Method and system for identifying dynamic surrounding string bidding gang based on bidder association network

Abstract

The application relates to the technical field of electronic bidding supervision, and particularly discloses a dynamic surrounding string bidding gang identification method and system based on a bidder correlation network, which integrates bidding business data and external correlation data to complete bidder entity normalization and evidence event construction; candidate correlation edges are established based on multiple types of correlation characteristics, and the correlation edges are screened through a correlation edge conflict resolution model to obtain credible correlation relationships; the comprehensive correlation strength of the bidders is calculated by combining evidence weight, credibility and time decay mechanism, time sequence correlation edges and a dynamic correlation network are constructed; incremental updating is performed on new projects, time smoothing constraints are introduced to perform dynamic community discovery, continuously evolving candidate gangs are identified, the risk level is calculated by combining state characteristics, and a gang risk report is output. The application can convert scattered bidding information into a dynamic correlation network, continuously track the relationship changes between bidding subjects in the time dimension, and improve the accuracy of surrounding string bidding behavior discovery.

Description

Technical Field

[0001] This invention relates to the field of electronic bidding supervision technology, and more specifically, to a method and system for identifying dynamic collusion groups based on bidder association networks. Background Technology

[0002] With the rapid development of electronic bidding, bidding activities have gradually shifted from offline transactions to digital transactions based on internet platforms, resulting in a continuous accumulation of massive amounts of bidding data. To circumvent regulations, some bidders are making it difficult for traditional detection methods to promptly identify potential bid-rigging or collusion. Currently, existing technologies typically determine whether there are abnormal relationships between bidders by setting fixed rules or performing similarity analysis on single bidding data. However, this method relies on pre-set rules and struggles to integrate the complex relationships between multi-source, heterogeneous data. Furthermore, normal business cooperation, industry collaboration, and internal group coordination can also generate similar correlation characteristics, easily leading to false alarms.

[0003] Existing graph-based bidding relationship analysis methods typically treat enterprises as nodes and the relationships between enterprises as fixed edges, identifying anomalous groups through static network analysis. However, actual bidding relationships exhibit significant temporal evolution; the same group of bidders may experience relationship formation, strengthening, stabilization, weakening, or even disappearance at different times. Fixed-linkage networks cannot accurately reflect the dynamic process of bidding relationships changing over time. Furthermore, existing methods often directly calculate weights based on single or multiple types of association evidence when constructing enterprise association edges, failing to adequately consider potential conflicts between different pieces of evidence. If the credibility of different evidence cannot be distinguished and the impact of conflicts eliminated, false association edges are easily included in subsequent analysis, reducing the accuracy of group identification. Simultaneously, in the construction of dynamic association networks, existing technologies lack effective analytical mechanisms for analyzing trends in relationship changes. They cannot determine whether certain association edges are rapidly strengthening or identify dormant relationships that have not had effective connections for a long time, resulting in a large number of historically invalid connections in the dynamic graph, affecting subsequent community discovery and risk assessment results.

[0004] Therefore, it is necessary to provide a method and system for identifying dynamic bid-rigging groups based on the bidder's network of connections to solve the above-mentioned technical problems. In order to solve the above problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, this invention provides a dynamic method and system for identifying bid-rigging groups based on bidder association networks. This method addresses the problems of existing bid-rigging identification methods failing to integrate multi-source heterogeneous association evidence and accurately characterize the implicit associations between bidders over time, leading to misjudgments of associations, inability to identify dynamic group evolution processes, and untimely detection of potential bid-rigging groups.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A dynamic method for identifying bid-rigging groups based on bidder networks includes the following steps: By accessing the bidding and tendering business system and external data sources, bidding and tendering business data and external related data are obtained respectively, and field standardization processing is performed. By combining deterministic rules and similarity models, normalization processing is performed on entities in bidding business data and external related data, and the relationships between entities are converted into evidentiary events. Based on the evidence events, the relationship characteristics between bidders are extracted, and basic weights, confidence calculation methods, time decay parameters and false alarm deduction conditions are configured for each relationship characteristic to form a candidate set of related edges; For any two bidders, a conflict resolution model for the candidate association edges generated between them is established. Consistency analysis is performed on the evidence supporting the association and the evidence weakening the association to screen credible association edges. Based on the association characteristics of credible association edges, the comprehensive association strength between any two bidders is calculated, time-series association edges are generated, the weight increment of time-series association edges is calculated, enhanced edges and dormant edges are identified, and a dynamic association network is constructed. Based on the newly added project data, locate the affected nodes and related edges in the dynamic association network, and perform incremental updates for the affected nodes and related edges; By introducing time smoothing constraints to perform dynamic community discovery, candidate overlapping groups in dynamic association networks are identified, state components of candidate overlapping groups are obtained, a group risk scoring model is constructed based on the state components to calculate the group risk score, and the risk level is determined based on the group risk score. By identifying multiple types of target members among candidate overlapping gangs and recording gang evolution events, a gang risk report is generated after summarizing the data.

[0007] As a further aspect of the present invention, the relationship features include project co-occurrence features, shared personnel features, equipment network features, document similarity features, financial guarantee features, bidding behavior features, winning bid rotation features, static relationship features, and abnormal collaborative behavior features.

[0008] As a further aspect of the present invention, for candidate association edges generated between any two bidders, an association edge conflict resolution model is established. Consistency analysis is performed on evidence supporting the association relationship and evidence weakening the association relationship to screen credible association edges. The specific steps are as follows: Based on the type of evidence, the evidence events corresponding to the candidate related edges are divided into supporting evidence sets and conflicting evidence sets. The number of pieces of evidence, their weight, occurrence time, and credibility are statistically analyzed for each type of evidence in the supporting evidence set and the conflicting evidence set. The strength of the supporting association and the strength of the conflicting association are then calculated. Trustworthy association edges are selected based on the strength of supporting association and the strength of conflicting association. Specifically, the strength difference between supporting association and conflicting association is calculated. When the strength difference is greater than a preset difference threshold, the corresponding candidate association edge is marked as a trustworthy association edge. Otherwise, the corresponding candidate association edge is marked as an association edge to be manually reviewed.

[0009] As a further aspect of the present invention, the screening of credible association edges specifically involves: calculating the strength difference between the supporting association strength and the conflicting association strength; when the strength difference is greater than a preset difference threshold, the corresponding candidate association edge is marked as a credible association edge; otherwise, the corresponding candidate association edge is marked as an association edge to be manually reviewed.

