Bidding total element compliance examination method based on multi-modal model space-time sequence analysis

Abstract

The invention provides a bidding total factor compliance examination method based on multi-modal model space-time sequence analysis, which relates to the technical field of data processing and comprises the steps of acquiring space-time sequence data and auxiliary data of bidding items, constructing a space-time double-flow feature network, constructing a knowledge graph based on output, and calculating an abnormal score by adopting a space-time mode encoder. And optimizing an anomaly detection result through a graph attention network, and dynamically generating a bidding compliance verification path by using deep reinforcement learning. According to the method, comprehensive analysis of bidding and tendering data can be realized, the abnormal behavior detection accuracy is improved, the manual review cost is reduced, and the verification path efficiency is optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a multi-modal model spatio-temporal sequence analysis method for bidding full-element compliance review. BACKGROUND

[0002] With the continuous development of the bidding market, bidding compliance review, as an important link to maintain market fair competition, has become increasingly important. Traditional bidding compliance review mainly relies on manual methods, which is time-consuming and inefficient. In recent years, with the development of artificial intelligence and big data analysis technology, bidding compliance review has gradually introduced automated and intelligent technical means. Current bidding review systems mainly use text analysis and rule matching methods to review bidding documents and find abnormal bidding behaviors through simple time series analysis.

[0003] However, the existing bidding compliance review technology has obvious shortcomings. The existing technology lacks comprehensive analysis capability in time and space dimensions, and cannot effectively capture the spatio-temporal correlation patterns in bidding activities, resulting in limited recognition ability for complex irregular behaviors such as collusion bidding and rotation bidding. Existing anomaly detection methods mostly use fixed thresholds or simple statistical models, which lack adaptive ability and are difficult to cope with complex and variable abnormal patterns and interference factors in the bidding scene, resulting in high false positive rates. Existing systems usually separate the anomaly detection and compliance verification, lack a systematic method to combine the detection results with specific compliance requirements, and are difficult to form a complete review loop, resulting in a lack of persuasiveness and explainability in the final compliance evaluation results.

[0004] To solve the above problems, a multi-modal bidding compliance review method that can comprehensively utilize spatio-temporal sequence data and auxiliary text information and has adaptive analysis capability is needed to improve the accuracy of abnormal behavior recognition and the efficiency of compliance review. SUMMARY

[0005] The embodiments of the present application provide a multi-modal model spatio-temporal sequence analysis method for bidding full-element compliance review, which can solve the problems in the prior art.

[0006] In a first aspect, the embodiments of the present application provide a multi-modal model spatio-temporal sequence analysis method for bidding full-element compliance review, comprising: obtaining spatio-temporal sequence data and auxiliary data of a bidding project to be reviewed, the spatio-temporal sequence data including bidding time data and bidding geographic location data, and the auxiliary data including bidding text content; constructing a spatio-temporal double-flow feature network, wherein the time sequence feature flow adopts an adaptive time window to dynamically segment the time sequence and extract the time sequence feature representation of the event, and the spatial feature flow adopts density clustering to adaptively partition the geographic location and extract the spatial distribution features; constructing a knowledge graph based on the output of the spatio-temporal dual-flow feature network, the knowledge graph comprising nodes representing bidding events and geographical locations and edges representing temporal dependency relationships and spatial constraint relationships, the attributes of the edges being supplemented by the auxiliary data; based on the knowledge graph, calculating the similarity of node features with a normal pattern library using a spatio-temporal pattern encoder at a local level to obtain an anomaly score, and using a graph attention network to optimize the anomaly score in a context at a global level to obtain an anomaly detection result; based on the anomaly detection result and the auxiliary data, dynamically generating a bidding and tender compliance verification path based on deep reinforcement learning, obtaining a compliance evaluation result by optimizing the verification path, and generating an audit report.

