CN-121996714-A - Supervision object graph analysis method and device oriented to complex relation mining

CN121996714ACN 121996714 ACN121996714 ACN 121996714ACN-121996714-A

Abstract

The application provides a supervision object graph analysis method and device for complex relation mining, and relates to the field of machine learning. The method comprises the steps of obtaining member basic attribute data of a target supervision object in a preset business data database, mining a knowledge graph through pre-established relation semantics based on the member basic attribute data, obtaining abnormal resource use structural features corresponding to the target supervision object, inputting the abnormal resource use structural features into an abnormal association evaluation model, outputting a resource use mode deviation score corresponding to the target supervision object, and outputting a risk identification result corresponding to the target supervision object based on the resource use mode deviation score. The application solves the problems that the existing supervision and analysis computing platform can easily misjudge the conditions of normal centralized traveling, centralized consumption, centralized approval or centralized cooperation as abnormal conditions only according to the surface statistics result, thereby generating a large number of misinformation and leading the supervision and analysis result of the computer system to have low reliability.

Inventors

ZHOU JINLIANG
HUO YITAO
ZHOU JING
YE SONG
YAN YONGTAO
ZHANG HAIBAO
CHENG HUAN

Assignees

武汉爱迪科技股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (10)

1. A supervised object graph analysis method for complex relation mining, the method comprising: Acquiring member basic attribute data of a target supervision object from a preset service data database; Based on the member basic attribute data, acquiring abnormal resource use structural features corresponding to the target supervision object through pre-constructed relation semantic mining knowledge graph, wherein the abnormal resource use structural features comprise business expenditure associated object features, resource allocation circulation path features and resource use time distribution features; inputting the abnormal resource use structural characteristics into an abnormal association evaluation model, and outputting a resource use mode deviation score corresponding to the target supervision object; And outputting a risk identification result corresponding to the target supervision object based on the resource use mode deviation score.
2. The method according to claim 1, wherein constructing the preset service data database specifically comprises: The method comprises the steps of uniformly collecting original business data from a plurality of business systems, classifying the original business data based on predefined data topics to form an original data set covering different business dimensions, wherein the data topics comprise member main subject topics, resource use topics and relationship interaction topics; Performing semantic standardization processing on data fields in the original data set; performing association integration on the standardized data records in the original data set, and constructing a multi-table association data view taking the target supervision object as a core; and executing data integrity check, outlier removal and time sequence alignment processing on the multi-table associated data view, and storing the multi-table associated data view into the preset service data database.
3. The method according to claim 2, wherein constructing the relational semantic mining knowledge graph specifically comprises: based on the multi-table associated data view, semantic mapping is carried out on the member main subject, the resource using subject and the data records corresponding to the relation interaction subject; constructing entity nodes in the relation semantic mining knowledge graph based on the data records after semantic mapping; According to the relation semantic relation of different entity nodes in the same resource usage record or relation interaction record, constructing semantic association edges among the entity nodes in the relation semantic mining knowledge graph; and constructing the relation semantic mining knowledge graph based on the entity nodes and the semantic association edges.
4. The method according to claim 3, wherein the obtaining, based on the member base attribute data, the abnormal resource usage structure feature corresponding to the target supervision object through pre-constructed relation semantic mining knowledge graph specifically includes: Acquiring a first entity node in the relation semantic mining knowledge graph, wherein the first entity node and the target supervision object have a resource use co-occurrence association relation in the resource use record; extracting the business expenditure associated object features through an associated concentration analysis algorithm based on the first entity node; Acquiring a second entity node in the relation semantic mining knowledge graph, wherein a resource circulation reachable association relation exists between the second entity node and the target supervision object; Extracting the resource allocation circulation path characteristics through path traversal and path complexity analysis algorithm based on the second entity node; Acquiring a third entity node in the relation semantic mining knowledge graph, wherein a time-stamp association relation exists between the third entity node and the target supervision object; And extracting the resource use time distribution characteristics through a time distribution statistics and time sequence aggregation analysis algorithm based on the third entity node.
5. The method according to claim 2, wherein constructing the anomaly association assessment model specifically comprises: extracting an abnormal resource use structure feature sample based on the multi-table associated data view in the preset service data database; Constructing an abnormal association evaluation training set according to the abnormal resource use structure feature sample, and calculating a resource use mode corresponding to the abnormal resource use structure feature sample, wherein the resource use mode is a reference mode corresponding to the target supervision object in a normal resource use state; Constructing a deviation loss function based on the deviation value; and carrying out iterative optimization on model parameters based on the abnormal association evaluation training set and the deviation loss function, and constructing the abnormal association evaluation model.
6. The method of claim 1, wherein the inputting the abnormal resource usage structural feature into an abnormal association assessment model and outputting a resource usage pattern deviation score corresponding to the target supervision object: respectively constructing vector representations corresponding to the business expenditure associated object features, the resource allocation circulation path features and the resource use time distribution features; And cross-feature interaction fusion is conducted on the vector representation based on the abnormal association evaluation model, and the resource use pattern deviation score is calculated based on the fused vector representation.
7. The method according to claim 1, wherein the outputting the risk identification result corresponding to the target supervision object based on the resource usage pattern deviation score specifically includes: If the resource use mode deviation score is confirmed to be in the first interval, outputting the risk identification result to be free of abnormal use risks; If the resource use mode deviation score is confirmed to be in the second interval, outputting the risk identification result as a low-consistency abnormal use risk; If the deviation score of the resource use mode is confirmed to be in a third interval, outputting a risk identification result to be high-consistency abnormal use risk, wherein the first interval, the second interval and the third interval are adjacent intervals and are sequentially arranged from small to large; When the resource use mode deviation score is confirmed to be in a second interval, carrying out preset consistency verification on the risk identification result based on the abnormal resource use structural characteristics; And if the verification result of the preset consistency verification meets consistency convergence, changing the risk identification result into the high-consistency abnormal use risk.
8. A supervision object graph analysis device facing complex relation mining is characterized in that the device comprises an acquisition module and a processing module, wherein, The acquisition module is used for acquiring member basic attribute data of a target supervision object from a preset service data database, and acquiring abnormal resource use structure characteristics corresponding to the target supervision object through a pre-constructed relation semantic mining knowledge graph based on the member basic attribute data, wherein the abnormal resource use structure characteristics comprise service expenditure associated object characteristics, resource allocation circulation path characteristics and resource use time distribution characteristics; The processing module is used for inputting the abnormal resource use structural characteristics into an abnormal association evaluation model, outputting a resource use mode deviation score corresponding to the target supervision object, and outputting a risk identification result corresponding to the target supervision object based on the resource use mode deviation score.
9. An electronic device comprising a processor, a communication bus, a user interface, a network interface, and a memory, the memory for storing instructions, the user interface and the network interface for communicating to other devices, the processor for executing instructions stored in the memory to cause the electronic device to perform the method of any of claims 1-7.
10. A non-transitory computer readable storage medium storing instructions which, when executed, perform the method of any one of claims 1 to 7.

