Method and system for detecting abnormal aggregation of medical insurance cards

Abstract

The application provides a medical insurance card abnormal gathering detection method and system, including: step 1: storing medical insurance case treatment data in a structured manner; step 2: converting the structured medical insurance case treatment data into a graph form by a graph reconstruction and entity relationship representation method, and directly reflecting the relationship between entities; step 3: using a graph algorithm to divide blocks; step 4: analyzing the entity and relationship characteristics in each divided block to identify potential medical insurance card abnormal gathering, and researching the entity behavior, association rules in the block to extract information about abnormal gathering; and step 5: outputting abnormal gathering characteristics. The application directly represents the relationship between entities by using a graph structure, avoids the construction of complex numerical characteristics, and reduces the difficulty and complexity of feature engineering.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine learning and anomaly detection, in particular, to a medical insurance card anomaly gathering detection method and system. BACKGROUND

[0002] Medical insurance fund is the "life-saving money" of the people, but in real life, driven by illegal interests, fraud and insurance fraud occurs from time to time. At present, the illegal and irregular behaviors of medical insurance fund are becoming more and more complex, hidden and dispersed, and it is more and more difficult to find through traditional inspection means.

[0003] In the process of medical insurance management, there are some medical gathering behaviors. The occurrence of medical gathering behavior may be caused by some special disease groups, or there may be fraud. Finding out these groups with gathering behavior characteristics can provide targeted management for special disease groups on the one hand, and effectively improve the supervision of irregular behaviors on the other hand.

[0004] Patent document CN111427926A discloses an abnormal medical insurance group identification method, device, computer equipment and storage medium. Patent document CN112991079A discloses a multi-card co-occurrence medical fraud behavior detection method, system, cloud and medium.

[0005] At present, the detection means for medical insurance card gathering behavior mainly adopts machine learning method. The medical insurance card gathering detection method based on machine learning mainly identifies medical insurance card abnormal gathering through training classifier. These methods use historical data (including normal and abnormal cases) to train the model, so as to learn the potential fraud behavior characteristics. Common machine learning algorithms include decision tree, support vector machine, random forest, logistic regression, etc. Machine learning method usually needs to perform feature engineering on data in order to construct a representative feature set.

[0006] There are some limitations in the method of using machine learning to realize the detection of medical insurance card gathering behavior:

[0007] 1. Difficulty in establishing feature engineering: In machine learning methods, the quality of feature engineering has a great influence on the performance of the model. However, it may require a lot of domain knowledge and experience to construct representative and discriminative features.

[0008] 2. Data imbalance problem: In the medical insurance card gathering detection scenario, abnormal gathering cases are usually much less than normal cases. This may cause the model to overfit to normal cases during training, thereby reducing the recognition ability of abnormal cases.

[0009] 3. Poor model interpretability: Some complex machine learning models (such as neural networks) lack interpretability of detection results, making it difficult to understand and verify the card gathering detection results.

[0010] 4. Adaptability to new frauds: Machine learning methods mainly rely on historical data for training and cannot fully capture new or unknown frauds. SUMMARY

[0011] In view of the defects in the prior art, the purpose of the present application is to provide a medical insurance card abnormal aggregation detection method and system.

[0012] The medical insurance card abnormal aggregation detection method provided by the present application comprises:

[0013] Step 1: Store the medical insurance case visit data in a structured manner. The medical insurance case data is organized in the form of a table of a relational database in structured storage, each field corresponds to a specific attribute, and each record corresponds to a visit case. By setting constraint conditions including primary keys and foreign keys, the integrity and consistency of the data are ensured.

[0014] Step 2: Convert the structured medical insurance case visit data into a graph form by a graph reconstruction and entity relationship representation method, which intuitively reflects the relationship between entities.

[0015] Step 3: Use a graph algorithm to divide the blocks.

[0016] Step 4: For each divided block, analyze the entity and relationship characteristics therein to identify potential medical insurance card abnormal aggregation, and study the entity behavior, association rules in the block to extract information about abnormal aggregation.

[0017] Step 5: Output the abnormal aggregation characteristics.

