Abnormal behavior correlation analysis method and system for medical insurance data

Abstract

The present invention provides an abnormal behavior association analysis method and system for medical insurance data, the method comprising: based on the medical insurance scenario knowledge graph, according to the target medical treatment path, determining multiple medical insurance nodes to be analyzed that are associated with the target medical insurance node, and obtaining the medical treatment paths to be analyzed of each medical insurance node to be analyzed in the medical insurance scenario knowledge graph; according to the number of medical treatment information, weighting the target medical treatment path and the subpaths in each medical treatment path to be analyzed, and calculating the similarity value between the weighted target medical treatment path and each weighted medical treatment path to be analyzed; selecting the medical insurance node to be analyzed with a similarity value of TOP‑K as the target analysis medical insurance node, and generating the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node. The present invention improves the efficiency of medical insurance abnormal medical record review and the accuracy of abnormal behavior analysis results.

Description

Technical Field

[0001] The present invention relates to the technical field of medical insurance data analysis, and in particular to a method and system for abnormal behavior association analysis of medical insurance data. Background Art

[0002] With the gradual improvement of medical insurance management, medical insurance fraud is more likely to manifest itself as gang fraud, which is mainly divided into false medical treatment incidents, false impersonation and illegal medical service behaviors.

[0003] In the mining of medical insurance fraud, the most traditional method is rule matching, that is, medical experts formulate simple rules for data analysis and then manually review; or through outlier detection, using data mining methods to find abnormal data that is very different from general data in a specific mode. However, the above two methods have great limitations. They usually examine fraudulent behavior in a specific mode. At the same time, they are not scalable and lack immunity to new fraud modes. Therefore, it is difficult to meet the current needs of medical insurance fraud analysis.

[0004] Therefore, there is an urgent need for an abnormal behavior correlation analysis method and system for medical insurance data to solve the above problems. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a method and system for abnormal behavior association analysis of medical insurance data.

[0006] The present invention provides an abnormal behavior association analysis method for medical insurance data, comprising:

[0007] Based on the medical insurance scenario knowledge graph, according to the target medical treatment path, multiple medical insurance nodes to be analyzed that are associated with the target medical insurance node are determined, and the medical treatment paths to be analyzed of each of the medical insurance nodes to be analyzed in the medical insurance scenario knowledge graph are obtained;

[0008] According to the visit frequency information, weighting is performed on the target visit path and each subpath in the visit path to be analyzed, and the similarity value between the weighted target visit path and each weighted visit path to be analyzed is calculated;

[0009] Select the to-be-analyzed medical insurance node with the similarity value TOP-K as the target analysis medical insurance node, and generate the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node;

[0010] Among them, the target medical treatment path is obtained from the medical insurance scenario knowledge graph based on the medical insurance information and node type of the target medical insurance node; the path types of the target medical treatment path and the medical treatment path to be analyzed include meta-paths and / or meta-structures.

[0011] According to a method for abnormal behavior association analysis of medical insurance data provided by the present invention, after selecting the medical insurance node to be analyzed with a similarity value TOP-K as the target analysis medical insurance node, and generating the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node, the method further includes:

[0012] According to the connection relationship between each medical insurance node in the medical insurance data association analysis result, the topological structure between the medical insurance nodes of the same node type is obtained;

[0013] According to the attribute information of the medical insurance node of each node type in the medical insurance data association analysis result, construct an attribute coding structure corresponding to the attribute information;

[0014] Through the community discovery algorithm, the global representation of the nodes in the medical insurance data is learned according to the topological structure and the attribute coding structure, and the medical insurance nodes in the medical insurance data association analysis results are clustered according to the global representation to obtain the medical insurance data association analysis results within the same community.

[0015] According to a method for abnormal behavior association analysis of medical insurance data provided by the present invention, after selecting the medical insurance node to be analyzed with a similarity value TOP-K as the target analysis medical insurance node, and generating the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node, the method further includes:

[0016] Select any two medical insurance nodes from the medical insurance data association analysis results;

[0017] According to the preset query stage path, the association path between any two medical insurance nodes in the medical insurance data association analysis result is obtained through a random walk algorithm.

[0018] According to a method for abnormal behavior association analysis of medical insurance data provided by the present invention, the method further includes:

[0019] Generate a visualization view of similarity exploration of medical insurance nodes based on the medical insurance data association analysis results;

[0020] Generate a visualization view of the community discovery of medical insurance nodes based on the results of the association analysis of the medical insurance data in the same community;

[0021] Generate a correlation path exploration visualization view according to the correlation path;

[0022] The abnormal behavior correlation analysis of medical insurance data is visualized and displayed through the medical insurance node similarity exploration visualization view, the medical insurance node community discovery visualization view and the association path exploration visualization view.

[0023] According to an abnormal behavior association analysis method for medical insurance data provided by the present invention, the similarity value between the weighted target medical treatment path and each weighted medical treatment path to be analyzed is calculated, including:

[0024] If the path types of the target medical path and the to-be-analyzed medical path are meta-paths, the similarity value is calculated according to a first similarity formula, which is:

[0025]

[0026] Among them, S sum (P x→y ) represents the total score of the weighted meta-path P between the medical insurance node x and the medical insurance node y; S sum (P x→x ) represents the sum of the scores of the weighted meta-paths P between medical insurance nodes x and medical insurance nodes x; S sum (P y→y ) represents the sum of the scores of the weighted meta-paths P between the medical insurance node y and the medical insurance node y; Represents each sub-path in the weighted meta-path, R i Represents node A i and node A i+1 The link relationship between i Represents the weight of the i-th link relationship.

[0027] According to an abnormal behavior association analysis method for medical insurance data provided by the present invention, the similarity value between the weighted target medical treatment path and each weighted medical treatment path to be analyzed is calculated, including:

[0028] If the path type of the target medical treatment path and the medical treatment path to be analyzed is a meta structure, the similarity value is calculated according to the second similarity formula, and the second similarity formula is:

[0029]

[0030] Among them, i Represents the i-th instance medical insurance node set N of the weighted meta-structure S i In the instance medical insurance node, {o i+1 |(o i ,o i+1 )∈E} represents the instance medical insurance node o on the i+1th layer iThere is an associated instance medical insurance node; Represents the instance medical insurance node o i Go to the instance medical insurance node o i+1 The path weight of Represents each subpath in the weighted metastructure S, M i Represents two groups of nodes N i and N i+1 The link between i Represents the weight of the connection between objects.

