Search

CN-122022999-A - Cross-domain risk collaborative assessment method and system based on artificial intelligence

CN122022999ACN 122022999 ACN122022999 ACN 122022999ACN-122022999-A

Abstract

The invention relates to the technical field of risk assessment, in particular to a cross-domain risk collaborative assessment method and a system thereof based on artificial intelligence. According to the method, the multi-dimensional modeling capability of account behavior and credit behavior is enhanced by constructing tag behavior feature mapping and implicit structure vectors, the security fusion effect of multi-source tag data in a collaborative process is improved by combining the same latitude structure space mapping and position reservation encryption mechanism, the potential association extraction capability among inter-domain tags is enhanced by cross labeling frequency statistics and tag duty ratio analysis, the mapping sequence construction in a collaborative space is introduced, unique identification management of node behavior features is realized, mapping expression and tracing of tag association relationship are promoted, the comprehensive modeling and dynamic expression capability of cross-domain risks is enhanced, and the accuracy and stability of collaborative evaluation results are improved.

Inventors

  • HE YUANJING
  • CHEN HONGBO

Assignees

  • 中国社会科学院大学

Dates

Publication Date
20260512
Application Date
20260126

Claims (9)

  1. 1. The cross-domain risk collaborative assessment method based on artificial intelligence is characterized by comprising the following steps of: S1, acquiring transaction flow information of a user bank account, executing transaction time ascending arrangement, executing judgment based on a large-amount interval occupation ratio on daily transaction amount, and matching with daily transaction amount to generate an account behavior continuity expression label; S2, acquiring user credit investigation breach records and loan application records, carrying out breach times statistics and application period repeatability calculation, carrying out commodity category preference clustering and daily average consumption balance calculation, and generating a user behavior credit balance judging label; S3, based on the account behavior continuity expression label and the behavior credit balance judging label, mapping label data to a same latitude structure space to form an implicit structure vector, carrying out position preservation encryption, then sending to a collaborative service node for splicing to form evaluation data, and outputting a collaborative mapping fusion expression structure; S4, based on the collaborative mapping fusion expression structure, performing cross labeling counting statistics on the label occupation proportion, extracting label association states when the cross frequency exceeds a set threshold, and establishing a source structure mark by combining a user dimension to generate a label inter-domain coverage association map; And S5, traversing mapping drop point positions of the structural marks in a collaborative space according to the inter-label domain coverage association map, constructing a unique mapping number sequence for all the participating nodes, generating an evaluation structure according to a collaborative sequence, carrying out sealing calibration, and outputting a cross-domain risk collaborative evaluation filing record.
  2. 2. The method for collaborative assessment of risk across domains based on artificial intelligence according to claim 1, wherein the performing comprises calculating a proportion of the number of transactions falling within a 1000-ary interval from the daily transaction number based on the large-forehead interval duty ratio.
  3. 3. The artificial intelligence-based cross-domain risk collaborative assessment method is characterized in that the account behavior continuity representation tag comprises high-frequency large transaction behaviors, transaction time distribution consistency and account fund change stability, the user behavior credit balance discrimination tag comprises abnormal behavior of default frequency, consumption category deviation modes and daily consumption fluctuation characteristics, the collaborative mapping fusion expression structure comprises tag dimension structured mapping characteristics, position keeping encryption index information and inter-institution potential behavior difference characterization, the tag inter-domain coverage association map comprises a tag association strength matrix, user source identification marks and inter-domain tag linkage path structures, and the cross-domain risk collaborative assessment archiving record comprises a collaborative node identification sequence, an assessment flow source identification and archiving data unique numbers.
  4. 4. The artificial intelligence based cross-domain risk collaborative assessment method according to claim 1, wherein the account behavior continuity performance tag obtaining step specifically comprises: S111, acquiring transaction flow information of a user bank account, performing ascending arrangement according to transaction time, combining the ordered time sequence, extracting transaction amount values in daily transaction records as a transaction amount list, calculating a daily large transaction ratio, and constructing a daily large transaction ratio sequence; S112, based on the daily large-scale transaction proportion sequence, reading daily transaction strokes to construct a stroke comparison sequence, comparing and calculating the corresponding daily numerical values in the sequence, executing logic judgment according to a set large-scale judgment threshold, marking the high-scale behavior day sequence index position larger than the set large-scale judgment threshold, and generating a high-duty transaction day index list; S113, according to the high-duty-ratio transaction day index list, carrying out aggregation operation on index positions which continuously appear, carrying out segment coding on continuous day sequences, binding account numbers corresponding to each segment of continuous behaviors, and generating an account behavior continuity representation tag.
  5. 5. The artificial intelligence-based cross-domain risk collaborative assessment method according to claim 1, wherein the user behavior credit balance discrimination tag obtaining step specifically comprises: S211, acquiring user credit investigation violating records and loan application records, arranging the records in ascending order according to the recording time, counting and accumulating the violating behaviors corresponding to each account number by combining the account number and the behavior type corresponding to each behavior in the violating records, counting the maximum value and the minimum value of the application times of each account in a plurality of periods, calculating the difference between the maximum value and the minimum value and taking the difference as a period repeatability index to generate a credit investigation behavior fluctuation factor matrix; s212, extracting commodity classification browsing records and daily consumption amount sequences of a user electronic commerce platform based on the credit investigation behavior fluctuation factor matrix, carrying out clustering mapping on the commodity classification records according to commodity numbers, marking preference categories, synchronously calculating variance values of daily consumption amounts as daily consumption fluctuation indexes, and generating a consumption behavior balance factor set; S213, according to the consumption behavior balance factor set and the credit investigation behavior fluctuation factor matrix, carrying out normalization processing on each behavior index in the two factors and constructing a combined index vector, marking the combined index vector as unbalanced credit behavior if the average value of the combined index vector is smaller than or equal to a set credit balance evaluation threshold value, and summarizing to generate a user behavior credit balance judgment label.