[0018] Preferably, the step 2 comprises:

[0019] For medical insurance case visit data, first identify different types of entities, including medical insurance visit patients, medical institutions, pharmacies and doctors, represent these entities as nodes in the graph, wherein each node contains attribute information related to the entity, including visit number, diagnosis code, visit time and medical category, and mark the nodes according to the entity type;

[0020] The relationship between entities is represented in the graph, including specific behaviors such as visit, transaction and payment. Add edges in the graph to connect related entity nodes, each edge carries attribute information related to the relationship, including transaction amount and visit time. Mark according to the relationship type to distinguish different types of relationships.

[0021] Through graph reconstruction and entity relationship representation, the medical insurance case visit data is displayed in the form of a graph.

[0022] Preferably, the step 3 comprises:

[0023] The method of relationship subgraph mining is used to divide the blocks according to the multiple co-occurrences of the medical personnel, medical time, and medical institutions, and the characteristics of gathering in a specific institution and prescribing a specific drug;

[0024] The frequently occurring structural patterns are defined, which should reflect the relationships between multiple entities including medical personnel, medical time, and medical institutions;

[0025] Search for subgraphs matching the defined frequent patterns in the medical insurance case medical data graph, including traversing the nodes and edges in the graph to find subgraph instances that meet the pattern; the search process uses optimization strategies, including limiting the search range and pruning;

[0026] Count each found subgraph instance to evaluate the frequency of the pattern in the entire graph, to determine whether it meets the condition of a frequent pattern;

[0027] Cluster or merge the found frequent subgraph instances to form blocks; in the merging process, consider the similarity and overlap between subgraphs to ensure that the entity relationships within the blocks are close and the entity relationships between the blocks are relatively weak.

[0028] Preferably, the step 5 comprises:

[0029] For each identified block, extract features related to abnormal aggregation, including the number of medical personnel, the number of medical institutions, the number of card swipes, the distribution of medical time, and the spatial distribution;

[0030] According to the extracted abnormal features, evaluate each block by setting a threshold to determine its abnormality degree, and the evaluation methods include statistical analysis and machine learning models. The evaluation results are integrated into an abnormal aggregation feature report, including the list of medical personnel in the block, the list of medical institutions, the card swipe records, the abnormal feature values, the abnormality degree evaluation results, the fraud type, and the cause analysis.

[0031] Preferably, the medical insurance case medical data is analyzed using a clustering algorithm, which clusters similar cases into the same category to find abnormal aggregations with similar characteristics;

[0032] Association rule mining is used to find frequent item sets and association rules in the data set; set the support and confidence thresholds to obtain the association rules, and find the rules of abnormal aggregation by mining the frequent item sets in the medical insurance case medical data;

[0033] Anomaly detection algorithm is used to identify abnormal points in the data set; the medical insurance case data is input into the anomaly detection algorithm to identify cases with abnormal characteristics, and then further analyze these abnormal cases to find abnormal aggregations.

[0034] The medical insurance card abnormal aggregation detection system provided by the present invention includes:

[0035] Module M1: Stores medical insurance case visit data in a structured manner. The medical insurance case data is organized into a table format of a relational database in the structured storage. Each field corresponds to a specific attribute, and each record corresponds to one visit case. By setting constraints including primary keys and foreign keys, the integrity and consistency of the data are ensured.

[0036] Module M2: Through graph reconstruction and entity relationship representation, structured medical insurance case visit data is converted into graph form, which intuitively reflects the relationships between entities;

[0037] Module M3: Uses graph algorithms for block partitioning;

[0038] Module M4: For each partitioned block, analyze the entity and relationship features within it to identify potential abnormal clusters of medical insurance cards, and study the entity behavior and association rules in the block to extract information about abnormal clusters;

[0039] Module M5: Outputs abnormal clustering characteristics.

[0040] Preferably, the module M2 includes:

[0041] For medical insurance case visit data, firstly, different types of entities are identified, including medical insurance patients, medical institutions, pharmacies and doctors. These entities are represented as nodes in the graph, where each node contains attribute information related to the entity, including visit number, diagnosis code, visit time and medical category, and the nodes are labeled according to the entity type.

[0042] The graph represents the relationships between entities, including specific behaviors such as medical visits, transactions, and payments; edges are added to the graph to connect related entity nodes, and each edge carries attribute information related to the relationship, including transaction amount and medical visit time; the graph is labeled according to the relationship type to distinguish different types of relationships;

[0043] Through graph reconstruction and entity relationship representation, medical insurance case visit data is presented in graph form.