[0007] In an alternative embodiment, The steps of constructing a spatio-temporal dual-flow feature network, wherein the temporal feature flow adopts an adaptive time window to dynamically segment time series and extract temporal feature representations of events, and the spatial feature flow adopts density clustering to adaptively partition geographical locations and extract spatial distribution features include: constructing a temporal feature flow that identifies bidding time points by designing a time sensitivity function for the bidding time data, constructing a non-uniform time window sequence based on the time points, determining the boundaries of the non-uniform time window by maximizing the difference in bidding behavior patterns between windows, and extracting temporal dependency features of bidding events within each time window using an attention-enhanced recurrent network to obtain hierarchical temporal feature representations; constructing a spatial feature flow that dynamically partitions bidding regions using an adaptive spatial partitioning algorithm for the bidding geographical location data, the adaptive spatial partitioning algorithm being optimized by minimizing the bidding behavior entropy within the region and maximizing the difference in bidding behavior between regions, constructing a bidding geographical correlation network based on the spatial partitioning, and extracting spatial distribution features; using a spatio-temporal collaborative attention mechanism to fuse the outputs of the temporal feature flow and the spatial feature flow, the spatio-temporal collaborative attention mechanism dynamically calculating the association strength of the hierarchical temporal feature representations and the spatial distribution features to adaptively adjust the weights of different dimensional features, and obtaining a unified spatio-temporal feature representation.

[0008] In an alternative embodiment, The steps of constructing a knowledge graph based on the output of the spatio-temporal dual-flow feature network, the knowledge graph comprising nodes representing bidding events and geographical locations and edges representing temporal dependency relationships and spatial constraint relationships, the attributes of the edges being supplemented by the auxiliary data include: The fusion feature of the spatio-temporal double-flow feature network is obtained, and the fusion feature includes a time sequence feature representation of the bidding event and a spatial distribution feature of the geographic location. A node set of a knowledge graph is constructed, and the node set includes a bidding event node, a geographic location node, and a bidding subject node. The bidding event node includes event identification, occurrence time, and bidding amount attributes. The geographic location node includes city area division attributes. The bidding subject node includes subject identification, registered capital, and establishment time attributes. The time sequence component and the spatial component of the fusion feature are mapped to a feature space of a corresponding node through a nonlinear mapping function, and a node feature vector is obtained. An edge set of the knowledge graph is constructed, and the edge set includes a time sequence dependent edge, a spatial constraint edge, and a subject association edge. The weight of the time sequence dependent edge is obtained by calculating the inner product of adjacent node feature vectors. The weight of the spatial constraint edge is obtained by calculating the distance between geographic location nodes. The bidding and tender text content is converted into a text semantic vector. The weight vector of the edge is spliced and fused with the text semantic vector to obtain a complete attribute vector of the edge. The edge set is optimized based on the importance score of the complete attribute vector. The spatio-temporal consistency of the edge attribute is ensured by setting the proportion coefficient of the time sequence feature weight and the spatial feature weight.

[0009] In an optional implementation, Based on the knowledge graph, the similarity of the node features and the normal mode library is calculated by using a spatio-temporal mode encoder at a local level to obtain an anomaly score. The anomaly score is contextually optimized by using a graph attention network at a global level to obtain an anomaly detection result. The steps include: Node features and neighborhood information in the knowledge graph are obtained. The node features include time sequence features and spatial features. The neighborhood information includes dependent relationships between nodes. A spatio-temporal mode encoder is used to encode the node features and the neighborhood information to obtain a spatio-temporal encoding representation of the node. A normal mode feature library is constructed based on the spatio-temporal encoding representation. The normal mode feature library includes a plurality of normal mode center representations. The similarity of the spatio-temporal encoding representation of the node and the normal mode center representations is calculated to obtain a local anomaly score of the node. A multi-head graph attention network is constructed. The attention weight between nodes is calculated based on a query matrix, a key matrix, and a value matrix. The node features are weighted and aggregated according to the attention weight to obtain a context representation of the node. The local anomaly score of the node and the context representation of the node are fused to obtain a global anomaly score of the node. The time sequence smoothing constraint is constructed based on the time sequence dependency relationship, the time sequence smoothing constraint is realized by minimizing the difference between global anomaly scores of time sequence related nodes, the global anomaly scores are optimized according to the time sequence dependency strength, and the final anomaly score of the node is obtained; the detection threshold is determined according to the final anomaly score, and the anomaly label is performed, so as to obtain the anomaly detection result.