  6. 6. The artificial intelligence-based cross-domain risk collaborative assessment method according to claim 1, wherein the collaborative mapping fusion expression structure obtaining step specifically comprises: s311, based on the account behavior continuity expression label and the user behavior credit balance judging label, performing field alignment operation on the two labels, performing mapping aggregation according to an account number as a main key, encoding label values into one-dimensional vector representations and projecting the one-dimensional vector representations into a same-dimensional structural space, combining each label vector sequence to form a unified account behavior embedding structure, performing aggregation encoding on all account behavior vectors in the structure, and generating an account label mapping vector matrix; s312, based on the account label mapping vector matrix, reading a mechanism attribution field to construct a mechanism distribution vector, executing aggregation processing on an account vector tensor, obtaining an account behavior representation mean value under each mechanism as a mechanism distribution vector, setting a target distribution vector as a reference behavior distribution reference vector, calculating Wasserstein distance between each mechanism distribution and the target distribution as an auxiliary loss participation item, executing position preserving encryption operation on each element value, and uploading to a cooperative service node to generate a cooperative structure embedded representation vector; S313, embedding the expression vectors according to the collaborative structure, calling account number indexes in collaborative service nodes to execute matrix splicing operation, aligning a spliced account embedded vector set with organization attribution structuring, establishing a joint feature expression structure, carrying out coding aggregation processing on all joint feature vectors, and outputting a collaborative mapping fusion expression structure.
  7. 7. The artificial intelligence-based cross-domain risk collaborative assessment method according to claim 1, wherein the step of acquiring the inter-label domain coverage correlation map is specifically as follows: s411, based on the collaborative mapping fusion expression structure, extracting account numbers and corresponding label vector information, analyzing and marking reconstruction according to element content by combining a label vector, aggregating the times of the same label value in an account, executing occupation proportion calculation on independent count values of each label, and counting the combined frequency under a field to generate a label cross frequency count table; S412, based on the label cross frequency count table, reading all label combination frequencies and logarithmic values, screening all label combinations meeting the frequency more than or equal to a set frequency judgment threshold value, recording label names, executing intersection analysis operation on each group of label combinations, extracting record sections with shared account numbers between cross labels, and generating a label association state index set; s413, performing source aggregation processing on the cross records under each label combination according to the label association state index set and combining source platform type identification information corresponding to the account number, and performing serialization coding processing on the connection relation of each edge set by taking a label field as a main index node to generate a label inter-domain coverage association map.
  8. 8. The artificial intelligence-based cross-domain risk collaborative assessment method according to claim 1, wherein the cross-domain risk collaborative assessment archive record obtaining step specifically comprises: s511, extracting label nodes and side set information in a map structure based on the label inter-domain coverage association map, calling mapping records of all nodes in a collaborative space by taking account numbers as main indexes, performing traversal operation on each label node, searching mapping position coordinates in a collaborative vector structure, numbering each mapping position according to the original sequence of the nodes, and performing summarization processing on all structure nodes to generate a node mapping falling point sequence set; S512, based on the node mapping drop point sequence set, combining each mapping number with an account source field to perform combined coding treatment, establishing a global unique identification number system, adopting an account number and structure mapping number splicing format by a numbering structure, sorting and merging the splicing structure according to a collaborative service sequence, constructing a continuous numbering vector sequence, embedding a sorting result into a collaborative account index structure, and generating an evaluation vector structure mapping table; s513, according to the evaluation vector structure mapping table, performing structure sealing calibration operation on each group of account collaborative identifiers, performing archiving coding processing on all sealing structures, outputting a whole structure storage result, and generating a cross-domain risk collaborative evaluation archiving record.
  9. 9. An artificial intelligence based cross-domain risk collaborative assessment system, characterized in that the system is configured to implement the artificial intelligence based cross-domain risk collaborative assessment method according to any one of claims 1-8, comprising: The behavior sequence construction module is used for executing S1, namely acquiring transaction flow information of a user bank account, executing transaction time ascending arrangement, executing judgment based on the large-amount interval occupation ratio on daily transaction amount, and matching with daily transaction number to generate an account behavior continuity expression label; The feature tag fusion module is used for executing S2, namely acquiring user credit investigation breach records and loan application records, carrying out breach times statistics and application period repeatability calculation, carrying out commodity category preference clustering and daily average consumption balance calculation, and generating a user behavior credit balance judging tag; The tag hidden-mapping encryption module is used for executing S3, based on the account behavior continuity expression tag and the behavior credit balance judging tag, mapping tag data to a same latitude structure space to form an hidden structure vector, carrying out position preservation encryption, then sending to a collaborative service node for splicing and forming evaluation data, and outputting a collaborative mapping fusion expression structure; the association graph extraction module is used for executing S4, based on the collaborative mapping fusion expression structure, carrying out cross labeling counting statistics on the label occupation proportion, carrying out label association state extraction on the cross frequency exceeding a set threshold, and establishing a source structure mark in combination with the user dimension to generate a label inter-domain coverage association graph; And S5, the evaluation mapping archiving module is used for constructing a unique mapping number sequence for all the participating nodes according to the label inter-domain coverage association map, traversing the mapping drop point positions of the structure marks in the collaborative space, generating an evaluation structure according to the collaborative sequence, conducting sealing calibration, and outputting a cross-domain risk collaborative evaluation archiving record.

