CN-122020480-A - AI model and data analysis-based bidding abnormal behavior identification method and system

CN122020480ACN 122020480 ACN122020480 ACN 122020480ACN-122020480-A

Abstract

The invention relates to the technical field of abnormal behavior identification, in particular to a bidding abnormal behavior identification method and system based on an AI model and data analysis, comprising the steps of acquiring bidding transaction data, taking a bidding enterprise as a node, taking a common bid of two nodes in the same project as an edge, and constructing a bidding association network map; and constructing the accompanied bidding viscosity index of two nodes based on the node pairs with connection relations in the network map, wherein the accompanied bidding viscosity index is used for representing the tightness degree of any two nodes participating in bidding together. According to the invention, the heterogeneous risk topological potential energy is used as a bias term to be forcedly injected into coefficient calculation of a graph attention model, so that the defect that the existing model only depends on feature similarity is overcome, the model is guided to penetrate through surface camouflage by using service priori knowledge, and hidden partner with high-frequency accompaniment and standard book bottom homology of behaviors is accurately locked.

Inventors

YIN JING
XU XUBO
LI YONGPING
TAN HAIHUAN
FU LE
ZENG YUFEI

Assignees

广州交易集团有限公司

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. A bid abnormal behavior identification method based on AI model and data analysis is characterized by comprising the steps of acquiring bid transaction data, taking a bid enterprise as a node, taking a common bid of two nodes in the same project as an edge, and constructing a bid association network map; Constructing an accompanying bidding viscosity index of two nodes based on node pairs with connection relations in the network map, wherein the accompanying bidding viscosity index is used for representing the tightness degree of any two nodes participating in bidding together; Calculating entropy spectrum consistency coefficients of the two standard entropy spectrum vectors in the same project, wherein the entropy spectrum consistency coefficients are positively correlated with cosine similarity of the standard entropy spectrum vectors and negatively correlated with absolute value of difference of mean values of the standard entropy spectrum vectors; The method comprises the steps of calculating heterogeneous risk topological potential energy of two nodes based on concomitance coefficients of a bidding viscosity index and an entropy spectrum, processing the bidding association network map by adopting a graph attention network model, wherein the heterogeneous risk topological potential energy is used as a bias term to be fused into the attention coefficient calculation process of the graph attention network model, and judging whether abnormal bidding behaviors exist or not according to an output result of the graph attention network model.
2. The AI model and data analysis-based bid anomaly behavior recognition method of claim 1, wherein constructing the companion bid viscosity index for two nodes comprises: And acquiring the number of the items which are jointly participated in the bidding by the two nodes in a preset time window and the total number of the items which are independently participated in the bidding by the two nodes, wherein the accompanying bidding viscosity index is positively correlated with the number of the items which are jointly participated in the bidding by the node in the preset time window and is negatively correlated with the product of the total number of the items which are independently participated in the bidding by the node, and the number of the items which are jointly participated in the bidding reflects the tightness degree between the two nodes.
3. The AI model and data analysis-based bidding anomaly behavior recognition method of claim 1, wherein the standard entropy spectrum vector acquisition method comprises: Traversing binary stream of bidding documents by using time windows with set sizes, and calculating shannon entropy of each time window to obtain an original entropy spectrum vector; and carrying out resampling processing on the original entropy spectrum vector, and mapping the original entropy spectrum vector into a standard entropy spectrum vector with a set length.
4. The AI model and data analysis-based bid anomaly behavior recognition method of claim 1, further comprising, prior to calculating the heterogeneous risk topology potential: normalizing the concomitantly bidding viscosity index and entropy spectrum concordance coefficient to distribute the values thereof in Within the interval.
5. The AI model and data analysis-based bidding abnormal behavior recognition method of claim 1, wherein determining whether there is a bidding abnormal behavior comprises: And if the abnormal probability value output by the drawing meaning network model for a certain bidding enterprise node is larger than a preset judging threshold value, judging that the bidding enterprise has abnormal behaviors.
6. The AI model and data analysis-based bidding anomaly behavior recognition method of claim 1, wherein the outputting of the graph-meaning network model results comprises: And inputting the constructed bidding association network map into a graph attention network model, and outputting an abnormal probability value by each bidding node in the bidding association network map.
7. The AI model and data analysis-based bid anomaly behavior recognition method of claim 6, wherein in response to the anomaly probability value of a bidding node being greater than a decision threshold, it is determined that there is a bid anomaly behavior and a high risk partner node associated with the bidding node is concurrently output.
8. The AI model and data analysis-based bid anomaly behavior recognition method of claim 1, wherein the accompanying bid viscosity index is calculated by the formula: ; In the formula, Indicating at the moment of analysis Lower node And node A concomitant bid viscosity index therebetween; indicated within a time window, nodes Sum node The number of items that are co-engaged in bidding; Representing nodes The total number of items participating in the bid within the time window; Representing nodes The total number of items participating in the bid within the time window; is a smoothing constant for preventing denominator from being zero; Is a constant.
9. The AI model and data analysis-based bid and ask abnormal behavior recognition method of claim 4, wherein calculating two node heterogeneous risk topological potentials comprises: and taking the product of the normalized concomitantly bidding viscosity index and the entropy spectrum consistency coefficient as heterogeneous risk topology potential energy.
10. A bidding abnormal behavior recognition system based on AI model and data analysis, comprising a processor and a memory, characterized in that the memory stores a computer program, the processor executing the computer program to implement the bidding abnormal behavior recognition method based on AI model and data analysis as claimed in any one of claims 1-9.