[0010] In an optional implementation, The step of constructing the time sequence smoothing constraint based on the time sequence dependency relationship, the time sequence smoothing constraint being realized by minimizing the difference between global anomaly scores of time sequence related nodes, and the final anomaly score of the node being obtained according to the time sequence dependency strength includes: The time sequence dependency relationship is constructed, and the time sequence dependency strength of a time sequence related node pair is calculated by using a dynamic graph neural network, the dynamic graph neural network including a time sequence query matrix and a time sequence key matrix, a node feature vector being transformed by using the time sequence query matrix and the time sequence key matrix, and the time sequence dependency strength of the node pair being calculated by combining a time interval attenuation function; A smoothing constraint target function is constructed based on the time sequence dependency strength, the smoothing constraint target function including a smoothing loss term and a causal loss term, the smoothing loss term being used to calculate the square difference of global anomaly scores of related nodes based on the time sequence dependency strength, and the causal loss term being used to calculate the deviation of the global anomaly score of the node from an expected value based on an intervention operation; An anomaly score propagation function is designed, the global anomaly score of the node is iteratively updated according to the time sequence dependency strength, the smoothing constraint target function is taken as an optimization target, the global anomaly score is optimized by using a gradient descent method, and a penalty term is introduced in each iteration to constrain the change amplitude of the global anomaly score; The optimized global anomaly score is subjected to multi-scale fusion, and the final anomaly score of the node is calculated based on the global anomaly scores of different scales and corresponding weights.

[0011] In an optional implementation, The step of modeling the anomaly verification as a sequential decision process based on deep reinforcement learning, dynamically generating a bidding compliance verification path based on the anomaly detection result and the auxiliary data, obtaining a compliance evaluation result by optimizing the verification path, and generating an audit report includes: A verification feature vector is constructed based on the anomaly detection result, the verification feature vector being used to construct a verification state space and a verification action space in combination with the auxiliary data, and the verification state space including the verification feature vector and a historical verification decision sequence; construct a verification policy generation network based on the verification state space and the verification action space, the verification policy generation network adopts a variational autoencoder to encode the verification feature vector to obtain a latent representation of an abnormal pattern, a verification policy distribution is constructed based on the latent representation, and a verification action sequence is generated according to the verification policy distribution; a verification evaluation function is constructed to include a verification accuracy term, a verification cost term and a decision consistency term, a Thompson sampling method is used to sample and optimize the verification evaluation function to obtain an optimal verification action sequence, and a verification path is determined according to the optimal verification action sequence; a verification process is performed along the verification path, a time sequence compliance index, a spatial consistency index and a text semantic similarity index are calculated for each verification node, a multi-dimensional risk dependence matrix is constructed based on the time sequence compliance index, the spatial consistency index and the text semantic similarity index, a Copula function is used to calculate a multi-dimensional risk joint distribution, and a compliance evaluation result is output; based on the compliance evaluation result, abnormal spatio-temporal pattern information and its propagation path are extracted, and an audit report is generated.