Description

Cross-domain risk collaborative assessment method and system based on artificial intelligence Technical Field The invention relates to the technical field of risk assessment, in particular to a cross-domain risk collaborative assessment method and a system based on artificial intelligence. Background The technical field of risk assessment mainly relates to identification, analysis and quantitative assessment of various risk sources, and provides decision support based on assessment results, including risk identification, risk analysis, risk assessment model construction and application thereof in finance, network security, supply chain management and other scenes, and known or potential risks are quantitatively or qualitatively assessed, and risk control and resource configuration optimization are assisted by means of establishing a risk index system, constructing a mathematical model, applying a statistical method and the like. The traditional cross-domain risk collaborative assessment method refers to risk relevance among multiple fields, and generally relies on a rule base or a fixed weight model established in advance to collect, share and integrate risk information in a unified assessment mode after integrating risk information of each domain through manual experience, static data table analysis or a simple rule model, and performs collaborative analysis and assessment on cross-domain risks in a linear weighting mode, an analytic hierarchy process or a static causal map mode. In the existing cross-domain risk collaborative assessment process, manual experience and static data analysis are mainly relied on, dynamic complexity generated by time variation of risk features is difficult to deal with, a rule base or a weight model is difficult to adapt to nonlinear features of multi-domain risk association, so that the problems of weak fusion capability, information redundancy, low tag coupling degree and the like exist in the multi-source heterogeneous data integration process, potential association among tags is difficult to effectively identify, collaborative assessment results are often influenced by artificial subjective judgment, deep semantic expression capability is lacked, error accumulation is easy to occur in assessment results, risk pre-judgment accuracy and response efficiency are reduced, and requirements of refinement and personalized risk identification and control under multiple scenes are difficult to support. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides a cross-domain risk collaborative assessment method based on artificial intelligence. In order to achieve the purpose, the invention adopts the following technical scheme that the cross-domain risk collaborative assessment method based on artificial intelligence comprises the following steps: S1, acquiring transaction flow information of a user bank account, executing transaction time ascending arrangement, executing judgment based on a large-amount interval occupation ratio on daily transaction amount, and matching with daily transaction amount to generate an account behavior continuity expression label; S2, acquiring user credit investigation breach records and loan application records, carrying out breach times statistics and application period repeatability calculation, carrying out commodity category preference clustering and daily average consumption balance calculation, and generating a user behavior credit balance judging label; S3, based on the account behavior continuity expression label and the behavior credit balance judging label, mapping label data to a same latitude structure space to form an implicit structure vector, carrying out position preservation encryption, then sending to a collaborative service node for splicing to form evaluation data, and outputting a collaborative mapping fusion expression structure; S4, based on the collaborative mapping fusion expression structure, performing cross labeling counting statistics on the label occupation proportion, extracting label association states when the cross frequency exceeds a set threshold, and establishing a source structure mark by combining a user dimension to generate a label inter-domain coverage association map; And S5, traversing mapping drop point positions of the structural marks in a collaborative space according to the inter-label domain coverage association map, constructing a unique mapping number sequence for all the participating nodes, generating an evaluation structure according to a collaborative sequence, carrying out sealing calibration, and outputting a cross-domain risk collaborative evaluation filing record. As a further scheme of the invention, the execution is specifically based on the large-amount interval section duty ratio judgment, and the number proportion of the daily transaction number falling into the interval of more than or equal to 1000 yuan is calculated. As a further scheme of the inve