[0031] The present invention also provides an abnormal behavior association analysis system for medical insurance data, comprising:

[0032] A medical insurance knowledge graph node path acquisition module is used to determine multiple medical insurance nodes to be analyzed that are associated with a target medical insurance node based on the medical insurance scenario knowledge graph and the target medical treatment path, and obtain the medical treatment paths to be analyzed of each of the medical insurance nodes to be analyzed in the medical insurance scenario knowledge graph;

[0033] A weighted path similarity calculation module is used to perform weighted processing on the target medical path and each subpath in the medical path to be analyzed according to the number of medical visits information, and calculate the similarity value between the weighted target medical path and each weighted medical path to be analyzed;

[0034] A medical insurance node similarity association analysis result generation module is used to select the medical insurance nodes to be analyzed with similarity values ​​TOP-K as target analysis medical insurance nodes, and generate corresponding medical insurance data association analysis results according to the co-occurrence medical treatment paths between the target medical insurance nodes and the target analysis medical insurance nodes;

[0035] Among them, the target medical treatment path is obtained from the medical insurance scenario knowledge graph based on the medical insurance information and node type of the target medical insurance node; the path types of the target medical treatment path and the medical treatment path to be analyzed include meta-paths and / or meta-structures.

[0036] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-described methods for abnormal behavior correlation analysis of medical insurance data are implemented.

[0037] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described methods for abnormal behavior association analysis of medical insurance data.

[0038] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned methods for abnormal behavior association analysis of medical insurance data.

[0039] The abnormal behavior association analysis method and system for medical insurance data provided by the present invention address the problem that in existing medical insurance fraud detection scenarios, many clues are cumbersome and have weak surface correlation, and medical insurance fraud analysis can only rely on a single clue. An association exploration analysis method based on weighted similarity calculation is proposed, which innovates the medical insurance audit model and can expand the clue chain based on existing fraud cases, so as to find cases that are highly similar to known fraud cases, greatly improve the efficiency of medical insurance abnormal medical record review, and improve the accuracy of abnormal behavior analysis results of medical insurance data. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A flow chart of the abnormal behavior correlation analysis method for medical insurance data provided by the present invention;

[0042] Figure 2 A schematic diagram of weighted subpaths based on the number of visits information provided by the present invention;

[0043] Figure 3 A schematic diagram of the structure of the abnormal behavior correlation analysis system for medical insurance data provided by the present invention;

[0044] Figure 4 A schematic diagram of the implementation process of the abnormal behavior correlation analysis system for medical insurance data provided by the present invention;

[0045] Figure 5 This is a schematic structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] In the existing medical insurance fraud mining, it is often limited by the level of experts and lacks scalability. Some studies have begun to consider pattern mining, such as studying abnormal behaviors in the use of medical insurance cards or diagnosis and treatment methods, using frequent pattern mining, Linear Discriminant Analysis (LDA) basic model and other methods to detect; or modeling medical insurance data as a graph, proposing anomaly detection methods based on the internal features of medical record data sets and network graph exploration. However, there is not much research on association exploration analysis in the current medical insurance fraud analysis methods. Therefore, association exploration analysis and visualization will be a model innovation for medical insurance fraud.

[0048] For traditional relational data, node similarity algorithms have been widely studied, but most of these studies focus on homogeneous networks or bipartite graphs, ignoring the subtle differences between different types of objects and links. If these node similarity algorithms are used on heterogeneous graphs, they will not be able to handle the different semantic relationships brought by objects of different types and links.

[0049] Figure 1 A flow chart of the abnormal behavior correlation analysis method for medical insurance data provided by the present invention is shown in FIG. Figure 1 As shown, the present invention provides an abnormal behavior association analysis method for medical insurance data, comprising:

[0050] Step 101, based on the medical insurance scenario knowledge graph and according to the target medical treatment path, determine multiple medical insurance nodes to be analyzed that are associated with the target medical insurance node, and obtain the medical treatment paths to be analyzed of each of the medical insurance nodes to be analyzed in the medical insurance scenario knowledge graph.

[0051] In the present invention, the entities in the medical insurance scenario knowledge graph may be patients, doctors, diseases, departments, hospitals, etc., and the edges between entities, that is, the entity relationships, may be the number of visits, prescription behavior, and hospitalization behavior, etc.

[0052] Further, in the present invention, when the medical insurance auditor obtains clues of abnormal medical record reports or obtains confirmed fraud cases, the medical insurance information corresponding to the abnormal medical record reports or fraud cases is used to obtain the medical treatment path of the medical insurance information in the medical insurance scene knowledge map, that is, the target medical treatment path is obtained, for example, patient-hospital-diagnosis, indicating that the patient has been diagnosed in a certain hospital. Specifically, the medical insurance information includes the patient's name, gender, social insurance type, personnel category, payment area information, date of birth, work unit, social security card number and residential address information, etc. The node ID of the relevant patient can be obtained through the medical insurance information, so that the medical insurance node of the patient can be directly located in the medical insurance scene knowledge map through the node ID as the target medical insurance node. It should be noted that in the medical insurance scene knowledge map, each entity has its own node ID. When the medical insurance auditor receives a fraud report or confirms a fraud case through review, the node ID and node type will be input into the knowledge map system to obtain relevant medical insurance information, thereby performing association exploration.

[0053] Furthermore, after determining the target medical insurance node and the target medical treatment path of the target medical insurance node, other medical insurance nodes with the same path definition as the target medical treatment path are selected from the medical insurance scenario knowledge graph, and these other medical insurance nodes are used as medical insurance nodes to be analyzed, and the medical treatment paths of these medical insurance nodes to be analyzed are used as medical treatment paths to be analyzed.