[0012] In an alternative embodiment, the step of constructing a verification evaluation function to include a verification accuracy term, a verification cost term and a decision consistency term, and using a Thompson sampling method to sample and optimize the verification evaluation function to obtain an optimal verification action sequence includes: the verification accuracy term is calculated based on an abnormal detection rate corresponding to the verification action sequence, the verification cost term is calculated based on resource consumption of a verification depth and a verification range, and the decision consistency term is calculated based on a strategy difference between adjacent verification actions; a Bayesian neural network is used to implement the Thompson sampling method, a posterior distribution of the verification evaluation function is established, a weight distribution of the Bayesian neural network is trained based on historical verification decision sequences, and a probability representation of the verification evaluation function is obtained; at each round of verification decision, an instance of the current verification evaluation function is sampled from the posterior distribution based on the Thompson sampling principle, candidate verification action sequences are scored based on the instance of the current verification evaluation function, and the verification action sequence with the highest score is selected as the optimal verification action sequence; the posterior distribution is updated according to verification feedback information, exploration and utilization are balanced by adaptively adjusting a sampling temperature parameter, and the sampling temperature parameter decreases with an increase in verification rounds.

[0013] In a second aspect of the embodiment of the application, an electronic device is provided, including: a processor; a memory for storing processor-executable instructions; The processor is configured to invoke instructions stored in the memory to perform the method described above.

[0014] In a third aspect of the embodiments of the present application, a computer readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0015] The present application realizes fine processing of the bidding spatio-temporal data by constructing a spatio-temporal double-flow feature network, wherein the time sequence feature flow is dynamically segmented by using an adaptive time window, and the spatial feature flow is adaptively partitioned by using density clustering, so as to effectively capture the spatio-temporal pattern change in the bidding activities and improve the accuracy of the anomaly detection.

[0016] The present application combines the knowledge graph technology, and performs multi-modal fusion of the spatio-temporal features and the text content, so as to construct a knowledge structure containing bidding events, geographic location nodes and their relationships, realize semantic understanding and correlation analysis of all factors of the bidding, and enhance the depth and breadth of the compliance review.

[0017] The present application adopts an abnormality verification mechanism based on deep reinforcement learning, models the compliance review as a sequential decision-making process, dynamically generates a bidding compliance verification path according to the preliminary anomaly detection result, significantly improves the review efficiency and accuracy, reduces the artificial review cost, and provides an intelligent solution for the bidding supervision. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 FIG. 1 is a flowchart of a bidding full-factor compliance review method for spatio-temporal sequence analysis of a multi-modal model according to an embodiment of the present application; Figure 2 FIG. 2 is a spatio-temporal anomaly detection flowchart based on a knowledge graph according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0020] The technical scheme of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described in some embodiments.

[0021] Figure 1A flowchart of a bidding full-factor compliance review method based on multi-modal model spatio-temporal sequence analysis according to an embodiment of the present application is shown in FIG. 1, which includes the following steps: Figure 1 obtaining spatio-temporal sequence data and auxiliary data of a bidding project to be reviewed, the spatio-temporal sequence data including bidding time data and bidding geographic location data, and the auxiliary data including bidding text content; constructing a spatio-temporal double-flow feature network, wherein a time sequence feature flow adopts an adaptive time window to dynamically segment a time sequence and extract a time sequence feature representation of an event, and a spatial feature flow adopts a density clustering to adaptively partition a geographic location and extract a spatial distribution feature; constructing a knowledge graph based on an output of the spatio-temporal double-flow feature network, the knowledge graph including nodes representing bidding events and geographic locations and edges representing time sequence dependency relationships and spatial constraint relationships, and attributes of the edges being supplemented by the auxiliary data; based on the knowledge graph, calculating a similarity of node features with a normal pattern library by a spatio-temporal pattern encoder at a local level to obtain an anomaly score, and optimizing the anomaly score by a graph attention network at a global level to obtain an anomaly detection result; modeling anomaly verification as a sequential decision process based on deep reinforcement learning, dynamically generating a bidding compliance verification path based on the anomaly detection result and the auxiliary data, obtaining a compliance evaluation result by optimizing the verification path, and generating a review report.