[0010] As a further aspect of the present invention, based on the association characteristics of the trusted association edges, the comprehensive association strength between any two bidders is calculated to generate time-series association edges. The specific steps are as follows: By taking the bidders as core nodes, the comprehensive correlation strength between any two bidder nodes is calculated. Calculate edge weight increments based on the comprehensive correlation strength between adjacent time points; A dynamic association network is constructed based on the identification of enhanced and dormant edges using edge weight increments. Specifically, if the edge weight increment calculated by the association strength is greater than or equal to a preset increment threshold, the corresponding edge is marked as an enhanced edge; if the edge weight increment calculated by the association strength is less than the preset increment threshold, the corresponding edge is marked as a dormant edge.

[0011] As a further aspect of the present invention, dynamic community discovery is performed by introducing time smoothing constraints to identify candidate overlapping groups in dynamic association networks. The specific steps are as follows: Candidate edges are selected based on edge weight threshold, number of strong evidence, number of common projects, and recent enhancements to form a candidate association subgraph; Dynamic community discovery is performed by introducing time smoothing constraints into the candidate subgraph, resulting in the current set of community partitions. In the formula: Let be the k-th community at time t. Divide the communities into numbers; The community member overlap index is calculated for each community based on adjacent time windows, and the community evolution status is judged by comprehensively judging the community member overlap index.

[0012] As a further aspect of the present invention, the community member overlap index includes member overlap rate, core member overlap rate, and edge evidence overlap rate; the community evolution state includes community expansion, contraction, merger, and community splitting and disappearance.

[0013] As a further aspect of the present invention, the state components of candidate overlapping groups are obtained, a group risk scoring model is constructed based on the state components to calculate the group risk score, and the risk level is determined based on the group risk score. The specific steps are as follows: Obtain the state components of candidate overlapping groups, construct a group risk scoring model based on the state components, and calculate the group risk score. The calculation formula is as follows: ; In the formula: Assess the risk of the group. The weight coefficients are the weights corresponding to the p-th state component. For the p-th state component, The weighting coefficient corresponding to the deduction value for normal cooperation. This is a deduction for normal cooperation. The number of state components excluding normal cooperation deduction values; The risk level is determined based on the group's risk score, which includes Level 1 risk, Level 2 risk, and Level 3 risk.

[0014] The dynamic collusion and bid-rigging gang identification system based on the bidder association network includes a data acquisition and standardization module, an entity normalization module, a heterogeneous association feature extraction module, an association edge conflict resolution module, a dynamic association network generation module, a network dynamic evolution analysis module, a dynamic community discovery module, a gang risk scoring module, a gang mining and evolution tracking module, an evidence chain generation and visualization output module, and a feedback calibration module. The data acquisition and standardization module is used to acquire bidding business data and external related data by accessing the bidding business system and external data sources, and to perform field standardization processing; the bidding business data includes project data, bidder data, bidding behavior data, document evidence data, equipment and network data, and financial or guarantee data; The entity normalization module is used to perform normalization processing on entities in bidding business data and external related data by combining deterministic rules and similarity models, and to convert the relationships between entities into evidence events. The heterogeneous association feature extraction module is used to extract the relationship features between bidders based on evidence events, and to configure basic weights, confidence calculation methods, time decay parameters and false alarm deduction conditions for each relationship feature to form a candidate association edge set. The associated edge conflict resolution module is used to establish an associated edge conflict resolution model for candidate associated edges generated between any two bidders, perform consistency analysis on the evidence supporting the association relationship and the evidence weakening the association relationship, and screen credible associated edges. The dynamic association network generation module is used to calculate the comprehensive association strength between any two bidders based on the association characteristics of credible association edges, generate time-series association edges, calculate the weight increment of time-series association edges, identify enhanced edges and dormant edges, and construct a dynamic association network. The network dynamic evolution analysis module is used to locate the affected nodes and associated edges in the dynamic network based on the data of newly added projects, and to perform incremental updates on the affected nodes and associated edges. The dynamic community detection module is used to perform dynamic community detection by introducing time smoothing constraints, and to identify candidate overlapping groups in dynamic association networks; The gang risk scoring module is used to obtain the state components of candidate overlapping gangs, build a gang risk scoring model based on the state components to calculate the gang risk score, and determine the risk level based on the gang risk score; The gang discovery and evolution tracking module is used to identify multiple types of target members in candidate overlapping gangs and record gang evolution events; The evidence chain generation and visualization output module is used to generate gang risk reports; The feedback calibration module is used to receive and update gang risk reports.

[0015] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of the dynamic bid-rigging gang identification method based on a bidder association network as described above.

[0016] The technical effects and advantages of this invention, based on a dynamic bid-rigging group identification method and system using a bidder association network, are as follows: By constructing a time-evolving dynamic bidder association network, this invention achieves continuous tracking of implicit relationships across projects and entities. Compared to traditional static rule-based detection methods, it can effectively detect bid-rigging groups that gradually form or hide and develop. Through association evidence credibility calculation and conflict resolution mechanisms, it reduces false alarms caused by normal business cooperation and public resource sharing, improving the accuracy of association identification. Through enhanced edge and dormant edge identification and incremental update mechanisms, it improves the efficiency of large-scale bidding data processing. Through dynamic community evolution analysis and risk scoring, it achieves interpretable identification of the formation, expansion, and changes of these groups, providing reliable data for regulatory decisions. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the dynamic bid-rigging gang identification method based on a bidder association network provided in this embodiment of the invention; Figure 2 This is a system block diagram of a dynamic bid-rigging gang identification system based on a bidder association network provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described technical solutions are only a part of this invention, and not all of it. All other technical solutions obtained by those skilled in the art based on the technical solutions of this invention without inventive effort are within the scope of protection of this invention.