[0054] After determining the medical insurance nodes to be analyzed and the medical treatment paths to be analyzed, based on the type of medical treatment path (in the medical insurance scenario knowledge graph, the connection structure between entities has two structures: meta-path and meta-structure), according to the target medical treatment path and the medical treatment path to be analyzed, a symmetrical meta-path or symmetrical meta-structure is constructed. Taking the patient node association exploration scenario as an example, the present invention defines 10 meta-paths and meta-structure relationships for calculating the node similarity between two nodes in different semantic scenarios. For details, please refer to Table 1:

[0055] Table 1

[0056] Mining scene Meta-path / Meta-structure definition Semantic interpretation Patient-Patient Meta Path Patient-Doctor-Patient Two patients visited the same doctor Patient-Patient Meta Path Patient-Department-Patient Two patients visited the same department Patient-Patient Meta Path Patient-Hospital-Patient Both patients visited the same hospital Patient-Patient Meta Path Patient-Disease-Patient Both patients had the same diagnosis Patient-Patient Meta Path Patient-Doctor-Diagnosis-Doctor-Patient Two patients received the same diagnosis from two doctors Patient-Patient Meta Path Patient-Department-Diagnosis-Department-Patient Two patients received the same diagnosis in two departments Patient-Patient Meta Path Patient-Hospital-Diagnosis-Hospital-Patient Two patients received the same diagnosis in two hospitals Patient-Patient Metastructure patient-[doctor,disease]-patient Two patients received the same diagnosis from the same doctor Patient-Patient Metastructure Patient-[Department, Diagnosis]-Patient Two patients received the same diagnosis in the same department Patient-Patient Metastructure Patient-[hospital,diagnosis]-patient Two patients received the same diagnosis in the same hospital

[0057] In Table 1, the target medical insurance node (for example, the patient on the left side of the entity relationship in the mining scenario) and the medical insurance node to be analyzed (such as the patient on the right side of the entity relationship in the mining scenario) have the same path (meta-path or meta-structure) in the knowledge graph. For example, the two patients have visited the same doctor, so a symmetric meta-path can be constructed, which is used to calculate the similarity value between the two medical insurance nodes in the subsequent calculation. It should be noted that if the meta-paths of the two medical insurance nodes are asymmetric, such as two patients visiting different hospitals, the similarity value of the two medical insurance nodes under this meta-path can be determined to be 0.

[0058] Step 102, weighting the target medical path and each subpath in the medical path to be analyzed according to the medical visit frequency information, and calculating the similarity value between the weighted target medical path and each weighted medical path to be analyzed.

[0059] In the medical insurance data knowledge graph, heterogeneous networks can connect entities such as patients, doctors, diseases, departments, and hospitals. However, since there are many-to-many relationships between entities, and the relationship between the previous two entities may affect the relationship between the next two entities, traditional heterogeneous networks may produce some wrong information when expressing the relationship between medical insurance entities.

[0060] Since the knowledge graph in the medical insurance scenario is different from other types of knowledge graphs, that is, the number of relationships between medical insurance entity nodes usually carries certain semantic information, such as the number of visits, prescriptions issued by a certain doctor, etc., and these semantic information are often the key clues for mining abnormal medical records. The existing pathsim algorithm does not consider the weights in the node relationship when calculating the similarity value. Therefore, the present invention assigns weights to the edges between nodes, and innovates the similarity calculation method of meta-path and meta-structure nodes, and proposes a weighted meta-path and meta-structure similarity calculation method to adapt to the similarity calculation in the medical insurance scenario.

[0061] Figure 2 The weighted schematic diagram of the subpath based on the number of visits information provided by the present invention can be referred to Figure 2 As shown, in Figure 2 In (a), the actual medical records of a doctor are shown, where the doctor-disease line is labeled with {<patient, number of visits>}. For example, patient 1 was diagnosed with disease 1 twice by doctor 1, but was not diagnosed with disease 3 by doctor 1. However, in actual heterogeneous networks, Figure 2 (b) , it is impossible to know that patient 1 was diagnosed with 2nd diagnosis 1 by doctor 1. In addition, heterogeneous networks may have incorrect information that patient 1 was diagnosed with 3rd diagnosis by doctor 1. The correct diagnosis information can be found in Figure 2 As shown in (c), the information on the medical treatment paths of patient 1 and patient 2 can be obtained. Since the traditional heterogeneous network may ignore some frequency information when expressing, and may also produce erroneous information. Therefore, the present invention performs weighted processing on each subpath in the medical treatment path according to the number of medical treatment information, as shown in FIG. Figure 2As shown in (d), patient 1 was diagnosed with disease 1 twice by doctor 1 and was diagnosed with disease 2 three times by doctor 1. At this time, the weight of the path between doctor 1 and disease 1 is 2, and the link relationship is patient 1. Similarly, the weights of the paths between doctor 1 and disease 2 and disease 3 are 3 and 0 respectively, and the link relationship is patient 1. In another embodiment, the number of visits of a patient to different hospitals can be used to determine the weights of the paths between hospitals and different diagnosed diseases based on the same method as above, so that each path can carry more information to participate in the similarity value calculation.

[0062] After weighting the subpaths in the target medical path and each of the medical paths to be analyzed, similarity values ​​are calculated based on the symmetric meta-path formed by the target medical path and each of the medical paths to be analyzed or on this meta-structure. The present invention improves the calculation method of the Pathsim algorithm and the Strucsim algorithm, and constructs a weighted meta-path similarity calculation method, namely weighted-strusim and weighted-pathsim, which can support the calculation of weighted meta-paths / weighted meta-structures. Specifically, on the basis of the above embodiment, the similarity value between the weighted target medical path and each of the weighted medical paths to be analyzed is calculated, including:

[0063] If the path types of the target medical path and the to-be-analyzed medical path are meta-paths, the similarity value is calculated according to a first similarity formula, which is:

[0064]

[0065] Among them, S sum (P x→y ) represents the total score of the weighted meta-path P between the medical insurance node x and the medical insurance node y; S sum (P x→x ) represents the sum of the scores of the weighted meta-paths P between medical insurance nodes x and medical insurance nodes x; S sum (P y→y ) represents the sum of the scores of the weighted meta-paths P between the medical insurance node y and the medical insurance node y; Represents each sub-path in the weighted meta-path, R i Indicates medical insurance node A i and medical insurance node A i+1 The link relationship between i Represents the weight of the i-th link relationship.

