanism to obtain a feature vector of the patient evaluation data, the feature vector of the current case number and the feature vector of the patient evaluation data are spliced to obtain a combined feature vector, the combined feature vector is dimensionally reduced to obtain a reduced combined feature vector, and the reduced combined feature vector is selected as a dynamic feature vector of the doctor node. The medical knowledge graph is accessed, and the feature vectors of the first-order neighbors and the second-order neighbors of the doctor node are obtained from the medical knowledge graph, the feature vectors of the first-order neighbors and the second-order neighbors of the doctor node are aggregated using a graph neural network algorithm to obtain a knowledge feature vector of the doctor node. The feature vectors of the static attribute information, the dynamic feature vector and the knowledge feature vector are spliced to obtain a self feature vector of the doctor node.

[0086] In some embodiments, the processor 20 is configured to: The time trajectory attention between the patient comprehensive feature vector and the doctor comprehensive feature vector is calculated through the time sequence trajectory attention model, and the semantic association attention between the patient comprehensive feature vector and the doctor comprehensive feature vector is calculated through the semantic association attention model.

[0087] In some embodiments, the processor 20 is configured to: The matching score of the patient node and the doctor node is obtained by processing the time trajectory attention and the semantic association attention through the trained fully connected layer; The dynamic threshold value is obtained through the dynamic threshold function, when the matching score of the patient node and the doctor node is greater than the dynamic threshold value, the doctor information corresponding to the doctor node is obtained, the doctor name, department, appointment time and registration fee are obtained from the doctor information, the doctor name, department, appointment time and registration fee are packaged into recommendation information, and the recommendation information is sent to the client corresponding to the patient node.

[0088] In some embodiments, the processor 20 is configured to: The sample set is obtained, and the fully connected layer is trained using a plurality of training samples in the sample set, in the training process, the total loss value of the fully connected layer is calculated through the target loss function, when the total loss value is less than the preset loss value, the training of the fully connected layer is stopped, and the trained fully connected layer is saved.

[0089] The processor 20 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0090] The memory 21 can be an internal storage unit of the electronic device 2 in some embodiments, such as a hard disk or a memory of the electronic device 2. The memory 21 can also be an external storage device of the electronic device 2 in other embodiments, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the memory 21 can include both the internal storage unit and the external storage device of the electronic device 2. The memory 21 is used to store an operating system, an application program, a boot loader, data, and other programs, such as program codes of the computer program, etc. The memory 21 can also be used to temporarily store data that has been output or will be output.

[0091] It should be noted that the information interaction, execution process, etc. between the above apparatuses / units, since based on the same concept as the method embodiments of the present application, the specific functions and the technical effects brought by the specific functions can be referred to the method embodiments part, and will not be described here.

[0092] The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps in each of the above method embodiments.

[0093] The computer readable storage medium stores program codes, and the program codes can be called and executed by the processor to implement the intelligent matching method described in the above method embodiments.

[0094] The computer readable storage medium has a storage space for storing program codes.

[0095] The program codes include codes of any steps in the intelligent matching method described in the above method embodiments.

[0096] For example, the program code is invoked by the processor, and the following steps can be performed: obtain the basic information, the illness record, and the medical record of the patient node, determine the final feature vector of the patient node based on the basic information, the illness record, and the medical record of the patient node; obtain the static attribute information, the dynamic attribute information, and the knowledge information of the doctor node, process the dynamic attribute information and the knowledge information based on a predefined manner to obtain a dynamic feature vector and a knowledge feature vector, and splice the feature vector of the static attribute information, the dynamic feature vector, and the knowledge feature vector to obtain the self-feature vector of the doctor node; obtain the collaboration feature vector of the doctor node from the collaboration graph, and fuse the self-feature vector of the doctor node and the collaboration feature vector of the doctor node to obtain the final feature vector of the doctor node; fuse the final feature vector of the patient node and the feature vector of the neighbor node of the patient node through the gate graph convolution network to obtain the comprehensive feature vector of the patient node, and fuse the final feature vector of the doctor node and the feature vector of the neighbor node of the doctor node to obtain the comprehensive feature vector of the doctor node; calculate the time trajectory attention between the patient comprehensive feature vector and the doctor comprehensive feature vector, and calculate the semantic association attention between the patient comprehensive feature vector and the doctor comprehensive feature vector; process the time trajectory attention and the semantic association attention through the trained full connection layer to obtain the matching score of the patient node and the doctor node, and when the matching score meets a preset condition, add the doctor node to the recommendation list.

[0097] The specific implementation of each operation can be referred to the foregoing embodiments, which will not be described here.

[0098] The computer readable storage medium can also be an external storage device of the intelligent matching device or the electronic device, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, a non-transitory computer readable storage medium, and the like.