[0019] Example 1: As Figure 1 The diagram shown is a flowchart of a dynamic bid-rigging gang identification method based on a bidder association network provided in an embodiment of the present invention. Figure 1 The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not limit this. Steps S1 to S9 are detailed as follows: Step S1: By accessing the bidding and tendering business system and external data sources, bidding and tendering business data and external related data are obtained respectively, and field standardization processing is performed. The bidding and tendering business data includes project data, bidder data, bidding behavior data, document evidence data, equipment and network data, and funding or guarantee data; Step S2 involves performing normalization processing on entities in bidding business data and external related data by combining deterministic rules and similarity models, and converting the relationships between entities into evidentiary events. The entities include the same bidder, the same personnel, the same equipment, or the same account. The normalization process specifically involves: merging companies with identical unified social credit codes into a single entity; generating candidate merger relationships and assigning confidence levels to companies with highly similar names and matching registered addresses, legal representatives, or historical names; combining and matching personnel names, mobile phone numbers, ID card suffixes, and authorization letter information to identify the same authorized representative or contact person; hashing technical fields such as IP addresses, device fingerprints, CA certificates, and file authors to form comparable evidence identifiers; and splitting the consortium into host nodes, member nodes, and consortium relationship edges for consortium bidding scenarios. Evidence events include event number, project number, occurrence time, evidence type, source bidder, target bidder, related objects, evidence value, confidence level, original data source, and traceable fields. Step S3: Extract the relationship features between bidders based on the evidence events, and configure basic weights, confidence calculation methods, time decay parameters and false alarm deduction conditions for each relationship feature to form a candidate association edge set; the relationship features include project co-occurrence features, shared personnel features, equipment network features, document similarity features, financial guarantee features, bidding behavior features, winning bid rotation features, static relationship features and abnormal collaborative behavior features; Step S4: For any two candidate association edges generated between bidders, establish an association edge conflict resolution model, perform consistency analysis on the evidence supporting the association relationship and the evidence weakening the association relationship, and screen credible association edges. Step S5: Based on the association characteristics of the trusted association edges, calculate the comprehensive association strength between any two bidders, generate time-series association edges, calculate the weight increment of the time-series association edges, identify enhanced edges and dormant edges, and construct a dynamic association network. Step S6: Based on the newly added project data, locate the affected nodes and associated edges in the dynamic association network, and perform incremental updates for the affected nodes and associated edges. Step S7: Dynamic community discovery is performed by introducing time smoothing constraints to identify candidate overlapping groups in the dynamic association network; Step S8: Obtain the state components of candidate overlapping groups, construct a group risk scoring model based on the state components to calculate the group risk score, and determine the risk level based on the group risk score; Step S9 involves identifying multiple types of target members among candidate overlapping gangs, recording gang evolution events, and then generating a gang risk report.

[0020] It should also be noted that all data parameters involved in the calculation must undergo preprocessing operations including dimensionless transformation.

[0021] Multiple target members are categorized into core members, peripheral members, bridging members, and suspected control members. Core members typically meet the following criteria: high-weight edges with multiple members, participation in multiple key projects, a central position in bid rotations, and strong evidentiary connections to personnel or equipment. Peripheral members usually appear only in some projects or have a relationship with only one core member. Bridging members are bidders connecting two or more candidate groups; they may be agents, actual controllers, or intermediaries. Suspected control members can be inferred through the strength of connections between personnel, equipment, accounts, document creation environments, and multiple companies. The system records group evolution events, including: initial formation, member joining, member leaving, core member changes, group expansion, group contraction, group splitting, group merging, risk level increase, and risk level decrease. Each evolution event is associated with a specific project and evidentiary event, facilitating retrospective analysis by regulatory personnel.

[0022] The gang risk report includes: gang number, risk level, risk score, first discovery time, and last update time; a list of gang members, including company name, unified social credit code, member role, core status, and risk contribution; a list of key projects, including project number, project name, bidding time, project amount, participating members, and winning members; a member relationship network diagram, including edge type, edge weight, number of evidence, and last evidence time; an evidence details table, including evidence type, evidence description, occurrence time, project source, and original field location; a bid and winning rotation table, showing bid ranking, winning results, and bid rejection / abandonment behavior across multiple projects; a document similarity and equipment account evidence summary, showing similar fragments, metadata, upload environment, or margin path; a gang evolution timeline, showing relationship strengthening, member changes, and risk level changes; a system explanation, explaining the main reasons why the gang was judged as a Level 1 or Level 2 risk; and a manual review entry point for marking confirmation, exclusion, supplementing evidence, or adjusting weights.

[0023] Preferably, for any two candidate association edges generated between bidders, an association edge conflict resolution model is established. Consistency analysis is performed on the evidence supporting the association and the evidence weakening the association to screen credible association edges. The specific steps are as follows: Based on the type of evidence, the evidentiary events corresponding to the candidate related edges are divided into supporting evidence sets and conflicting evidence sets. Supporting evidence sets include evidence that strengthens the relationship, such as shared equipment, shared margin accounts, shared contacts, shared authorized representatives, highly similar documents, rotating winning bids, and consistent pricing tiers. Conflicting evidence sets include evidence that weakens the relationship, such as long-term independent participation in bidding by the enterprise, long-term independent use of equipment, no overlap in personnel, no shared projects in the past, normal industry cooperation, legitimate collaborative relationships within the group, and manually confirmed normal cooperation records. The number of pieces of evidence, their weight, occurrence time, and credibility are statistically analyzed for each type of evidence in the supporting evidence set and the conflicting evidence set. The strength of the supporting association and the strength of the conflicting association are then calculated. Trustworthy association edges are selected based on the strength of supporting association and the strength of conflicting association. Specifically, the strength difference between supporting association and conflicting association is calculated. When the strength difference is greater than a preset difference threshold, the corresponding candidate association edge is marked as a trustworthy association edge. Otherwise, the corresponding candidate association edge is marked as an association edge to be manually reviewed.

[0024] Specifically, the calculation steps for supporting correlation strength are as follows: For each piece of evidence in the supporting evidence set, a basic correlation weight is determined based on the evidence type. This basic correlation weight characterizes the contribution of different types of evidence to the bidder's correlation relationship. The credibility of the first piece of evidence is calculated by weighting the data source credibility, data integrity, multi-source consistency, and time decay factor. The correlation contribution value of each piece of supporting evidence is then calculated using both the basic correlation weight and the first piece of evidence credibility. All correlation contribution values ​​in the supporting evidence set are summed, and the synergistic enhancement coefficient generated when multiple pieces of supporting evidence appear together within the same time window is used to enhance and compensate for the mutual corroboration effect among multiple pieces of evidence, ultimately yielding the supporting correlation strength. Supporting correlation strength not only comprehensively considers the importance of evidence types but also fully integrates evidence quality, evidence timeliness, and the synergistic relationship among multiple pieces of evidence, accurately reflecting the possibility of a genuine correlation between two bidders.