[0044] Preferably, the module M3 includes:

[0045] The relationship subgraph mining method is used to divide the data into blocks based on the characteristics of multiple co-occurrences of patients, consultation time, and medical institutions, as well as clustering in specific institutions and prescribing specific drugs.

[0046] Define frequently occurring structural patterns that reflect the relationships between multiple entities, including patients, consultation times, and healthcare institutions.

[0047] Searching for subgraphs matching the defined frequent patterns in the medical insurance case visit data graph, including traversing the nodes and edges in the graph to find subgraph instances that match the patterns; the search process uses optimization strategies, including limiting the search range and pruning;

[0048] Counting each found subgraph instance to evaluate the frequency of the pattern in the entire graph, to determine whether it meets the conditions of a frequent pattern;

[0049] Clustering or merging the found frequent subgraph instances to form blocks; in the merging process, the similarity and overlap degree between subgraphs are considered to ensure that the entity relationships within the blocks are close and the entity relationships between the blocks are relatively weak.

[0050] Preferably, the module M5 comprises:

[0051] For each identified block, extract features related to abnormal aggregation, including the number of visitors, the number of medical institutions, the number of card swipes, the distribution of visit time and space;

[0052] According to the extracted abnormal features, evaluate each block by setting a threshold to determine its abnormality degree, the evaluation method includes statistical analysis and machine learning model, and integrate the evaluation results into an abnormal aggregation feature report, including the list of visitors within the block, the list of medical institutions, the card swipe record, the abnormal feature value, the abnormality degree evaluation result, the fraud type and the cause analysis.

[0053] Preferably, the medical insurance case visit data is analyzed using a clustering algorithm, which clusters similar cases into the same category to find abnormal aggregations with similar characteristics;

[0054] Using association rule mining to find frequent item sets and association rules in the data set; set the support and confidence thresholds to obtain the association rules, and find the rules of abnormal aggregation by mining the frequent item sets in the medical insurance case visit data;

[0055] Using anomaly detection algorithm to identify abnormal points in the data set; input the medical insurance case data into the anomaly detection algorithm to identify cases with abnormal characteristics, and then further analyze these abnormal cases to find abnormal aggregations.

[0056] Compared with the prior art, the present application has the following beneficial effects:

[0057] (1) The present application directly represents the relationship between entities using graph structure, avoiding the construction of complex numerical features and reducing the difficulty and complexity of feature engineering;

[0058] (2) Graph algorithms can discover abnormal clusters in the entire data space without being affected by the data imbalance problem;

[0059] (3) The graph structure and algorithm are relatively intuitive and interpretable, which helps to understand the detected abnormal card clusters and related entity relationships;

[0060] (4) The present application can capture new or unknown fraudulent behavior through block division and analysis of same group and spatial clustering characteristics, because it focuses on the relationship and clustering pattern between entities, rather than relying solely on known violation features in historical data. BRIEF DESCRIPTION OF DRAWINGS

[0061] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments, made with reference to the accompanying drawings:

[0062] Figure 1 Flow chart of the medical insurance card abnormal cluster detection method of the present application. DETAILED DESCRIPTION

[0063] The present application will be described in detail below with specific examples. The following examples will help those skilled in the art to further understand the present application, but do not limit the present application in any form. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of changes and improvements can be made. These are within the scope of protection of the present application.

[0064] Example 1

[0065] The present application provides a medical insurance card abnormal cluster detection method, which aims to effectively identify potential medical insurance fraud, thereby improving the efficiency and accuracy of medical insurance management. First, the medical insurance case visit data is structured and stored, including medical insurance cost and other related information. The case data is reconstructed using a graph structure, where nodes represent entities (such as medical insurance card holders, medical service providers, etc.), and edges represent relationships between entities (such as transactions, diagnoses, etc.). Graph algorithms are used to divide the overall medical record data into blocks, making the entity relationships within the blocks more closely related, and the relationships between the blocks relatively weak. Analyze the characteristics of nodes and edges in each block to determine whether the same group and spatial clustering characteristics are present within the block. If medical insurance card abnormal clusters are detected, output the relevant features for further analysis and processing.