[0066] Based on the above embodiment, the similarity value between the weighted target medical treatment pathway and each weighted medical treatment pathway to be analyzed is calculated, including:

[0067] If the path type of the target medical treatment path and the medical treatment path to be analyzed is a meta structure, the similarity value is calculated according to the second similarity formula, and the second similarity formula is:

[0068]

[0069] Among them, i Represents the i-th instance medical insurance node set N of the weighted meta-structure S i In the instance medical insurance node, {o i+1 |(o i ,o i+1 )∈E} is the node o on the i+1th layer that corresponds to the instance medical insurance node i There is an associated instance medical insurance node; Represents the instance medical insurance node o i Go to the instance medical insurance node o i+1 The second similarity formula provided by the present invention is a recursive calculation, and the termination condition is when i=n+1, where n is the number of layers of the meta-structure. i With o j When on the same layer, if o i =o j ,P(o i ,o j )=1, if o i ≠O j ,P(o i ,o j )=0. The weighted element structure can be expressed as Each subpath in the weighted metastructure Two groups of nodes N are defined i and N i+1 The compound relationship between i Represents two groups of nodes N i and N i+1 The link between i Represents the weight of the connection between objects.

[0070] Step 103, select the to-be-analyzed medical insurance node with a similarity value of TOP-K as the target analysis medical insurance node, and generate the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node.

[0071] Among them, the target medical treatment path is obtained from the medical insurance scenario knowledge graph based on the medical insurance information and node type of the target medical insurance node; the path types of the target medical treatment path and the medical treatment path to be analyzed include meta-paths and / or meta-structures.

[0072] In the present invention, first, the nodes of the same type that have a first-order association relationship with the target medical insurance node are screened as the candidate set for node similarity mining (i.e., the set of medical insurance nodes to be analyzed). It should be noted that the present invention can also screen the nodes of the same type that have multi-order association relationships (i.e., define multiple stage entity relationships, improve the data features in the symmetric meta-path or symmetric meta-structure), which can be determined according to the actual association analysis needs; then, according to the meta-path and meta-structure relationships defined based on semantic relationships, the weighted-pathsim algorithm and the weighted-strusim algorithm are used on the medical insurance scenario knowledge graph to calculate the medical The similarity feature vector between the candidate nodes in the medical insurance scenario knowledge graph and the target medical insurance node; preferably, in the present invention, the labeled training set is used to use the LambdaMART algorithm to optimize the weight of the similarity feature vector to obtain the final similarity between the nodes; finally, the node with TOP-K similarity to the query node (i.e., the target medical insurance node) and the ID of the node (the medical insurance information of the node can be obtained based on the ID) are selected, thereby determining multiple target analysis medical insurance nodes, and returning the association path (i.e., the co-occurrence treatment path) between the target medical insurance node and the target analysis medical insurance node to generate a corresponding visualization display.

[0073] In the present invention, the generated similarity visualization display results can display the attribute information of the node, the medical record and the details of the top-n nodes with similarity through a list view (visual view). The visualization view displays the node similarity results in the form of a graph visualization, so that medical insurance auditors can intuitively understand the entire similar node network through the graph, or click on a single node to understand the node attributes and initiate the next step of exploration.

[0074] The abnormal behavior association analysis method for medical insurance data provided by the present invention aims to solve the problem that in existing medical insurance fraud detection scenarios, many clues are cumbersome and have weak surface correlation, and medical insurance fraud analysis can only rely on a single clue. An association exploration analysis method based on weighted similarity calculation is proposed, which innovates the medical insurance audit model and can expand the clue chain based on existing fraud cases, so as to find cases that are highly similar to known fraud cases, greatly improve the efficiency of medical insurance abnormal medical record review, and improve the accuracy of abnormal behavior analysis results of medical insurance data.

[0075] On the basis of the above embodiment, after selecting the to-be-analyzed medical insurance node with the similarity value TOP-K as the target analysis medical insurance node, and generating the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node, the method further includes:

[0076] According to the connection relationship between each medical insurance node in the medical insurance data association analysis result, the topological structure between the medical insurance nodes of the same node type is obtained;

[0077] According to the attribute information of the medical insurance node of each node type in the medical insurance data association analysis result, construct an attribute coding structure corresponding to the attribute information;

[0078] Through the community discovery algorithm, the global representation of the nodes in the medical insurance data is learned according to the topological structure and the attribute coding structure, and the medical insurance nodes in the medical insurance data association analysis results are clustered according to the global representation to obtain the medical insurance data association analysis results within the same community.

[0079] Existing community discovery algorithms can be roughly divided into seven categories: graph partitioning, statistical inference, hierarchical clustering, dynamic methods, spectral clustering, density-based methods, and optimization methods. However, existing methods can only obtain shallow connections between nodes in a direct way, while deep learning methods can simulate complex relationships from high-dimensional to low-dimensional vectors and reveal deep network information. Community discovery based on deep learning can obtain effective community detection results because it can embed unstructured features and can jointly identify information groups from nodes, edges, neighborhoods, or multiple graphs. Since the Graph Convolutional Network (GCN) can aggregate node neighborhood information in the convolutional layer and globally capture complex features, it can be used in community discovery scenarios in attribute graphs. For attribute graphs, GCN-based community detection can rely on the attribute information of the graph (including patient name, identity information, residential area information, and medical information, etc.) and the attribute encoding structure information obtained based on the attribute information encoding, and cluster nodes with similar attributes in a community. Among them, the more representative one is Attributed Graph Clustering (AGC), which can embed low-pass filters into adaptive graph convolutions to smoothly filter out high-frequency noise and realize community discovery for attribute graphs.