[0025] The specific steps for calculating the conflict association strength are as follows: determine the corresponding basic weakening weight based on the type of conflict evidence; calculate the corresponding second evidence credibility based on the data source credibility, data integrity, multi-source consistency, and time decay factor of each conflict evidence; calculate the weakening contribution value of each conflict evidence based on the basic weakening weight and the second evidence credibility; and sum up all the weakening contribution values ​​in the conflict evidence set to obtain the conflict association strength.

[0026] It should be noted that when there is a large amount of conflicting evidence or highly credible conflicting evidence appears continuously, the strength of the conflict association will gradually increase, thereby effectively weakening the false association edges caused by accidental joint bidding, normal industry cooperation, or historical relationships. Conversely, when there is a small amount of conflicting evidence or the credibility of the second piece of evidence is low, it will not have a significant impact on the supporting association. By constructing supporting association strength and conflicting association strength separately, the relative contributions between strengthening and weakening association evidence can be comprehensively measured in the subsequent association consistency analysis process, realizing the dynamic adjustment of association edge weights. This provides a more realistic and reliable basis for association relationships in dynamic graph construction, community discovery, and identification of bid-rigging groups.

[0027] Preferably, based on the association characteristics of the credible association edges, the comprehensive association strength between any two bidders is calculated to generate time-series association edges. The specific steps are as follows: By taking the bidders as core nodes, the comprehensive correlation strength between any two bidder nodes is calculated using the following formula:

[0028] In the formula: Let be the overall correlation strength between bidder nodes i and j at time t. For the weight of evidence type, Evidence type within the current time window The normalized intensity, For evidence type The most recent occurrence time, The most recent type of evidence The time of occurrence, The total number of evidence types, As a type of evidence; Calculate edge weight increment based on the comprehensive correlation strength between adjacent time points. ,in, Let be the overall correlation strength between bidder nodes i and j at time t. Let be the comprehensive correlation strength between bidder nodes i and j at time t-1; Based on edge weight increment identification of enhanced edges and dormant edges, a dynamic association network is constructed. ,in, For a set of nodes, It is the set of edges connecting nodes. This represents the set of attributes for nodes and associated edges; specifically: if the increment of the edge weight calculated for the association strength is greater than or equal to a preset increment threshold, the corresponding edge is marked as an enhanced edge; if the increment of the edge weight calculated for the association strength is less than the preset increment threshold, the corresponding edge is marked as a dormant edge.

[0029] Preferably, based on the newly added project data, the affected nodes and associated edges in the dynamic association network are located, and incremental updates are performed on the affected nodes and associated edges. The specific steps are as follows: Read new projects and related bidders; extract new evidence events; map evidence events to related edges; update the weights, most recent occurrence times, and evidence lists of affected edges; update the degree, weighted degree, nearest neighbor list, and historical number of bids of bidders; submit the affected subgraph to the community discovery module; match the new community results with the community results in the previous window to generate evolution events.

[0030] This invention calculates the comprehensive association strength between any two bidders based on the association characteristics of credible association edges, and further generates temporal association edges and a dynamic association network. This overcomes the limitations of traditional bidding relationship analysis methods that rely on fixed rules or single time windows for association judgment, enabling a continuous portrayal of the changing relationship between bidders over time. Specifically, by treating bidders as core nodes in the dynamic association network, it integrates multi-dimensional evidence features such as joint bidding, shared personnel, shared equipment, similar documents, linked financial accounts, synchronized bidding behavior, and rotating winning bids. The comprehensive association strength is calculated by combining evidence type weights, evidence strength, and a time decay mechanism, allowing association evidence from different sources, with varying degrees of credibility, and occurring at different times to participate in the bidder relationship evaluation. Recently occurring and highly credible association evidence receives a higher contribution value, while historical evidence that has not been updated for a long time or whose association value has decreased gradually loses its influence through time decay, thereby improving the accuracy of association edge construction and avoiding false associations caused by historical coincidences.

[0031] By calculating the incremental edge weights between adjacent time windows, the development trend of relationships among bidders is dynamically analyzed. When the edge weight of a certain relationship continuously increases, recently strengthening edges can be identified in a timely manner, revealing new resource sharing, collaborative bidding, or increased abnormal behavior among bidding entities, providing a basis for the early detection of potential bid-rigging groups. When a relationship shows no new evidence for a long period and its edge weight gradually decreases, it is identified as a dormant edge, reducing its impact on subsequent community discovery and risk assessment, and minimizing the interference of historical invalid relationships on the analysis results. Based on the dynamic adjustment mechanism of strengthening and dormant edges, the constructed dynamic relationship network can truly reflect the evolutionary process of the formation, strengthening, weakening, and disappearance of relationships among bidding entities.

[0032] By incrementally updating newly added project data, only the affected bidder nodes and related edges are located. The weights, most recent occurrence times, and evidence lists of the related edges corresponding to newly added evidence events are updated, eliminating the need to recalculate the entire network. This reduces the computational burden under continuously growing large-scale bidding data and improves the real-time processing capabilities of this invention. After updating the related edges, the node attributes of the bidders are updated synchronously, including node degree, weighted degree, nearest neighbor relationships, and historical participation counts. This allows node characteristics to be adjusted in real time as business data changes. The affected subgraph is then input into the community discovery module to achieve continuous tracking of potential related groups.

[0033] This invention enables the formation of a dynamic bidder association network with time-aware and continuously updated capabilities. Compared to traditional static association analysis methods, it can not only identify existing abnormal associations but also track the development trends of these associations, discovering suspected collusion groups that are forming or expanding. Furthermore, based on the community evolution results output by the dynamic network, the formation, expansion, contraction, merger, and demise of these groups can be further analyzed. This provides a more accurate, continuous, and interpretable data foundation for subsequent risk scoring, evidence chain generation, and regulatory decisions, improving the real-time performance, accuracy, and reliability of bidding risk identification.

[0034] Preferably, dynamic community detection is performed by introducing time smoothing constraints to identify candidate overlapping groups in dynamic association networks. The specific steps are as follows: Candidate edges are selected based on edge weight threshold, number of strong evidence, number of common projects, and recent enhancements to form a candidate association subgraph; Dynamic community discovery is performed by introducing time smoothing constraints into the candidate subgraph, resulting in the current set of community partitions. In the formula: Let be the k-th community at time t. Divide the communities into numbers; Based on adjacent time windows, community member overlap indicators are calculated separately, including member overlap rate, core member overlap rate, and peripheral evidence overlap rate. The community evolution status is comprehensively judged by the community member overlap indicators. The community evolution status includes community expansion, contraction, merger, and community splitting and disappearance.