Description

AI model and data analysis-based bidding abnormal behavior identification method and system Technical Field The invention relates to the technical field of abnormal behavior identification. More particularly, the invention relates to a bid and ask abnormal behavior identification method and system based on AI model and data analysis. Background With the comprehensive popularization and application of electronic bidding systems, bidding activities have become a core mode of resource allocation in the fields of engineering construction, government purchasing, material transaction and the like, and the disclosure, fairness and fairness of the flow are directly related to the maintenance of social public interests and the healthy development of market economic order. In the face of the increasing mass transaction data of bidding, the supervision by utilizing a digital means has become an industry consensus. Currently, the supervision department often utilizes big data analysis or artificial intelligence technology to construct an enterprise association network in an attempt to mine bidder ring, string labels and other abnormal behaviors from the enterprise association network. In this scenario, the identification system typically uses bidding enterprises as nodes, common bidding behaviors as edges, and analyzes the network structure through algorithms such as a graph neural network to lock out offending partners. However, existing technical approaches face serious challenges in the actual abnormal behavior identification and supervision process. Mature bidder ring group partners are increasingly covert and specialized in order to circumvent supervision, and not only do they superficially forge irrelevant enterprise portraits, such as differentiated registration information, business scope, but also deliberately adjust macroscopic features of the taggant (e.g. file size, hash value) to make the illusion of independent competition. The existing graph attention network model often relies on the similarity of the characteristics of the nodes to calculate the attention coefficient when processing the correlation map. This leads to a practical problem in that when the surface features are deliberately disguised to be very different, the recognition model misjudges that the correlation between these nodes is weak, giving it very low attention weight, although there is a high frequency of concomitant bidding phenomena between the offending enterprises and the underlying microstructures of the bidding documents (such as entropy spectra) are highly homologous. This mechanical mismatch makes it difficult for the model to penetrate the surface camouflage to capture the true benefit conveying chain behind, which is extremely prone to missed decisions on high risk covert bidder ring partners. Disclosure of Invention The invention provides a bid and ask abnormal behavior recognition method and system based on an AI model and data analysis, and aims to solve the problem that the model is difficult to penetrate through a surface camouflage to capture a real benefit conveying chain behind when an existing graphic annotation network model processes such a correlation map in the related art, and is very easy to cause missed judgment on a high-risk concealment bidder ring partner. In a first aspect, the invention provides a bidding abnormal behavior identification method based on an AI model and data analysis, which comprises the steps of obtaining bidding transaction data, taking a bidding enterprise as a node, taking a common bid of two nodes in the same item as an edge, constructing a bidding associated network map, constructing an accompanying bidding sticky index of the two nodes based on a node pair with a connection relation in the network map, wherein the accompanying bidding sticky index is used for representing the tightness degree of the bidding of any two nodes in common participation, extracting binary streams of bidding files in the two nodes in common participation item and calculating to obtain a standard entropy spectrum vector, calculating entropy spectrum coincidence coefficients of the two standard entropy spectrum vectors in the same item, wherein the entropy spectrum coincidence coefficients are positively correlated with cosine similarity of the standard entropy spectrum vector and negatively correlated with the absolute value of the difference of the mean value of the standard entropy spectrum vector, calculating the two node heterogeneous risk topology based on the accompanying sticky index and the entropy spectrum coincidence coefficients, processing the bidding associated network, wherein the heterogeneous risk topology potential energy is used as a bias risk item, calculating the bidding abnormal risk map according to the bidding abnormal behavior map, and judging whether the bidding abnormal behavior is output to the network map. By extracting the concomitance coefficients of the bidding viscosity index