[0066] Through the detection of the technical scheme of the present application, the accuracy and efficiency of medical insurance card abnormal cluster detection can be improved to timely detect and prevent potential medical insurance fraud; reduce the difficulty and complexity of feature engineering, reduce the dependence on domain knowledge and experience; enhance the interpretability of the model, making the detection results easier to understand and verify; improve the adaptability to new or unknown fraudulent behavior, making the detection method more robust.

[0067] Specifically comprising the following steps:

[0068] 1. Structured storage of medical insurance case data

[0069] In the present invention, medical insurance case data is stored in a structured manner. This structured storage method makes data easy to manage, query and analyze, providing convenience for subsequent abnormal cluster detection. Medical insurance case data includes but is not limited to the following information: patient medical record code, medical insurance visit card identification, medical institution identification, visit time, diagnosis information, treatment measures information, drug and consumable use information, medical expense information, medical insurance reimbursement amount, etc. These information can be organized into table form of relational database in structured storage, each field corresponds to a specific attribute, and each record corresponds to a visit case. In addition, by setting primary key, foreign key and other constraint conditions, the integrity and consistency of data can be ensured.

[0070] By structuring the medical insurance case data, data query, filtering and sorting can be easily performed, and relevant entities and their relationships can be extracted as needed. This helps to reconstruct case data through graph structure and perform abnormal cluster detection using graph algorithms.

[0071] 2. Graph reconstruction and entity relationship representation

[0072] In the present invention, the structured medical insurance case data is converted into a graph form through the method of graph reconstruction and entity relationship representation. This graph representation can intuitively reflect the relationship between entities, providing strong support for abnormal cluster detection.

[0073] For medical insurance case data, first identify different types of entities, such as medical insurance patients, medical institutions, pharmacies, doctors, etc. These entities are represented as nodes in the graph, where each node contains attribute information related to the entity, such as visit number, diagnosis code, visit time, medical category, etc. In addition, nodes can be labeled according to entity type for easy differentiation and analysis.

[0074] The relationship between entities is represented in the graph. These relationships can be specific behaviors such as visits, transactions, payments, etc. To represent these relationships, edges can be added to the graph to connect related entity nodes. Each edge can carry attribute information related to the relationship, such as transaction amount, visit time, etc. In addition, edges can be labeled according to relationship type to distinguish different types of relationships.

[0075] By graph reconstruction and entity relationship representation, medical insurance case data is presented in the form of a graph. This makes the data more intuitive, easy to analyze, and facilitates the use of graph algorithms for anomaly clustering detection. In subsequent processing, features related to entity relationships can be extracted based on the graph structure to further identify potential medical insurance card anomaly clustering.

[0076] 3. Block division using graph algorithms

[0077] In this invention, block division is performed based on features such as "multiple co-occurrences of patients, visit times, and medical institutions, as well as clustering in specific institutions and issuing specific drugs."

[0078] Definition of frequent patterns: For the characteristics of this invention, first define the frequently occurring structural patterns. These patterns should reflect the relationships between multiple entities such as patients, visit times, and medical institutions. For example, a frequent pattern can represent multiple patients visiting the same medical institution at similar times and being prescribed the same drug.

[0079] Subgraph search: Search for subgraphs in the case data graph that match the defined frequent patterns. This may require traversing the nodes and edges in the graph to find subgraph instances that match the pattern. The search process can employ some optimization strategies, such as limiting the search range, pruning, etc., to improve search efficiency.

[0080] Subgraph counting: Count each found subgraph instance. Counting can be used to assess the frequency of the pattern in the entire graph, thereby determining whether it meets the conditions for a frequent pattern. A threshold can be set to determine whether the pattern is frequent enough.

[0081] Subgraph matching: Cluster or merge the found frequent subgraph instances to form blocks. In the merging process, factors such as the similarity between subgraphs and the degree of overlap can be considered to ensure that the entity relationships within the blocks are tight and the entity relationships between the blocks are relatively weak.

[0082] Result analysis: For each divided block, analyze the entity and relationship characteristics to identify potential medical insurance card anomaly clustering. Further research on entity behavior, association rules, etc. in the block can be conducted to extract information related to anomaly clustering.

[0083] Through the relationship subgraph mining method, subgraphs with specific characteristics in the medical insurance case data graph can be found and divided into blocks. This helps to detect medical insurance card anomaly clustering and thus address medical insurance fraud risks.