[0035] Furthermore, the objective function for performing dynamic community detection is: ; In the formula: Optimize target values ​​for dynamic communities within a time window. The weighted modularity of the time window t. This is the time smoothing coefficient. The degree of difference between the community divisions of two adjacent time windows. The set of community divisions obtained within the time window. This is the set of community divisions obtained from time window - 1.

[0036] The community member overlap index is calculated separately for each community based on adjacent time windows, including member overlap rate, core member overlap rate, and peripheral evidence overlap rate. The formula for calculating the member overlap rate is as follows: ; In the formula: For member overlap rate, The set of community divisions obtained within the time window. This is the set of community divisions obtained within the time window t-1.

[0037] The formula for calculating the overlap rate of core members is: ; In the formula: The overlap rate of edge evidence. The set of associated edges for the time window. Let be the set of associated edges for time window t-1.

[0038] The evolutionary status of a community is determined by comprehensively assessing the overlap indicators of its members. These evolutionary statuses include community expansion, community contraction, community merger, community splitting, and community disappearance. The specific criteria for judging community expansion are as follows: when the member overlap rate is greater than the preset overlap rate threshold, and the number of communities obtained in time window t is greater than the number of communities obtained in time window t-1, community expansion is judged to have occurred. The specific criteria for judging community contraction are as follows: when the member overlap rate is greater than the preset overlap rate threshold, and the number of communities obtained in time window t is less than the number of communities obtained in time window t-1, then community expansion is judged to have occurred. The specific criteria for determining community merging are as follows: if there are several historical community division sets that correspond to the current community division set, and the member overlap rate is greater than a preset overlap rate threshold, then community merging is determined to exist. The specific criteria for judging community splitting are as follows: if there is a historical community division set that corresponds to several current community division sets, and the overlap rate of members is greater than a preset overlap rate threshold, then community splitting is judged to exist. The criteria for determining the disappearance of a community are as follows: if a historical community has no valid associated edges within a number of consecutive time windows and the overlap rate of its members is lower than a preset overlap rate threshold, then the community is marked as extinct.

[0039] Preferably, the state components of candidate overlapping groups are obtained, a group risk scoring model is constructed based on the state components to calculate the group risk score, and the risk level is determined based on the group risk score. The specific steps are as follows: Obtain the state components of candidate overlapping groups, including: community density; proportion of strong evidence; persistence indicators; edge weight growth rate; bid rotation rate; bid anomaly rate; document similarity; community isolation rate; historical risk data; and normal cooperation deduction value. A gang risk scoring model is constructed based on state components to calculate the gang risk score. The calculation formula is as follows: ; In the formula: Assess the risk of the group. The weight coefficients are the weights corresponding to the p-th state component. For the p-th state component, The weighting coefficient corresponding to the deduction value for normal cooperation. This is a deduction for normal cooperation. The number of state components excluding normal cooperation deduction values; The risk level is determined based on the group's risk score, which includes Level 1, Level 2, and Level 3 risk. Specifically, when the group's risk score... At that time, it was at level one risk; when the group's risk score was... At that time, it was at level two risk; when the group's risk score was... At that time, it was at level three risk.

[0040] Example 2: In this embodiment of the invention, a large amount of bidding data for engineering construction, equipment procurement, and service procurement is generated daily on a public resource transaction supervision platform in a certain region. The regulatory department needs to conduct correlation analysis on bidding companies participating in multiple projects to identify whether there are corporate groups engaging in bid-rigging through methods such as sharing qualifications, collaborative bidding, document copying, and rotating winning bids. A dynamic bid-rigging group identification method based on the bidder association network is applied to the above scenario.

[0041] Multi-source heterogeneous data is obtained by accessing the bidding and tendering business system and external data sources. The bidding and tendering business system includes an electronic trading platform, a bid evaluation system, a bid security management system, and a bid document management system, which are used to obtain business data such as bidding company information, project number, bidding time, bid price, bid documents, review results, bid security payment information, and login operation logs. External data sources include business registration databases, enterprise credit information platforms, personnel information databases, equipment access log databases, and historical penalty databases, which are used to obtain corporate legal person information, equity relationships, registered addresses, contact information, historical penalty records, and other related information.

[0042] Because data formats differ across data sources—for example, the same company may use different names in different systems, and the same person may have abbreviated names or missing ID information—the first step is to standardize the fields of the acquired data. Specifically, company names, unified social credit codes, personnel identity information, equipment identifiers, account identifiers, time formats, and project number formats are uniformly converted to eliminate data format differences between different data sources, providing a unified data foundation for subsequent entity association analysis.

[0043] After data standardization, entities in bidding and tendering data and externally related data are normalized. In real-world scenarios, some bidding companies may participate in bidding by changing their names, registering different entities, or using affiliated companies. Therefore, a combination of deterministic rules and similarity models is used for entity matching. Deterministic matching is performed based on strong unique identifiers such as the unified social credit code, business registration number, and legal representative identity information. For data lacking unique identifiers, a similarity model is constructed based on company name similarity, registered address similarity, contact person relationships, historical name associations, and personnel cross-relationships. A comprehensive score is then used to determine whether different records belong to the same entity.

[0044] After entity normalization, the associated behaviors between entities are converted into standardized evidentiary events. For example, when Company A and Company B submit bids for the same project simultaneously, a "joint bidding event" is generated; when two companies log in to the bidding system using the same device, a "shared device event" is generated; when the bid documents submitted by two companies have highly similar content, a "document similarity event" is generated; and when two companies use the same deposit account to pay fees, an "account association event" is generated. Each evidentiary event records the event number, associated company, event type, occurrence time, evidence source, evidence value, and credibility information, transforming raw business data into calculable relationships.

[0045] Based on the generated evidence events, relationship characteristics between bidders are extracted to construct a candidate set of related edges. For example, for bidder A and bidder B, if they jointly participate in a project, a project co-occurrence relationship is formed; if they use the same login device, a device relationship is formed; if their bid documents are highly similar, a textual relationship is formed; and if their price change patterns are consistent, a behavioral relationship is formed. For different types of relationship characteristics, basic weights, confidence calculation methods, time decay parameters, and false alarm deduction conditions are configured respectively. For example, shared security deposit accounts, due to their strong exclusivity, are assigned a higher basic weight; while registered addresses in the same region, which may indicate a normal business partnership, are assigned a lower weight, and a false alarm deduction parameter is added.