[0080] In the present invention, the community discovery algorithm model is constructed using the AGC algorithm, while taking into account the topological structure of the graph and the semantic information on the graph. In the topological structure, the connection relationship between nodes is used to calculate the topological distance between nodes of the same type; in the semantic information of the graph, the attribute information required for each type of node is selected, weights are assigned, and the semantic similarity between nodes is analyzed. After processing the topological structure and semantic information of the node, other nodes in the same community as the target medical insurance node and their associated paths are obtained.

[0081] Specifically, in the present invention, because heterogeneous networks are difficult to cluster, the community discovery algorithm provided by the present invention first maps the heterogeneous network to a homogeneous network for processing. During the mapping process, the community discovery algorithm adds associations to nodes of the same type with the same neighbors, deletes intermediate neighbors, so that the original medical insurance knowledge graph is mapped to an isomorphic graph with only one type of node. Taking the mapping of the patient isomorphic graph as an example, the community discovery algorithm adds associations between patients who have visited the same doctor, the same department, the same hospital, and the same diagnosis, and deletes the remaining intermediate nodes except the patient node, so that the heterogeneous graph containing multiple types of nodes is mapped to a patient node isomorphic graph.

[0082] Furthermore, the AGC algorithm is used when learning the global representation of nodes using neural networks. The AGC algorithm understands GCN from the perspective of the spectral graph theory of graph signal processing. The AGC algorithm is based on the principle that if adjacent nodes on the graph have similar feature representations, the graph signal is smooth. A low-pass filter is designed to filter high-frequency signals in the graph signal F and capture low-frequency signals, which is more suitable for clustering tasks. Among them, the K-order graph convolution is defined as:

[0083]

[0084]

[0085] The iterative calculation formula of K-order graph convolution is:

[0086]

[0087]

[0088] …

[0089]

[0090] in, For node v i The K-order representation of (v i ,v j )∈ε is v i ,v j The connection between ij is the adjacency matrix The value in d i For node v i Degree, x i For node v i The attribute vector of .

[0091] From the above K-order graph convolution iteration formula, we can see that the K-order graph convolution updates the features of each node by iteratively aggregating the features of its K-hop neighbors. Since the size of the K value determines the effect of graph convolution learning, in order to obtain the best effect, the community discovery algorithm uses a heuristic method to design an adaptive K value. Based on the principle of minimum intra-class distance and maximum inter-class distance, the algorithm gives a clustering performance indicator. When the value of intra(C) reaches a local minimum, the current K value is the optimal K value, that is:

[0092]

[0093] After AGC learns the node features, the community discovery algorithm performs spectral clustering on the extracted features to cluster the medical insurance nodes, and returns the clustering results in the form of an offline table to generate corresponding visual display results.

[0094] In the present invention, possible medical insurance fraud gangs are explored through community discovery algorithms, and the obtained visualization results are displayed in a list view to show the details of other nodes in the same community as the target node. This visualization view displays the entire community structure in a graph visualization way, so that auditors can understand the co-occurrence characteristics of the entire community and dig out clues hidden in the community.

[0095] On the basis of the above embodiment, after selecting the to-be-analyzed medical insurance node with the similarity value TOP-K as the target analysis medical insurance node, and generating the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node, the method further includes:

[0096] Select any two medical insurance nodes from the medical insurance data association analysis results;

[0097] According to the preset query stage path, the association path between any two medical insurance nodes in the medical insurance data association analysis result is obtained through a random walk algorithm.

[0098] In the present invention, the exploration of associated paths is performed based on two medical insurance nodes, so as to discover the hidden associated paths between the medical insurance nodes, thereby exploring new co-occurring medical treatment paths that may exist between the two medical insurance nodes. Since the associated path query runs in a heterogeneous graph composed of different medical insurance entities, the present invention first needs to define the required query meta-path, and through the random walk algorithm, starting from the two input points, initiate searches to each other according to the query meta-path, intersect the multi-order neighbors of the two points, realize the discovery of associated paths, and return the matched path for generating visual display results.

[0099] In the present invention, the exploration of the associated path displays all the mined path results in a list view. The visual view displays all the associated paths between two target nodes, so that auditors can easily find the key nodes hidden in the graph that are associated with both nodes and initiate an audit on them.

[0100] Based on the above embodiment, the method further includes:

[0101] Generate a visualization view of similarity exploration of medical insurance nodes based on the medical insurance data association analysis results;

[0102] Generate a visualization view of the community discovery of medical insurance nodes based on the results of the association analysis of the medical insurance data in the same community;

[0103] Generate a correlation path exploration visualization view according to the correlation path;

[0104] The abnormal behavior correlation analysis of medical insurance data is visualized and displayed through the medical insurance node similarity exploration visualization view, the medical insurance node community discovery visualization view and the association path exploration visualization view.

[0105] The current knowledge graph visualization tools are divided into three categories according to the language category: JavaScript, R / Python, and Business Intelligence. Among them, the representative tools of JavaScript are d3.js, Echart, and Cytoscape. JavaScript tools are highly flexible and extensible, and can handle massive amounts of visual data, but such tools often have high technical barriers and require programming language foundations such as Html, Css, and JavaScript. Representative tools based on the R / Python language include Plotly and rCharts, which are highly flexible and extensible and suitable for data analysts to use, but the technical barriers are high, and the layout algorithm is not built-in, which is not flexible enough. Representative tools of the Business Intelligence category include Tableau, FineBI, etc., but they have low flexibility and extensibility.

[0106] In the present invention, since the visualization of the knowledge graph involves the rendering of a large number of nodes, there are extremely high requirements for the rendering speed and accuracy of the visualization component. In the visualization selection, the present invention selects Cytoscape.js combined with the front-end VUE framework for visualization rendering and constructs the association relationship between nodes.