[0084] 4. Same group and spatial clustering identification

[0085] A frequent pattern is defined according to given parameters, representing at least 5 different patients visiting at least 3 different medical institutions within a single month, with each patient having at least 5 card swipes. Additionally, the time intervals between visits of these patients must be less than or equal to 15 minutes.

[0086] Search for subgraphs matching the defined frequent pattern in the case data graph. The search process can employ optimization strategies such as limiting the search range, pruning, etc., to improve search efficiency. To simplify the search process, the search can first filter out eligible patients, medical institutions, and visit time nodes, and then search for subgraphs.

[0087] Count the found subgraph instances that match the frequent pattern. These subgraphs represent potential risk groups that meet the given parameters. A threshold can be set to determine whether a subgraph instance is frequent enough to further filter out high-risk subgraphs.

[0088] Cluster or merge the found frequent subgraph instances to form clusters. During the merging process, factors such as the degree of overlap and similarity between subgraphs can be considered to ensure that the entities within the cluster are closely related.

[0089] For each divided cluster, analyze its spatial distribution. If multiple patients visit multiple medical institutions within a short period of time and there is a spatial clustering phenomenon (e.g., visiting adjacent medical institutions), it can be considered that the cluster has a spatial clustering feature.

[0090] Further analyze the blocks with cluster and spatial clustering features to identify potential abnormal aggregation of medical insurance cards. The behavior of entities in the block, association rules, etc. can be studied to extract information about abnormal aggregation.

[0091] 5、Output abnormal aggregation features

[0092] According to the results of relationship subgraph mining, blocks with cluster and spatial clustering features have been identified. These blocks may contain potential abnormal aggregation of medical insurance cards.

[0093] For each identified block, extract features related to abnormal aggregation. These features may include the number of patients, the number of medical institutions, the number of card swipes, the distribution of visit times, and the spatial distribution (such as the distance between medical institutions). Extracting these features helps quantify the degree of abnormal aggregation and possible risks.

[0094] According to the extracted abnormal features, each block is evaluated to determine its abnormality degree. A threshold can be set, and if it exceeds the threshold, it is considered as abnormal aggregation. The evaluation method can include statistical analysis, machine learning model, etc. to obtain more accurate abnormality degree evaluation. The evaluation results are integrated into an abnormal aggregation feature report. Including the list of patients in the block, the list of medical institutions, the card swiping records, etc. Abnormal feature values (such as the number of patients, the number of medical institutions, etc.), abnormality degree evaluation results, possible fraud types and cause analysis.

[0095] Through the above process, abnormal aggregation features can be output and subsequent processing can be performed to deal with medical insurance fraud risks.

[0096] Alternative 1: Abnormal aggregation detection based on clustering algorithm

[0097] In this scheme, clustering algorithms (such as K-means, DBSCAN, etc.) can be used to analyze medical insurance case data. By clustering similar cases into the same category, abnormal aggregation with similar features can be found. This scheme requires appropriate feature extraction and similarity measurement to ensure the accuracy of the clustering results.

[0098] Alternative 2: Abnormal aggregation detection based on association rule mining

[0099] Association rule mining (such as Apriori, FP-Growth, etc.) can be used to find frequent item sets and association rules in a dataset. By mining frequent item sets in medical insurance case data, the rules of abnormal aggregation can be found. This method requires setting appropriate support and confidence thresholds to obtain meaningful association rules.

[0100] Alternative 3: Abnormal aggregation detection based on anomaly detection algorithm

[0101] Anomaly detection algorithms (such as LOF, Isolation Forest, etc.) can be used to identify outliers in a dataset. By inputting medical insurance case data, these algorithms can identify cases with abnormal features. Then, these abnormal cases can be further analyzed to find abnormal aggregation. This method requires appropriate anomaly detection algorithms and parameter settings to ensure the accuracy of the detection results.

[0102] The present invention adopts the method of relationship subgraph mining, combined with specific scenarios and parameters, to analyze medical insurance case data to identify medical insurance card abnormal aggregation. Compared with the prior art, the present invention has the following advantages and effects:

[0103] Efficient anomaly cluster identification: Through the relationship subgraph mining method, the present application can effectively identify blocks with the same group and spatial aggregation characteristics. Compared with traditional rule-based or statistical methods, relationship subgraph mining can better capture complex entity relationships, thereby improving the accuracy of anomaly cluster identification. In addition, relationship subgraph mining can use optimization strategies (such as pruning, search range limitation, etc.) to improve search efficiency and shorten analysis time.