[0046] After obtaining candidate related edges, to avoid the influence of some normal cooperative relationships or accidental relationships on the analysis results, a related edge conflict resolution model is further established. Specifically, for any two candidate related edges between bidders, the evidence of association is divided into a set of supporting evidence and a set of conflicting evidence. Among them, supporting evidence includes evidence that can strengthen the association, such as shared equipment, shared accounts, joint control personnel, highly similar documents, synchronized bidding, and abnormal bidding rotation; conflicting evidence includes evidence that can weaken the association, such as long-term independent bidding, long-term independent use of equipment, no joint project records, and manual confirmation of normal cooperative relationships.

[0047] For the supporting evidence set, the supporting correlation strength is calculated by combining evidence type weights, evidence credibility, and time decay factors; for the conflicting evidence set, the corresponding conflict correlation strength is calculated. When the supporting correlation strength is significantly higher than the conflict correlation strength, the candidate correlation edge is identified as a credible correlation edge; when the conflict correlation strength is high, the weight of the correlation edge is reduced; when the two types of evidence are close, they are marked as correlation edges awaiting manual review. Through this method, false correlations caused by factors such as public office networks and industry partnerships are reduced, improving the accuracy of the correlation network.

[0048] The overall association strength between any two bidders is calculated based on the filtered trusted association edges. Using bidders as nodes, the overall association strength is calculated based on different relationship characteristics such as joint bidding, shared equipment, similar files, account association, and bidding behavior.

[0049] Based on the calculated comprehensive association strength, temporal association edges are generated, and the change in edge weight between adjacent time windows is further calculated.

[0050] When the edge weight increment continues to increase and exceeds the set threshold, the associated edge is marked as an enhanced edge. For example, if Company A and Company B have recently participated in multiple identical projects and new evidence such as shared equipment and synchronized pricing has emerged, it indicates that their relationship is strengthening. When the edge weight continues to decrease and there is no new evidence for a long period of time, it is marked as a dormant edge, eventually forming a dynamic association network.

[0051] When a new bidding project enters the system, the entire network is not recalculated. Instead, the affected nodes and related edges are located based on the new project data. For example, if companies A and B jointly bid in a new project, the system only updates the related edges involving A and B. Simultaneously, new evidence events are extracted, and the weights, most recent occurrence times, and evidence lists of related edges are updated. The node degree, weighted degree, nearest neighbor list, and historical participation count of each bidder are also updated concurrently. The updated affected subgraph is submitted to the community discovery module, achieving incremental updates to the dynamic related network.

[0052] Based on the dynamic association network, a time-smoothing constraint is introduced to perform dynamic community discovery. Candidate association communities are identified based on the weight of associated edges, the number of strong evidences, the number of shared projects, and the state of reinforcing edges. Simultaneously, the continuity between the current community structure and the community structure of the previous time window is considered to avoid drastic changes in community structure due to a single abnormal bid. For example, if companies A, B, and C maintain high association strength for multiple consecutive time windows, the system identifies this as a continuously evolving community; if company D recently joins and generates multiple strong associations, it is identified as a community expansion event.

[0053] For identified candidate overlapping groups, further state components are extracted, including community density, proportion of strong evidence, cross-project persistence, relationship growth rate, bid rotation degree, degree of bid anomaly, document similarity, and historical risk information. These are then input into the group risk scoring model to calculate a risk score. Based on the scoring results, groups are classified as high-risk, medium-risk, and attention-level groups to achieve different levels of supervision.

[0054] Within the candidate overlapping groups, different types of target members are further identified, including core members, bridging members, peripheral members, and suspected controlling members. Evolutionary events such as group formation, member joining, member leaving, community expansion, community splitting, and risk changes are continuously recorded. Group member relationships, key related evidence, abnormal project records, bidding behavior characteristics, and evolutionary processes are summarized to generate a group risk report, providing regulators with traceable and interpretable data for analyzing bid-rigging risks.

[0055] This invention transforms every bidding activity, every shared personnel or equipment, every document similarity, every abnormal bid, and every joint bid or winning bid rotation into "evidence events." These evidence events are then mapped to type-, time-, and weighted association edges between bidders. The dynamic association network of bidders is continuously updated within a sliding time window. Finally, potential bid-rigging groups are identified through dynamic community discovery and group risk scoring. This approach transforms bidding information scattered across different business systems into a dynamic association network, continuously tracking changes in relationships between bidding entities over time. It shifts from "single-project anomaly detection" to "cross-project, cross-entity, dynamically evolving group identification," improving the accuracy, real-time nature, and regulatory efficiency of bid-rigging detection.

[0056] Example 3: A dynamic bid-rigging gang identification system based on the bidder association network, including a data acquisition and standardization module, an entity normalization module, a heterogeneous association feature extraction module, an association edge conflict resolution module, a dynamic association network generation module, a network dynamic evolution analysis module, a dynamic community discovery module, a gang risk scoring module, a gang mining and evolution tracking module, an evidence chain generation and visualization output module, and a feedback calibration module. The data acquisition and standardization module is used to acquire bidding business data and external related data by accessing the bidding business system and external data sources, and to perform field standardization processing; the bidding business data includes project data, bidder data, bidding behavior data, document evidence data, equipment and network data, and financial or guarantee data; The entity normalization module is used to perform normalization processing on entities in bidding business data and external related data by combining deterministic rules and similarity models, and to convert the relationships between entities into evidence events. The heterogeneous association feature extraction module is used to extract the relationship features between bidders based on evidence events, and to configure basic weights, confidence calculation methods, time decay parameters and false alarm deduction conditions for each relationship feature to form a candidate association edge set; the relationship features include project co-occurrence features, shared personnel features, equipment network features, document similarity features, financial guarantee features, bidding behavior features, winning bid rotation features, static relationship features and abnormal collaborative behavior features; The associated edge conflict resolution module is used to establish an associated edge conflict resolution model for candidate associated edges generated between any two bidders, perform consistency analysis on the evidence supporting the association relationship and the evidence weakening the association relationship, and screen credible associated edges. The dynamic association network generation module is used to calculate the comprehensive association strength between any two bidders based on the association characteristics of credible association edges, generate time-series association edges, calculate the weight increment of time-series association edges, identify enhanced edges and dormant edges, and construct a dynamic association network. Combining the basic weights of various association characteristics, time decay parameters, and evidence occurrence time, it calculates the comprehensive association strength between any two bidders and generates time-series association edges that include edge type, edge weight, most recent occurrence time, evidence set, credibility level, and time decay parameters. The network dynamic evolution analysis module is used to locate the affected nodes and associated edges in the dynamic network based on the data of newly added projects, and to perform incremental updates on the affected nodes and associated edges. The dynamic community detection module is used to perform dynamic community detection by introducing time smoothing constraints, and to identify candidate overlapping groups in dynamic association networks; The gang risk scoring module is used to obtain the state components of candidate overlapping gangs, build a gang risk scoring model based on the state components to calculate the gang risk score, and determine the risk level based on the gang risk score; The gang discovery and evolution tracking module is used to identify multiple types of target members in candidate overlapping gangs and record gang evolution events; The evidence chain generation and visualization output module is used to generate gang risk reports; The feedback calibration module is used to receive and update gang risk reports.