[0107] In one embodiment, the patient node with medical insurance node ID = 65963072 in the medical insurance scenario graph data is taken as an example to illustrate the similarity exploration of its running nodes. Since the node information involves sensitive information such as medical insurance card number, ID card number and address, the sensitive information is desensitized in the following embodiment. The technical solution implementation process is as follows:

[0108] 1. Retrieve node information: The node type selected for retrieval is PATIENT_ID, and the node ID is 65963072;

[0109] 2. Backend processing and query: After receiving the search request, the attribute information of the node and the detailed information of its 10 medical records are returned;

[0110] 3. Node similarity exploration: First, the patient's associated patients are circled, and 2440 patients are circled as the candidate set for node similarity calculation; then, the similarity value between the patient and the associated patients is calculated through the node similarity algorithm, and the top-50 nodes with the final similarity ranking are returned, among which there are 2440 associated patient nodes and 1583631 constructed meta-paths; the average similarity of the top-100 nodes is 0.576666, of which the maximum value is 1, that is, the treatment path of this node is exactly the same as that of the target patient;

[0111] 4. Result visualization rendering: After the front end obtains the returned data and associated paths, it performs visualization rendering on the similarity.

[0112] 5. Auxiliary audit: The auditor clicks on a related node to proceed to the next step of exploration, and will run the related exploration and analysis algorithm again to visualize the results.

[0113] In another embodiment, taking the patient node with ID=65963072 in the graph data as an example, a fraud group exploration is performed on it, that is, a community discovery algorithm is used to analyze whether there is a group medical insurance fraud. The technical solution implementation process is as follows:

[0114] 1. Retrieve node information: The node type selected for retrieval is PATIENT_ID, and the node ID is 65963072;

[0115] 2. Backend processing and query: After receiving the search request, the attribute information of the node and the detailed information of its 10 medical records are returned;

[0116] 3. Fraud gang exploration: The community discovery algorithm searches for the remaining 434 patient nodes in the same community as the patient node, and obtains the attribute information of each node. The average value of the comprehensive similarity is 0.5358, which means that when the community discovery algorithm is performed, the similarity between patient nodes is also calculated;

[0117] 4. Result visualization rendering: The front end obtains the returned data and visualizes the community network structure;

[0118] 5. Auxiliary audit: The auditor clicks on a related node to proceed to the next step of exploration, and will run the related exploration and analysis algorithm again to visualize the results.

[0119] In another embodiment, taking two patient nodes with ID1=41061500 and ID2=32286529 in the atlas data as an example, the associated path exploration is performed on them. The technical solution implementation process is as follows:

[0120] 1. Retrieve node information: The node type selected for retrieval is PATIENT_ID, node ID1 = 41061500, node ID2 = 32286529;

[0121] 2. Background processing and query: After receiving the search request, the background returns the attribute information of the two nodes.

[0122] 3. Association exploration analysis: After obtaining the node ID, the association path exploration algorithm, namely the random walk algorithm, explores the association path between two nodes according to the defined first-order, second-order and third-order meta-paths, and returns the explored 7411 path information, among which the total number of association paths is 7411, the total number of first-order relationships is 3, the total number of second-order relationships is 312, and the total number of third-order relationships is 7411;

[0123] 4. Result visualization rendering: The front end obtains the returned data and performs visualization rendering of the association path between the two nodes;

[0124] 5. Auxiliary audit: The auditor clicks on a related node to proceed to the next step of exploration, and will run the related exploration and analysis algorithm again to visualize the results.

[0125] The abnormal behavior association analysis method for medical insurance data provided by the present invention innovates the form of medical insurance auditing through association exploration analysis (including similarity calculation, community discovery algorithm and association path mining) and visualization system, and deeply mines the association clues between medical insurance nodes through data processing, storage management and visualization interaction technologies, expands the medical insurance fraud clue chain, and efficiently locates suspicious nodes.

[0126] The abnormal behavior association analysis system for medical insurance data provided by the present invention is described below. The abnormal behavior association analysis system for medical insurance data described below and the abnormal behavior association analysis method for medical insurance data described above can be referenced to each other.

[0127] Figure 3 The schematic diagram of the structure of the abnormal behavior correlation analysis system for medical insurance data provided by the present invention is as follows: Figure 3 As shown, the present invention provides an abnormal behavior association analysis system for medical insurance data, including a medical insurance knowledge graph node path acquisition module 301, a weighted path similarity calculation module 302 and a medical insurance node similarity association analysis result generation module 303, wherein the medical insurance knowledge graph node path acquisition module 301 is used to determine multiple medical insurance nodes to be analyzed that are associated with a target medical insurance node based on a medical insurance scenario knowledge graph and a target medical treatment path, and obtain the medical treatment paths to be analyzed of each of the medical insurance nodes to be analyzed in the medical insurance scenario knowledge graph; the weighted path similarity calculation module 302 is used to perform weighted processing on the target medical treatment path and each of the sub-paths in the medical treatment paths to be analyzed according to the number of medical treatment information, and calculate the similarity value between the weighted target medical treatment path and each weighted medical treatment path to be analyzed; the medical insurance node similarity association analysis result generation module 303 is used to select the medical insurance node to be analyzed with a similarity value of TOP-K as the target analysis medical insurance node, and generate the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node;

[0128] Among them, the target medical treatment path is obtained from the medical insurance scenario knowledge graph based on the medical insurance information and node type of the target medical insurance node; the path types of the target medical treatment path and the medical treatment path to be analyzed include meta-paths and / or meta-structures.

[0129] Figure 4 This is a schematic diagram of the implementation process of the abnormal behavior correlation analysis system for medical insurance data provided by the present invention, which can be referred to Figure 4 As shown in the figure, first, according to the existing clues, the remaining nodes that are highly similar to the medical treatment path of the target medical node are determined, and the known clues are used as input parameters. Through the similarity of the medical treatment path, the associated suspicious nodes are mined, and the determined target medical insurance node ID is input into the system; then, the back-end processes and retrieves the target medical insurance node ID, returns the attribute information and historical medical treatment records of the ID, and inputs the node information into the association exploration analysis algorithm; further, the association exploration analysis algorithm runs the node similarity algorithm, community discovery algorithm and association path mining algorithm according to the input node information, finds the associated nodes, and mines the nodes that have attributes and medical treatment relationships similar to the retrieved nodes or are suspected of being in the same fraud community from multiple angles; then, the association exploration analysis algorithm returns the mining results to the server layer, and the server transmits them to the visualization module after processing for graph visualization display, which intuitively shows the co-occurrence relationship between the suspicious medical insurance nodes and the medical treatment paths of the target medical insurance nodes; finally, the medical insurance auditors initiate node information retrieval and association exploration again through the mined suspicious cases to confirm the possibility of fraud in the suspicious cases.