[0104] Targeted parameter setting: The present application sets specific parameters (such as appointment time, number of medical institutions, number of medical personnel, etc.) according to actual application scenarios, making the analysis results more in line with actual needs. This helps to reduce false positives and false negatives, improving the accuracy of medical insurance fraud detection.

[0105] Systematic result analysis: The present application not only identifies abnormal aggregation, but also conducts in-depth analysis, including extracting abnormal features and evaluating abnormality. This makes the results more targeted and operable, helping relevant departments to intervene and handle medical insurance fraud in a timely manner.

[0106] Strong scalability: The relationship subgraph mining method has strong scalability and can adjust the definition and parameter setting of frequent patterns according to different scenarios and needs. This makes the present application capable of dealing with various types of medical insurance fraud, with strong practicality and broad application prospects.

[0107] In summary, the present application uses the relationship subgraph mining method combined with specific scenarios and parameters, which has higher anomaly cluster identification efficiency, more accurate medical insurance fraud detection capability, and stronger scalability compared with existing technologies. These advantages and effects help to more effectively deal with medical insurance fraud risks.

[0108] Example 2

[0109] The present application also provides a medical insurance card abnormal aggregation detection system, which can be realized by executing the process steps of the medical insurance card abnormal aggregation detection method, i.e. the medical insurance card abnormal aggregation detection method can be understood as the preferred embodiment of the medical insurance card abnormal aggregation detection system by those skilled in the art.

[0110] The medical insurance card abnormal clustering detection system provided by the present invention includes: Module M1: storing medical insurance case visit data in a structured manner, the medical insurance case data is organized into a table form of a relational database in the structured storage, each field corresponds to a specific attribute, each record corresponds to one visit case, and the integrity and consistency of the data are ensured by setting constraints including primary keys and foreign keys; Module M2: converting the structured medical insurance case visit data into a graph form through graph reconstruction and entity relation representation methods to intuitively reflect the relationships between entities; Module M3: using graph algorithms to divide the data into blocks; Module M4: analyzing the entity and relation features of each divided block to identify potential medical insurance card abnormal clustering, and studying the entity behavior and association rules in the block to extract information about abnormal clustering; Module M5: outputting abnormal clustering features.

[0111] Module M2 includes: for medical insurance case visit data, firstly identifying different types of entities, including medical insurance patients, medical institutions, pharmacies, and doctors, and representing these entities as nodes in a graph, where each node contains attribute information related to the entity, including visit number, diagnosis code, visit time, and medical category, and labeling the nodes according to the entity type; representing the relationships between entities in the graph, including specific behaviors such as visit, transaction, and payment; adding edges to the graph to connect related entity nodes, with each edge carrying attribute information related to the relationship, including transaction amount and visit time; labeling according to the relationship type to distinguish different types of relationships; and through graph reconstruction and entity relationship representation, the medical insurance case visit data is presented in graph form.

[0112] Module M3 includes: employing a relational subgraph mining method to divide blocks based on the multiple co-occurrences of patients, consultation times, and medical institutions, as well as the characteristics of clustering at specific institutions and prescribing specific drugs; defining frequently occurring structural patterns that reflect the relationships between multiple entities, including patients, consultation times, and medical institutions; searching the medical insurance case consultation data graph for subgraphs that match the defined frequent patterns, including traversing nodes and edges in the graph to find subgraph instances that conform to the pattern; employing optimization strategies in the search process, including limiting the search scope and pruning; counting each found subgraph instance to evaluate the frequency of the pattern in the entire graph, thereby determining whether it meets the conditions for a frequent pattern; clustering or merging the found frequent subgraph instances to form blocks; during the merging process, considering factors such as similarity and overlap between subgraphs to ensure that the entity relationships within a block are close and the entity relationships between blocks are relatively weak.

[0113] The module M5 comprises: for each identified block, extracting features related to abnormal aggregation, including the number of medical personnel, the number of medical institutions, the number of card swiping times, the distribution of medical treatment time and space; according to the extracted abnormal features, each block is evaluated, the abnormal degree is determined by setting a threshold, the evaluation method includes statistical analysis and machine learning model, and the evaluation result is integrated into an abnormal aggregation feature report, including the medical personnel list, the medical institution list, the card swiping record, the abnormal feature value, the abnormal degree evaluation result, the fraud type and the reason analysis in the block.