[0022] In an alternative embodiment, the step of constructing a spatio-temporal double-flow feature network, wherein a time sequence feature flow adopts an adaptive time window to dynamically segment a time sequence and extract a time sequence feature representation of an event, and a spatial feature flow adopts a density clustering to adaptively partition a geographic location and extract a spatial distribution feature, includes: constructing a time sequence feature flow, which identifies bidding time points by designing a time sensitivity function for the bidding time data, constructs a non-uniform time window sequence based on the time points, determines boundaries of the non-uniform time window by maximizing a difference in bidding behavior patterns between windows, and extracts time sequence dependency features of bidding events in each time window by using an attention-enhanced recurrent network to obtain a hierarchical time sequence feature representation; constructing a spatial feature flow, which dynamically divides bidding regions by using an adaptive spatial partitioning algorithm for the bidding geographic location data, optimizes the adaptive spatial partitioning algorithm by minimizing a bidding behavior entropy within a region and maximizing a difference in bidding behavior between regions, constructs a bidding geographic correlation network based on the spatial partitioning, and extracts spatial distribution features; ​The outputs of the time sequence feature flow and the space feature flow are fused by adopting a space-time collaborative attention mechanism. The space-time collaborative attention mechanism dynamically calculates the correlation strength of the hierarchical time sequence feature representation and the space distribution feature, adaptively adjusts the weights of different dimension features, and obtains a unified space-time feature representation.

[0023] Exemplarily, the embodiment constructs a space-time double-flow feature network, including a time sequence feature flow and a space feature flow, and a space-time collaborative attention fusion mechanism.

[0024] The time sequence feature flow processes the bidding time data, and a time sensitivity function is defined as the weighted combination of the bidding frequency and the bidding price change amplitude within a specific time window. For a given bidding data set, the sensitivity value S(t) at each time point t is calculated. When S(t) is greater than a preset threshold a (a can be set to 0.65), t is marked as a key time point. For example, in a certain engineering bidding project, it is detected that the sensitivity values at the time points of 24 hours, 12 hours and 2 hours before the bidding deadline are 0.72, 0.81 and 0.93 respectively, all of which exceed the threshold 0.65, so these three time points are marked as key time points.

[0025] Based on the identified key time points, a non-uniform time window sequence is constructed. The window boundary is determined by iterative optimization, and the goal is to maximize the difference between the bidding behavior patterns of adjacent windows. The difference is calculated by multi-dimensional indicators such as bidding frequency, price fluctuation and participant change. In the implementation process, the sliding window technology is adopted, and the window size is dynamically adjusted according to the data density, ranging from 2 hours to 24 hours. For the aforementioned engineering bidding project, four non-uniform time windows are finally determined: [bidding start, 24 hours before the deadline), [24 hours before the deadline, 12 hours before the deadline), [12 hours before the deadline, 2 hours before the deadline) and [2 hours before the deadline, bidding deadline].

[0026] In each time window, the attention-enhanced recurrent network is composed of a bidirectional long short-term memory network and a self-attention mechanism. The input features include bidding price, bidding time interval, bidder historical behavior, etc. The network contains 3 layers of bidirectional LSTM, each layer has 128 hidden units, and residual connection is set in the middle to avoid gradient disappearance problem. The self-attention mechanism dynamically adjusts the feature weights by calculating the correlation between different bidding events. The processed result is a hierarchical time sequence feature representation with a dimension of 256. Taking a certain bidding project as an example, in the time window of 2 hours before the deadline, the pattern of multiple bidders densely modifying the bidding price in a short time is captured, and the attention mechanism automatically increases the feature weight of this period to 0.87.

[0027] The spatial feature flow processes the bid location data. The adaptive spatial partitioning algorithm, based on the principle of density clustering, optimizes by minimizing the entropy of bidding behavior within a region and maximizing the variance of bidding behavior between regions. Bidding geographic coordinates are converted to a standardized format, and then the initial number of partitions is set to K (K can be set to 5-8, depending on the data size). The algorithm iteratively performs the clustering process, adjusting the cluster center and radius with each iteration until convergence or the maximum number of iterations (typically set to 100) is reached. For example, in a regional construction project bidding process, 300 bidding points were spatially partitioned with an initial setting of K=6. After 78 iterations, convergence was achieved, ultimately identifying five valid spatial partitions. The entropy values ​​of bidding behavior within each partition were 0.32, 0.41, 0.28, 0.36, and 0.39, respectively, and the average variance of behavior between partitions reached 0.76.