[0057] like Figure 2 The diagram shown is a system block diagram of a dynamic bid-rigging gang identification system based on a bidder association network according to an embodiment of the present invention, which can be used to execute... Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0058] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of the dynamic bid-rigging gang identification method based on a bidder association network as described above.

[0059] The readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of computer programs from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application-Specific Integrated Circuit (ASIC). Alternatively, the ASIC can be located in a user equipment. Of course, the processor and the readable storage medium can also exist as discrete components in a communication device. The readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0060] The present invention also provides a program product including executable instructions stored in a readable storage medium. At least one processor of the device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the device to implement the methods provided in the various embodiments described above.

[0061] In the embodiments of the above-described device, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0062] Through the above embodiments, this invention integrates bidding business data with external related data to complete the normalization of bidder entities and the construction of evidence events; establishes candidate association edges based on multi-type association features, and filters credible association relationships through an association edge conflict resolution model; calculates the comprehensive association strength of bidders by combining evidence weight, credibility, and time decay mechanism, and constructs temporal association edges and dynamic association networks; performs incremental updates for new projects, introduces time smoothing constraints to perform dynamic community discovery, identifies continuously evolving candidate groups, and calculates risk levels by combining state features, finally outputting a group risk report containing member relationships, evolution process, and key evidence.

[0063] By constructing a time-evolving dynamic network of bidder relationships, this invention enables continuous tracking of implicit relationships across projects and entities. Compared to traditional static rule-based detection methods, it can identify implicit bid-rigging groups across projects and time windows. It is no longer limited to single-project rule hits but continuously maintains the dynamic network of bidder relationships, enabling the discovery of hidden patterns such as long-term joint bidding, phased relationship strengthening, member rotation for dummy bidding, alternating winning bids, and cross-project collaborative pricing. This invention integrates scattered clues into a group-level evidence chain, unifying multi-source evidence such as personnel, equipment, accounts, documents, quotations, joint bidding, and winning bid rotation into evidence events and related edges, outputting them in a "group-member-project-evidence" structure. Supervisory personnel do not need to manually piece together relationships from numerous individual rules. This invention reduces false alarms. The system sets different weights for different evidence types and introduces time decay and normal cooperation deduction items to avoid simply judging historical accidental joint bidding, industry / regional clustering, and legitimate participation by the same group as bid-rigging. This invention can detect rapidly formed groups in recent times. The system identifies recent relationship enhancements through edge weight increments, recently enhanced edges, and sliding time windows, providing timely warnings even when multiple strong evidence clusters appear, despite limited historical samples of the group. This invention can track the evolution of groups. The system records the formation, expansion, contraction, merger, splitting, member joining and leaving, core member changes, and risk level changes of candidate communities, making it suitable for long-term supervision, special audits, and case lead accumulation. This invention improves interpretability by outputting not only risk scores but also score sources, key edges, key projects, evidence types, occurrence times, and original field locations, enabling regulators to review each item. This invention is adaptable to data from multiple industries, regions, and scales. Evidence weights and thresholds can be configured by industry, region, project amount, and procurement method, and parallel calculations can be performed by time window and regional partitioning, making it suitable for large-scale public resource transaction data. This invention enables a continuous calibration loop. Manual review results can be fed back to the system to adjust evidence weights, risk thresholds, and normal cooperation deduction rules, continuously optimizing the model in actual regulatory scenarios. This invention employs engineerable components such as data cleaning, graph edge updating, community discovery, rule scoring, and interpretable report generation, without relying on a single, uninterpretable black-box model, thus facilitating auditing and maintenance.

[0064] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

[0065] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying dynamic bid-rigging gangs based on bidder association networks, characterized in that, Includes the following steps: By accessing the bidding and tendering business system and external data sources, bidding and tendering business data and external related data are obtained respectively, and field standardization processing is performed. By combining deterministic rules and similarity models, normalization processing is performed on entities in bidding business data and external related data, and the relationships between entities are converted into evidentiary events. Based on the evidence events, the relationship characteristics between bidders are extracted, and basic weights, confidence calculation methods, time decay parameters and false alarm deduction conditions are configured for each relationship characteristic to form a candidate set of related edges; For any two bidders, a conflict resolution model for the candidate association edges generated between them is established. Consistency analysis is performed on the evidence supporting the association and the evidence weakening the association to screen credible association edges. Based on the association characteristics of credible association edges, the comprehensive association strength between any two bidders is calculated, time-series association edges are generated, the weight increment of time-series association edges is calculated, enhanced edges and dormant edges are identified, and a dynamic association network is constructed. Based on the newly added project data, locate the affected nodes and related edges in the dynamic association network, and perform incremental updates for the affected nodes and related edges; By introducing time smoothing constraints to perform dynamic community discovery, candidate overlapping groups in dynamic association networks are identified, state components of candidate overlapping groups are obtained, a group risk scoring model is constructed based on the state components to calculate the group risk score, and the risk level is determined based on the group risk score. By identifying multiple types of target members among candidate overlapping gangs and recording gang evolution events, a gang risk report is generated after summarizing the data.

2. The method for identifying dynamic bid-rigging groups based on bidder association networks according to claim 1, characterized in that, Relationship characteristics include project co-occurrence characteristics, shared personnel characteristics, equipment network characteristics, document similarity characteristics, financial guarantee characteristics, bidding behavior characteristics, winning bid rotation characteristics, static relationship characteristics, and abnormal collaborative behavior characteristics.