[0099] The computer readable storage medium stores the computer program, and can execute any one of the intelligent matching methods based on the convolution network provided in the embodiments of the present application. Therefore, the computer readable storage medium can achieve the beneficial effects of any one of the intelligent matching methods based on the convolution network provided in the embodiments of the present application. Details are described in the foregoing embodiments, which will not be described here.

[0100] The embodiment of the application provides a computer program product, when the computer program product is run on an electronic device, the electronic device is caused to execute the intelligent matching method.

[0101] The computer program product is loaded by the electronic device, and the following steps can be executed: obtaining basic information, illness records and treatment records of the patient node, determining a final feature vector of the patient node based on the basic information, the illness records and the treatment records of the patient node; obtaining static attribute information, dynamic attribute information and knowledge information of the doctor node, processing the dynamic attribute information and the knowledge information based on a predefined manner to obtain a dynamic feature vector and a knowledge feature vector, and splicing the feature vector of the static attribute information, the dynamic feature vector and the knowledge feature vector to obtain a self feature vector of the doctor node; obtaining a cooperation feature vector of the doctor node from the cooperation relationship graph, and fusing the self feature vector of the doctor node and the cooperation feature vector of the doctor node to obtain a final feature vector of the doctor node; fusing the final feature vector of the patient node and the feature vector of the neighbor node of the patient node through a gate graph convolution network to obtain a comprehensive feature vector of the patient node, and fusing the final feature vector of the doctor node and the feature vector of the neighbor node of the doctor node to obtain a comprehensive feature vector of the doctor node; calculating time trajectory attention between the patient comprehensive feature vector and the doctor comprehensive feature vector, and calculating semantic association attention between the patient comprehensive feature vector and the doctor comprehensive feature vector; processing the time trajectory attention and the semantic association attention through the trained full connection layer to obtain a matching score of the patient node and the doctor node, and adding the doctor node into a recommendation list when the matching score meets a preset condition.

[0102] The integrated unit, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium.

[0103] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0104] Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to an electronic device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0105] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0106] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. An intelligent matching method based on convolutional networks, characterized in that, The intelligent matching method, applied to electronic devices, includes: Obtain basic information, medical records, and consultation records of the patient node; and determine the final feature vector of the patient node based on these information. Obtain the static attribute information, dynamic attribute information, and knowledge information of the doctor node. Process the dynamic attribute information and knowledge information in a predefined way to obtain dynamic feature vectors and knowledge feature vectors. Concatenate the feature vectors of the static attribute information, dynamic feature vectors, and knowledge feature vectors to obtain the doctor node's own feature vector. The collaborative feature vector of the doctor node is obtained from the collaborative relationship graph. The doctor node’s own feature vector and the collaborative feature vector of the doctor node are fused to obtain the final feature vector of the doctor node. By using a gated graph convolutional network, the final feature vector of the patient node is fused with the feature vectors of the patient node's neighboring nodes to obtain the comprehensive feature vector of the patient node. Similarly, the final feature vector of the doctor node is fused with the feature vectors of the doctor node's neighboring nodes to obtain the comprehensive feature vector of the doctor node. Calculate the temporal trajectory attention between the patient's comprehensive feature vector and the doctor's comprehensive feature vector, and calculate the semantic association attention between the patient's comprehensive feature vector and the doctor's comprehensive feature vector; The trained fully connected layer processes temporal trajectory attention and semantic association attention to obtain matching scores for patient nodes and doctor nodes. When the matching scores meet preset conditions, the doctor node is added to the recommendation list.

2. The intelligent matching method according to claim 1, characterized in that, The process of acquiring basic information, medical records, and consultation records of patient nodes, and determining the final feature vector of patient nodes based on these records, includes: Obtain the basic information, medical condition records, and medical visit records of the patient node; perform feature encoding on the basic information to obtain the basic feature vector of the patient node; perform feature encoding on the medical condition records to obtain the first feature vector; and perform feature encoding on the medical visit records to obtain the second feature vector. The first and second feature vectors are concatenated to obtain the third feature vector. The third feature vector is then input into an LSTM unit, which processes it to obtain the fourth feature vector. The fourth feature vector is then processed using a Transformer model to obtain the fifth feature vector. The fourth and fifth feature vectors are then concatenated to obtain the sixth feature vector. Finally, the sixth feature vector is concatenated with the basic feature vector of the patient node to obtain the final feature vector of the patient node.