[0028] Based on the spatial partitioning results, a bidding geographic correlation network was constructed. The network nodes represent spatial partitions, and the edges indicate the strength of bidding relationships between partitions. The strength of relationships was calculated by the frequency and similarity of bidder activity across different partitions. The network was constructed using a graph convolution operation, consisting of two layers with 64 hidden units each. This process extracted spatial distribution features with a dimension of 128. In the aforementioned construction project example, an unusually high correlation of 0.92 was identified between partitions 2 and 4. Further analysis revealed a pattern of suspected collusive behavior among bidders in these two areas.

[0029] The spatiotemporal collaborative attention mechanism adaptively adjusts the weights of features of different dimensions by calculating the correlation strength matrix between the hierarchical temporal feature representation and the spatial distribution features. Temporal features (256 dimensions) and spatial features (128 dimensions) are projected into the same feature space (192 dimensions) through a linear transformation. An attention score is then calculated between the two features, and the attention weights are normalized using a softmax function. Finally, the two features are weighted and fused to form a unified spatiotemporal feature representation (192 dimensions).

[0030] The spatiotemporal dual-stream feature network is trained in an end-to-end manner, and the loss function combines classification loss and contrastive learning loss.

[0031] This invention, through the design of adaptive time windows and spatial partitioning, addresses the inability of traditional fixed windows to cope with dynamic changes in bidding behavior and overcomes the drawback of static spatial partitioning, which lacks the ability to capture regional features. It can automatically adjust feature extraction strategies based on the temporal and spatial variations in bidding behavior, significantly improving the accuracy of identifying abnormal patterns while reducing the workload of manual parameter adjustment.

[0032] In an alternative embodiment, a knowledge graph is constructed based on the output of the spatio-temporal dual-stream feature network, the knowledge graph comprising nodes representing the bidding events and geographical locations and edges representing the temporal dependency and spatial constraint relationships, the step of supplementing the attributes of the edges by the auxiliary data comprising: obtaining a fusion feature of the spatio-temporal dual-stream feature network, the fusion feature comprising a temporal feature representation of the bidding events and a spatial distribution feature of the geographical locations; constructing a node set of the knowledge graph, the node set comprising bidding event nodes, geographical location nodes and bidding subject nodes, the bidding event nodes comprising event identifiers, occurrence times and bidding amount attributes, the geographical location nodes comprising city area division attributes, and the bidding subject nodes comprising subject identifiers, registered capital and establishment time attributes; mapping the temporal component and the spatial component of the fusion feature to the feature space of the corresponding nodes by a nonlinear mapping function to obtain node feature vectors; constructing an edge set of the knowledge graph, the edge set comprising temporal dependency edges, spatial constraint edges and subject association edges, the weight of the temporal dependency edges being obtained by calculating the inner product of the feature vectors of adjacent nodes, and the weight of the spatial constraint edges being obtained by calculating the distance between the geographical location nodes; converting the bidding text content into a text semantic vector, fusing the weight vector of the edges with the text semantic vector to obtain a complete attribute vector of the edges, optimizing the edge set based on the importance score of the complete attribute vector, and ensuring the spatio-temporal consistency of the edge attributes by setting the proportion coefficient of the temporal feature weight and the spatial feature weight.

[0033] For example, a fusion feature of the spatio-temporal dual-stream feature network is obtained, the fusion feature comprising a temporal feature representation of the bidding events and a spatial distribution feature of the geographical locations. According to the output of the spatio-temporal dual-stream feature network, the dimension of this fusion feature is 192.