3. The method for identifying dynamic bid-rigging groups based on bidder association networks according to claim 1, characterized in that, For any two candidate association edges generated between bidders, an association edge conflict resolution model is established. Consistency analysis is performed on the evidence supporting the association and the evidence weakening the association to screen credible association edges. The specific steps are as follows: Based on the type of evidence, the evidence events corresponding to the candidate related edges are divided into supporting evidence sets and conflicting evidence sets. The number of pieces of evidence, their weight, occurrence time, and credibility are statistically analyzed for each type of evidence in the supporting evidence set and the conflicting evidence set. The strength of the supporting association and the strength of the conflicting association are then calculated. Trustworthy association edges are selected based on the strength of supporting association and the strength of conflicting association. Specifically, the strength difference between supporting association and conflicting association is calculated. When the strength difference is greater than a preset difference threshold, the corresponding candidate association edge is marked as a trustworthy association edge. Otherwise, the corresponding candidate association edge is marked as an association edge to be manually reviewed.

4. The method for identifying dynamic bid-rigging groups based on bidder association networks according to claim 3, characterized in that, The process of selecting trustworthy association edges involves calculating the strength difference between the supporting association strength and the conflicting association strength. When the strength difference is greater than a preset threshold, the corresponding candidate association edge is marked as a trustworthy association edge; otherwise, the corresponding candidate association edge is marked as an association edge awaiting manual review.

5. The method for identifying dynamic bid-rigging groups based on bidder association networks according to claim 1, characterized in that, Based on the association characteristics of credible association edges, the comprehensive association strength between any two bidders is calculated, and time-series association edges are generated. The specific steps are as follows: By taking the bidders as core nodes, the comprehensive correlation strength between any two bidder nodes is calculated. Calculate edge weight increments based on the comprehensive correlation strength between adjacent time points; A dynamic association network is constructed based on the identification of enhanced and dormant edges using edge weight increments. Specifically, if the edge weight increment calculated by the association strength is greater than or equal to a preset increment threshold, the corresponding edge is marked as an enhanced edge; if the edge weight increment calculated by the association strength is less than the preset increment threshold, the corresponding edge is marked as a dormant edge.

6. The method for identifying dynamic bid-rigging groups based on bidder association networks according to claim 1, characterized in that, By introducing time smoothing constraints to perform dynamic community discovery, candidate overlapping groups in dynamic association networks are identified. The specific steps are as follows: Candidate edges are selected based on edge weight threshold, number of strong evidence, number of common projects, and recent enhancements to form a candidate association subgraph; Dynamic community discovery is performed by introducing time smoothing constraints into the candidate subgraph, resulting in the current set of community partitions. In the formula: Let be the k-th community at time t. Divide the communities into numbers; The community member overlap index is calculated for each community based on adjacent time windows, and the community evolution status is judged by comprehensively judging the community member overlap index.

7. The method for identifying dynamic bid-rigging groups based on bidder association networks according to claim 6, characterized in that, Community member overlap indicators include member overlap rate, core member overlap rate, and peripheral evidence overlap rate; community evolution status includes community expansion, contraction, merger, and community splitting and disappearance.

8. The method for identifying dynamic bid-rigging groups based on bidder association networks according to claim 1, characterized in that, Obtain the state components of candidate overlapping groups, construct a group risk scoring model based on the state components to calculate the group risk score, and determine the risk level based on the group risk score. The specific steps are as follows: Obtain the state components of candidate overlapping groups, construct a group risk scoring model based on the state components, and calculate the group risk score. The calculation formula is as follows: ； In the formula: Assess the risk of the group. The weight coefficients are the weights corresponding to the p-th state component. For the p-th state component, The weighting coefficient corresponding to the deduction value for normal cooperation. This is a deduction for normal cooperation. The number of state components excluding normal cooperation deduction values; The risk level is determined based on the group's risk score, which includes Level 1 risk, Level 2 risk, and Level 3 risk.

9. A dynamic bid-rigging gang identification system based on a bidder association network, applied to the dynamic bid-rigging gang identification method based on a bidder association network as described in any one of claims 1-8, characterized in that, It includes modules for data acquisition and standardization, entity normalization, heterogeneous association feature extraction, association edge conflict resolution, dynamic association network generation, network dynamic evolution analysis, dynamic community discovery, gang risk scoring, gang mining and evolution tracking, evidence chain generation and visualization output, and feedback calibration. The data acquisition and standardization module is used to acquire bidding business data and external related data by accessing the bidding business system and external data sources, and to perform field standardization processing; the bidding business data includes project data, bidder data, bidding behavior data, document evidence data, equipment and network data, and financial or guarantee data; The entity normalization module is used to perform normalization processing on entities in bidding business data and external related data by combining deterministic rules and similarity models, and to convert the relationships between entities into evidence events. The heterogeneous association feature extraction module is used to extract the relationship features between bidders based on evidence events, and to configure basic weights, confidence calculation methods, time decay parameters and false alarm deduction conditions for each relationship feature to form a candidate association edge set. The associated edge conflict resolution module is used to establish an associated edge conflict resolution model for candidate associated edges generated between any two bidders, perform consistency analysis on the evidence supporting the association relationship and the evidence weakening the association relationship, and screen credible associated edges. The dynamic association network generation module is used to calculate the comprehensive association strength between any two bidders based on the association characteristics of credible association edges, generate time-series association edges, calculate the weight increment of time-series association edges, identify enhanced edges and dormant edges, and construct a dynamic association network. The network dynamic evolution analysis module is used to locate the affected nodes and associated edges in the dynamic network based on the data of newly added projects, and to perform incremental updates on the affected nodes and associated edges. The dynamic community detection module is used to perform dynamic community detection by introducing time smoothing constraints, and to identify candidate overlapping groups in dynamic association networks; The gang risk scoring module is used to obtain the state components of candidate overlapping gangs, build a gang risk scoring model based on the state components to calculate the gang risk score, and determine the risk level based on the gang risk score; The gang discovery and evolution tracking module is used to identify multiple types of target members in candidate overlapping gangs and record gang evolution events; The evidence chain generation and visualization output module is used to generate gang risk reports; The feedback calibration module is used to receive and update gang risk reports.

10. A readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it is used to implement the steps of the dynamic bid-rigging gang identification method based on the bidder association network as described in any one of claims 1-8.

Images