3. The intelligent matching method according to claim 1, characterized in that, The process involves acquiring static attribute information, dynamic attribute information, and knowledge information of the doctor node; processing the dynamic attribute information and knowledge information based on a predefined method to obtain dynamic feature vectors and knowledge feature vectors; and concatenating the feature vectors of the static attribute information, dynamic feature vectors, and knowledge feature vectors to obtain the doctor node's own feature vector, including: Obtain static attribute information, dynamic attribute information, and knowledge information of doctor nodes. Dynamic attribute information includes the number of cases treated at past time points and patient evaluation data. Knowledge information includes rare disease consultation data, rare disease paper data, and rare disease clinical trial data. The current number of cases at a doctor node is obtained by using a time decay model, decay coefficient, and the number of cases treated at past time points. Each evaluation text in the patient evaluation data is obtained, and a pre-trained sentiment analysis model is used to process each evaluation text to obtain a sentiment score for each evaluation text. By using a weighted model, the sentiment score of each evaluation text, the feature vector of each evaluation text, and time decay weight, a multi-head attention mechanism is used to fuse the weighted vectors of each evaluation text to obtain the feature vector of the patient evaluation data. The feature vector of the current number of cases and the feature vector of the patient evaluation data are concatenated to obtain a combined feature vector. The combined feature vector is then dimensionality-reduced to obtain a dimensionality-reduced combined feature vector. The dimensionality-reduced combined feature vector is selected as the dynamic feature vector of the doctor node. Access the medical knowledge graph, obtain the feature vectors of the first-order and second-order neighbors of the doctor node from the medical knowledge graph, and use the graph neural network algorithm to aggregate the feature vectors of the first-order and second-order neighbors of the doctor node to obtain the knowledge feature vector of the doctor node. The feature vectors of static attribute information, dynamic feature vectors, and knowledge feature vectors are concatenated to obtain the doctor node's own feature vector.

4. The intelligent matching method according to claim 1, characterized in that, The calculation of the temporal trajectory attention between the patient's comprehensive feature vector and the doctor's comprehensive feature vector, and the calculation of the semantic association attention between the patient's comprehensive feature vector and the doctor's comprehensive feature vector, include: The temporal trajectory attention between the patient's comprehensive feature vector and the doctor's comprehensive feature vector is calculated using a temporal trajectory attention model, and the semantic association attention between the patient's comprehensive feature vector and the doctor's comprehensive feature vector is calculated using a semantic association attention model.

5. The intelligent matching method according to claim 1, characterized in that, The process involves processing temporal trajectory attention and semantic association attention through a fully connected layer after training to obtain matching scores between patient nodes and doctor nodes. When the matching scores meet preset conditions, the doctor node is added to the recommendation list, including: The training-processed fully connected layer handles temporal trajectory attention and semantic association attention to obtain matching scores for patient and doctor nodes. A dynamic threshold is obtained through a dynamic threshold function. When the matching score between a patient node and a doctor node is greater than the dynamic threshold, the doctor node is added to the recommendation list.

6. The intelligent matching method according to claim 1, characterized in that, Before acquiring the basic information, medical records, and consultation records of the patient node, and determining the final feature vector of the patient node based on the basic information, medical records, and consultation records of the patient node, the intelligent matching method includes: Obtain a sample set and train the fully connected layer using multiple training samples from the sample set. During the training process, calculate the total loss value of the fully connected layer using the target loss function. When the total loss value is less than the preset loss value, stop training the fully connected layer and save the trained fully connected layer.

7. The intelligent matching method according to any one of claims 1 to 6, characterized in that, The static attribute information includes the doctor's educational background, job position, and professional title; the dynamic attribute information includes the number of cases treated at past time points and patient evaluation data; and the knowledge information includes rare disease consultation data, rare disease paper data, and rare disease clinical trial data.

8. A smart matching device based on convolutional networks, characterized in that, Applied to electronic devices, including: The first acquisition module is used to acquire the basic information, medical records, and medical records of the patient node, and to determine the final feature vector of the patient node based on the basic information, medical records, and medical records of the patient node. The second acquisition module is used to acquire the static attribute information, dynamic attribute information, and knowledge information of the doctor node. It processes the dynamic attribute information and knowledge information in a predefined way to obtain dynamic feature vectors and knowledge feature vectors. It concatenates the feature vectors of the static attribute information, dynamic feature vectors, and knowledge feature vectors to obtain the doctor node's own feature vector. The third acquisition module is used to obtain the collaborative feature vector of the doctor node from the collaborative relationship graph, and to fuse the doctor node’s own feature vector and the doctor node’s collaborative feature vector to obtain the final feature vector of the doctor node. The first fusion module is used to fuse the final feature vector of the patient node and the feature vectors of the patient node's neighboring nodes through a gated graph convolutional network to obtain the comprehensive feature vector of the patient node, and to fuse the final feature vector of the doctor node and the feature vectors of the doctor node's neighboring nodes to obtain the comprehensive feature vector of the doctor node. The second fusion module is used to calculate the temporal trajectory attention between the patient's comprehensive feature vector and the doctor's comprehensive feature vector, and to calculate the semantic association attention between the patient's comprehensive feature vector and the doctor's comprehensive feature vector. The matching module is used to process temporal trajectory attention and semantic association attention through the trained fully connected layer to obtain the matching score of patient nodes and doctor nodes. When the matching score meets the preset conditions, the doctor node is added to the recommendation list.

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, it implements the intelligent matching method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the intelligent matching method as described in any one of claims 1 to 7.

Images