[0130] The abnormal behavior association analysis system for medical insurance data provided by the present invention proposes an association exploration analysis method based on weighted similarity calculation to address the problem that in existing medical insurance fraud detection scenarios, many clues are cumbersome and have weak surface correlation, and medical insurance fraud analysis can only rely on a single clue. It innovates the medical insurance audit model and can expand the clue chain based on existing fraud cases to find cases that are highly similar to known fraud cases, greatly improving the efficiency of medical insurance abnormal medical record review and improving the accuracy of abnormal behavior analysis results of medical insurance data.

[0131] The system provided by the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific processes and detailed contents, which will not be repeated here.

[0132] Figure 5 A schematic diagram of the structure of an electronic device provided by the present invention, such as Figure 5 As shown, the electronic device may include: a processor (Processor) 501, a communication interface (Communications Interface) 502, a memory (Memory) 503 and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 communicate with each other through the communication bus 504. The processor 501 can call the logic instructions in the memory 503 to execute the abnormal behavior association analysis method for medical insurance data, which includes: based on the medical insurance scenario knowledge graph, according to the target medical treatment path, determine multiple medical insurance nodes to be analyzed that are associated with the target medical insurance node, and obtain the medical treatment paths to be analyzed of each of the medical insurance nodes to be analyzed in the medical insurance scenario knowledge graph; according to the number of medical treatment information, weight the sub-paths in the target medical treatment path and each of the medical treatment paths to be analyzed, and calculate the similarity value between the weighted target medical treatment path and each weighted medical treatment path to be analyzed; through the Top-K query method and the similarity value, select the medical insurance node to be analyzed with a similarity value of TOP-K as the target analysis medical insurance node, and generate the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node; wherein the target medical treatment path is obtained from the medical insurance scenario knowledge graph according to the medical insurance information and node type of the target medical insurance node; the path type of the target medical treatment path and the medical treatment path to be analyzed includes a meta-path and / or a meta-structure.

[0133] In addition, the logic instructions in the above-mentioned memory 503 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0134] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the abnormal behavior association analysis method for medical insurance data provided by the above methods, the method comprising: based on the medical insurance scenario knowledge graph, according to the target medical treatment path, determining multiple medical insurance nodes to be analyzed that are associated with the target medical insurance node, and obtaining the medical treatment path to be analyzed of each of the medical insurance nodes to be analyzed in the medical insurance scenario knowledge graph; according to the number of medical treatments information, respectively analyzing the target medical treatment path and each The sub-paths in the to-be-analyzed medical treatment paths are weighted, and the similarity values ​​between the weighted target medical treatment paths and each weighted medical treatment path to be analyzed are calculated; the to-be-analyzed medical insurance node with the similarity value TOP-K is selected as the target analysis medical insurance node, and according to the co-occurring medical treatment paths between the target medical insurance node and the target analysis medical insurance node, the corresponding medical insurance data association analysis result is generated; wherein, the target medical treatment path is obtained from the medical insurance scenario knowledge graph according to the medical insurance information and node type of the target medical insurance node; the path types of the target medical treatment path and the to-be-analyzed medical treatment path include meta-paths and / or meta-structures.

[0135] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the processor executes the abnormal behavior association analysis method for medical insurance data provided in the above-mentioned embodiments, the method comprising: based on the medical insurance scenario knowledge graph, according to the target medical treatment path, determining multiple medical insurance nodes to be analyzed that are associated with the target medical insurance node, and obtaining the medical treatment path to be analyzed of each of the medical insurance nodes to be analyzed in the medical insurance scenario knowledge graph; according to the number of medical treatment times information, respectively adding the sub-paths in the target medical treatment path and each of the medical treatment paths to be analyzed; Weight processing is performed, and the similarity value between the weighted target medical treatment path and each weighted medical treatment path to be analyzed is calculated; the medical insurance node to be analyzed with the similarity value TOP-K is selected as the target analysis medical insurance node, and according to the co-occurring medical treatment path between the target medical insurance node and the target analysis medical insurance node, the corresponding medical insurance data association analysis result is generated; wherein, the target medical treatment path is obtained from the medical insurance scenario knowledge graph according to the medical insurance information and node type of the target medical insurance node; the path type of the target medical treatment path and the medical treatment path to be analyzed includes meta-path and / or meta-structure.