[0114] The medical insurance case medical treatment data is analyzed using a clustering algorithm, similar cases are clustered into the same category, and abnormal aggregation with similar characteristics is found; association rule mining is used to find frequent item sets and association rules in the data set; support and confidence thresholds are set to obtain association rules, and the rules of abnormal aggregation are found by mining the frequent item sets in the medical insurance case medical treatment data; an anomaly detection algorithm is used to identify abnormal points in the data set; the medical insurance case data is input, and the abnormal cases with abnormal characteristics are identified by the anomaly detection algorithm, and then the abnormal cases are further analyzed to find abnormal aggregation.

[0115] Those skilled in the art know that, in addition to implementing the system, device and each module thereof provided by the present application in a pure computer readable program code manner, the same program can be realized in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and each module thereof provided by the present application can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing methods and structures within hardware components.

[0116] The specific embodiments of the present application are described above. It should be understood that the present application is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essential content of the present application. The embodiments of the present application and the features in the embodiments can be arbitrarily combined with each other without conflict.

Claims

1. A method of detecting abnormal aggregation of medical insurance cards, characterized by, The method comprises the following steps: Step 1: Store the medical insurance case visit data in a structured manner. The medical insurance case data is organized in the form of a table of a relational database in the structured storage. Each field corresponds to a specific attribute, and each record corresponds to a visit case. By setting constraint conditions including primary keys and foreign keys, the data integrity and consistency are ensured; Step 2: Convert the structured medical insurance case visit data into a graph form by using a graph reconstruction and entity relationship representation method, so as to directly reflect the relationship between entities; Step 3: Divide the graph into blocks by using a graph algorithm; Step 4: For each divided block, analyze the entity and relationship characteristics in the block to identify potential abnormal aggregation of medical insurance cards, and study the entity behavior and association rules in the block to extract information about the abnormal aggregation; Step 5: Output the abnormal aggregation characteristics. The step 3 comprises: A relationship subgraph mining method is used to divide the blocks according to the characteristics of the multiple co-occurrences of the visit personnel, the visit time and the medical institutions, and the aggregation in a specific institution and the prescription of a specific drug; Frequent structure patterns are defined. These patterns should reflect the relationship between multiple entities including the visit personnel, the visit time and the medical institutions; The subgraphs matching the defined frequent patterns are searched in the medical insurance case visit data graph, including traversing the nodes and edges in the graph to find the subgraph instances conforming to the patterns. The search process uses optimization strategies including limiting the search range and pruning; Each found subgraph instance is counted to evaluate the frequency of the pattern in the entire graph, so as to determine whether the condition of the frequent pattern is met; The found frequent subgraph instances are clustered or merged to form blocks. In the merging process, the similarity and overlap degree between the subgraphs are considered to ensure that the entity relationship in the block is close and the entity relationship between the blocks is relatively weak.

2. The method of claim 1, wherein the medical insurance card abnormal aggregation detection method is characterized by, The step 2 comprises: For the medical insurance case visit data, first, different types of entities are identified, including medical insurance patients, medical institutions, pharmacies and doctors. These entities are represented as nodes in the graph, wherein each node contains attribute information related to the entity, including the visit number, the diagnosis code, the visit time and the medical category. The nodes are labeled according to the entity type; The relationship between the entities is represented in the graph, including specific behaviors such as visit, transaction and payment. Edges are added in the graph to connect the related entity nodes. Each edge carries attribute information related to the relationship, including the transaction amount and the visit time. The edges are labeled according to the relationship type to distinguish different types of relationships; Through the graph reconstruction and entity relationship representation, the medical insurance case visit data is displayed in the form of a graph.

3. The method of claim 1, wherein the medical card abnormal aggregation detection method is characterized by, The step 5 comprises: For each identified block, the features related to the abnormal aggregation are extracted, including the number of visit personnel, the number of medical institutions, the number of card swiping times, the visit time distribution and the spatial distribution. According to the extracted abnormal features, each block is evaluated to determine its abnormality degree by setting a threshold, and the evaluation method includes statistical analysis and machine learning model. The evaluation results are integrated into an abnormal aggregation feature report, including the list of patients in the block, the list of medical institutions, the card swiping records, the abnormal feature value, the abnormality degree evaluation result, the fraud type and the cause analysis.