Description

Supervision object graph analysis method and device oriented to complex relation mining Technical Field The application relates to the field of machine learning, in particular to a supervision object graph analysis method and device for complex relation mining. Background With the expansion of organization scale and the complexity of business systems, data generated in the organization operation process has a development trend of multi-source, isomerization and association. The data analysis technology surrounding the organization member behaviors, resource allocation and business cooperation gradually evolves from an early statistical analysis mode with a single system and a single index as cores to an analysis direction integrating multi-business system data, emphasizing a relation structure and overall pattern recognition. When the analysis is carried out on the basis of a rule-driven supervision and analysis computing platform, the judgment is usually carried out around a local statistical result or a fixed judgment rule formed in a single information system, such as the statistics of times, the statistics of amounts or the analysis of simple comparison relations formed around a single financial system, a public service travel system or a matter declaration system, etc., the method can play a certain role in the scenes of single data source, smaller personnel scale and relatively stable behavior mode, but in the actual application environment represented by a large organization or a public service class organization, the staff's performance, association activities and asset variation are often recorded in a plurality of mutually independent information systems in a scattered manner, the different identities, different roles and related personnel relations of the same natural person are lack of a unified association expression mode, and the traditional supervision and analysis computing platform is easy to misjudge the conditions of normal centralized travel, centralized consumption, centralized approval or centralized approval, etc. as abnormal situations only according to the surface statistical results, so that a large number of misinformation is generated, and the supervision and analysis result of the computer system has the problem of low reliability. Therefore, there is a need for a method and apparatus for supervised object atlas analysis for complex relational mining. Disclosure of Invention The application provides a complex relation mining-oriented supervision object graph analysis method and device, which solve the problem that the existing supervision analysis computing platform is easy to misjudge the conditions of normal centralized travel, centralized consumption, centralized approval or centralized cooperation as abnormal conditions only according to surface statistics results, so that a large number of false positives are generated, and the supervision analysis results of a computer system have low reliability. The application provides a supervision object map analysis method for complex relation mining, which comprises the steps of obtaining member basic attribute data of a target supervision object in a preset business data database, mining a knowledge map through pre-constructed relation semantics based on the member basic attribute data, obtaining abnormal resource use structure characteristics corresponding to the target supervision object, wherein the abnormal resource use structure characteristics comprise business expenditure associated object characteristics, resource allocation circulation path characteristics and resource use time distribution characteristics, inputting the abnormal resource use structure characteristics into an abnormal association evaluation model, outputting a resource use mode deviation score corresponding to the target supervision object, and outputting a risk identification result corresponding to the target supervision object based on the resource use mode deviation score. The method comprises the steps of establishing a preset business data database, specifically comprising the steps of uniformly collecting original business data from a plurality of business systems, classifying the original business data based on predefined data topics to form an original data set covering different business dimensions, wherein the data topics comprise member main body topics, resource use topics and relationship interaction topics, performing semantic standardization processing on data fields in the original data set, performing association integration on data records in the standardized original data set, establishing a multi-table association data view taking a target supervision object as a core, performing data integrity verification, outlier rejection and time sequence alignment processing on the multi-table association data view, and storing the multi-table association data view in the preset business data database. The method comprises the steps of establishi