[0136] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0137] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0138] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for abnormal behavior association analysis of medical insurance data, characterized in that: include: Based on the medical insurance scenario knowledge graph, according to the target medical treatment path, multiple medical insurance nodes to be analyzed that are associated with the target medical insurance node are determined, and the medical treatment paths to be analyzed of each of the medical insurance nodes to be analyzed in the medical insurance scenario knowledge graph are obtained; According to the visit frequency information, weighting is performed on the target visit path and each subpath in the visit path to be analyzed, and the similarity value between the weighted target visit path and each weighted visit path to be analyzed is calculated; Select the to-be-analyzed medical insurance node with the similarity value TOP-K as the target analysis medical insurance node, and generate the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node; The target medical treatment path is obtained from the medical insurance scenario knowledge graph according to the medical insurance information and node type of the target medical insurance node; the path types of the target medical treatment path and the medical treatment path to be analyzed include meta-path and / or meta-structure; The calculating of the similarity value between the weighted target medical treatment pathway and each weighted medical treatment pathway to be analyzed includes: If the path types of the target medical path and the to-be-analyzed medical path are meta-paths, the similarity value is calculated according to a first similarity formula, which is: Among them, S sum (P x→y ) represents the total score of the weighted meta-path P between the medical insurance node x and the medical insurance node y; S sum (P x→x ) represents the sum of the scores of the weighted meta-paths P between medical insurance nodes x and medical insurance nodes x; S sum (P y→y ) represents the sum of the scores of the weighted meta-paths P between the medical insurance node y and the medical insurance node y; Represents each sub-path in the weighted meta-path, R i Represents node A i and node A i+1 The link relationship between them, w i Represents the weight of the i-th link relationship; The calculating of the similarity value between the weighted target medical treatment pathway and each weighted medical treatment pathway to be analyzed includes: If the path type of the target medical treatment path and the medical treatment path to be analyzed is a meta structure, the similarity value is calculated according to the second similarity formula, and the second similarity formula is: Among them, i Represents the i-th instance medical insurance node set N of the weighted meta-structure S i In the instance medical insurance node, {o i+1 |(o i ,o i+1 )∈E} represents the instance medical insurance node o on the i+1th layer i There is an associated instance medical insurance node; Represents the instance medical insurance node o i Go to the instance medical insurance node o i+1 The path weight of Represents each subpath in the weighted metastructure S, M i Represents two groups of nodes N i and N i+1 The link between i Represents the weight of the connection between objects.

2. The abnormal behavior association analysis method for medical insurance data according to claim 1 is characterized in that: After selecting the to-be-analyzed medical insurance node with the similarity value TOP-K as the target analysis medical insurance node, and generating the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node, the method further includes: According to the connection relationship between each medical insurance node in the medical insurance data association analysis result, the topological structure between the medical insurance nodes of the same node type is obtained; According to the attribute information of the medical insurance node of each node type in the medical insurance data association analysis result, construct an attribute coding structure corresponding to the attribute information; Through the community discovery algorithm, the global representation of the nodes in the medical insurance data is learned according to the topological structure and the attribute coding structure, and the medical insurance nodes in the medical insurance data association analysis results are clustered according to the global representation to obtain the medical insurance data association analysis results within the same community.

3. The abnormal behavior association analysis method for medical insurance data according to claim 2 is characterized in that: After selecting the to-be-analyzed medical insurance node with the similarity value TOP-K as the target analysis medical insurance node, and generating the corresponding medical insurance data association analysis result according to the co-occurrence medical treatment path between the target medical insurance node and the target analysis medical insurance node, the method further includes: Select any two medical insurance nodes from the medical insurance data association analysis results; According to the preset query stage path, the association path between any two medical insurance nodes in the medical insurance data association analysis result is obtained through a random walk algorithm.

4. The abnormal behavior association analysis method for medical insurance data according to claim 3 is characterized in that: The method further comprises: Generate a visualization view of similarity exploration of medical insurance nodes based on the medical insurance data association analysis results; Generate a visualization view of the community discovery of medical insurance nodes based on the results of the association analysis of the medical insurance data in the same community; Generate a correlation path exploration visualization view according to the correlation path; The abnormal behavior correlation analysis of medical insurance data is visualized and displayed through the medical insurance node similarity exploration visualization view, the medical insurance node community discovery visualization view and the association path exploration visualization view.

5. An abnormal behavior correlation analysis system for medical insurance data, characterized in that: include: A medical insurance knowledge graph node path acquisition module is used to determine multiple medical insurance nodes to be analyzed that are associated with a target medical insurance node based on the medical insurance scenario knowledge graph and the target medical treatment path, and obtain the medical treatment paths to be analyzed of each of the medical insurance nodes to be analyzed in the medical insurance scenario knowledge graph; A weighted path similarity calculation module is used to perform weighted processing on the target medical path and each subpath in the medical path to be analyzed according to the number of medical visits information, and calculate the similarity value between the weighted target medical path and each weighted medical path to be analyzed; The medical insurance node similarity association analysis result generation module selects the medical insurance nodes to be analyzed with similarity values ​​TOP-K as the target analysis medical insurance nodes, and generates corresponding medical insurance data association analysis results according to the co-occurrence medical treatment paths between the target medical insurance nodes and the target analysis medical insurance nodes; The target medical treatment path is obtained from the medical insurance scenario knowledge graph according to the medical insurance information and node type of the target medical insurance node; the path types of the target medical treatment path and the medical treatment path to be analyzed include meta-path and / or meta-structure; The weighted path similarity calculation module is also used to: If the path types of the target medical path and the to-be-analyzed medical path are meta-paths, the similarity value is calculated according to a first similarity formula, which is: Among them, S sum (P x→y ) represents the total score of the weighted meta-path P between the medical insurance node x and the medical insurance node y; S sum (P x→x ) represents the sum of the scores of the weighted meta-paths P between medical insurance nodes x and medical insurance nodes x; S sum (P y→y ) represents the sum of the scores of the weighted meta-paths P between the medical insurance node y and the medical insurance node y; Represents each sub-path in the weighted meta-path, R i Represents node A i and node A i+1 The link relationship between them, w i Represents the weight of the i-th link relationship; The weighted path similarity calculation module is also used to: If the path type of the target medical treatment path and the medical treatment path to be analyzed is a meta structure, the similarity value is calculated according to the second similarity formula, and the second similarity formula is: Among them, i Represents the i-th instance medical insurance node set N of the weighted meta-structure S i In the instance medical insurance node, {o i+1 |(o i ,o i+1 )∈E} represents the instance medical insurance node o on the i+1th layer i There is an associated instance medical insurance node; Represents the instance medical insurance node o i Go to the instance medical insurance node o i+1 The path weight of Represents each subpath in the weighted metastructure S, M i Represents two groups of nodes N i and N i+1 The link between i Represents the weight of the connection between objects.

6. 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 steps of the abnormal behavior association analysis method for medical insurance data as described in any one of claims 1 to 4 are implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the abnormal behavior association analysis method for medical insurance data as described in any one of claims 1 to 4 are implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the abnormal behavior association analysis method for medical insurance data as described in any one of claims 1 to 4 are implemented.

Images