4. The method of claim 1, wherein the medical card abnormal aggregation detection method is characterized by, Using clustering algorithm to analyze medical insurance case visit data, by clustering similar cases into the same category, to find abnormal aggregation with similar characteristics; Using association rule mining to find frequent item sets and association rules in the data set; set the support and confidence thresholds to obtain the association rules, and find the rules of abnormal aggregation by mining the frequent item sets in the medical insurance case visit data; Using anomaly detection algorithm to identify abnormal points in the data set; taking medical insurance case data as input, the anomaly detection algorithm is used to identify cases with abnormal features, and then further analyze these abnormal cases to find abnormal aggregation.

5. A system for detecting abnormal aggregation of medical insurance cards, characterized by, It includes: Module M1: Store medical insurance case visit data in a structured manner. Medical insurance case data is organized into table form in relational database in structured storage. Each field corresponds to a specific attribute, and each record corresponds to a visit case. By setting constraints including primary key and foreign key, the integrity and consistency of the data are ensured; Module M2: Convert structured medical insurance case visit data into graph form by graph reconstruction and entity relationship representation method, which directly reflects the relationship between entities; Module M3: Use graph algorithm for block division; Module M4: Analyze the entity and relationship characteristics in each divided block to identify potential medical insurance card abnormal aggregation, and study the entity behavior, association rules in the block to extract information about abnormal aggregation; Module M5: Output abnormal aggregation features; The module M3 includes: Using the method of relationship subgraph mining, the block division is carried out according to the characteristics of multiple common occurrence of patients, visit time and medical institutions, and aggregation in specific institutions and issuance of specific drugs; Define frequent structure patterns, which should reflect the relationship between multiple entities including patients, visit time and medical institutions; Search for subgraphs matching the defined frequent patterns in the medical insurance case visit data graph, including traversing the nodes and edges in the graph to find subgraph instances that meet the pattern; the search process uses optimization strategies, including limiting the search range and pruning; Count each found subgraph instance to evaluate the frequency of the pattern in the entire graph, to determine whether it meets the condition of frequent pattern; Cluster or merge the found frequent subgraph instances to form blocks; in the merging process, consider the similarity and overlap degree between subgraphs to ensure that the entity relationship within the block is close and the entity relationship between blocks is relatively weak.

6. The system for detecting abnormal aggregation of medical care cards according to claim 5, wherein The module M2 includes: For medical insurance case visit data, first identify different types of entities, including medical insurance patients, medical institutions, pharmacies and doctors, represent these entities as nodes in the graph, where each node contains attribute information related to the entity, including visit number, diagnosis code, visit time and medical category, and label the nodes according to the entity type; Represent the relationship between entities in the graph, including specific behaviors such as visits, transactions and payments; add edges to the graph to connect related entity nodes, each edge carrying attribute information related to the relationship, including transaction amount and visit time; label according to relationship type to distinguish different types of relationships; Through graph reconstruction and entity relationship representation, medical insurance case visit data is presented in the form of a graph.

7. The system for detecting abnormal aggregation of medical care cards according to claim 5, wherein The module M5 comprises: For each identified block, extract features related to abnormal aggregation, including the number of visitors, the number of medical institutions, the number of card swipes, the distribution of visit time and space; According to the extracted abnormal features, evaluate each block by setting a threshold to determine its abnormality, the evaluation methods include statistical analysis and machine learning model, integrate the evaluation results into abnormal aggregation feature report, including the list of visitors in the block, the list of medical institutions, the card swiping records, the abnormal feature value, the abnormality evaluation result, the fraud type and the cause analysis.

8. The system for detecting abnormal aggregation of medical care cards according to claim 5, wherein Use clustering algorithm to analyze medical insurance case visit data, by clustering similar cases into the same category, find abnormal aggregation with similar characteristics; Use association rule mining to find frequent item sets and association rules in the data set; set the support and confidence thresholds to obtain the association rules, by mining the frequent item sets in the medical insurance case visit data, find the rules of abnormal aggregation; Use anomaly detection algorithm to identify abnormal points in the data set; input the medical insurance case data into the anomaly detection algorithm to identify cases with abnormal characteristics, and then further analyze these abnormal cases to find abnormal aggregation.

Images