[0034] In constructing the node set of the knowledge graph, three types of nodes are defined: bid event nodes, geographic location nodes, and bid subject nodes. For bid event nodes, a unique identification code is assigned to each event, such as "Bid20230401001", and the event occurrence time (e.g., "2023-04-0110:30:00") and bid amount (e.g., "5,680,000 yuan") are recorded. For geographic location nodes, the city division is refined to the county level, and the boundary coordinate information is recorded, such as the boundary coordinate point set of a certain district is {(120.15, 30.28), (120.25, 30.28), (120.25, 30.38), (120.15, 30.38)}. For bid subject nodes, the subject identification (e.g., "Corp20230001"), registered capital (e.g., "1000 million yuan"), and establishment time (e.g., "2010-05-12") are recorded.

[0035] The temporal and spatial components of the fusion features are mapped to the feature space of the corresponding nodes through a nonlinear mapping function. A multilayer perceptron is used as the nonlinear mapping function, with a 192-dimensional fusion feature vector as input, a hidden layer dimension of 128, and an output dimension of 64. For bid event nodes, the part of the fusion features related to time (about 60% weight) is mainly used for mapping; for geographic location nodes, the part related to space (about 40% weight) is mainly used for mapping; for bid subject nodes, all 192-dimensional information is comprehensively used. The 64-dimensional vector obtained after mapping is the feature vector of the node, which is used for subsequent relationship modeling.

[0036] In constructing the edge set of the knowledge graph, three types of edges are defined: temporal dependence edges, spatial constraint edges, and subject association edges. For temporal dependence edges, the inner product of the feature vectors of adjacent bid event nodes is calculated to obtain the weight value. For example, for two bid events with a time interval of no more than 30 days, if the inner product of their feature vectors is greater than 0.7, a temporal dependence edge is established, and the weight value is the inner product value. For spatial constraint edges, the Euclidean distance between geographic location nodes is calculated, and the smaller the distance, the greater the weight. Specifically, the weight value is equal to 1 minus the normalized distance value, for example, the weight of adjacent regions is usually above 0.8. For subject association edges, the bid events participated by the bid subject are connected, and the weight value is calculated based on the number of participations and the amount.

[0037] The pre-trained text encoder is used to process the bidding documents, extract key information and convert it into a 384-dimensional text semantic vector. The edge weight vector has a dimension of 32, which is obtained by expanding the weight value of the edge. The 32-dimensional weight vector is spliced with the 384-dimensional text semantic vector to obtain a complete attribute vector with a dimension of 416. When optimizing the edge set based on the importance score of the complete attribute vector, a multilayer perceptron is used to calculate the importance score of each edge, with the input being the 416-dimensional complete attribute vector and the output being a 1-dimensional importance score. Edges with an importance score greater than 0.6 are retained, and edges with lower scores are discarded, thereby optimizing the structure of the knowledge graph. For example, a knowledge graph containing 1000 nodes initially has 4500 edges, and after optimization, 2800 high-importance edges are retained.

[0038] To ensure the spatio-temporal consistency of edge attributes, the proportion coefficient of the temporal feature weight and the spatial feature weight is set. In actual application, based on historical data analysis and the design of the spatio-temporal double-flow feature network, the proportion of the temporal feature weight and the spatial feature weight is set to 6:4, that is, in the calculation of the complete attribute vector, the temporal feature contributes 60% of the weight, and the spatial feature contributes 40% of the weight. This proportion is consistent with the weight distribution of the spatio-temporal collaborative attention mechanism, ensuring the technical coherence from feature extraction to knowledge graph construction. This proportion can be adjusted according to specific application scenarios to adapt to the characteristics of different regions and industries.

[0039] The construction scheme of the knowledge graph effectively solves the problem of the separation of temporal features, spatial features and text features in traditional methods, and overcomes the information loss caused by single feature representation. A complete multi-modal feature correlation system is established, which greatly improves the completeness of feature representation, enables the subsequent anomaly detection to consider multi-dimensional information comprehensively, reduces the false positive rate, and improves the reliability of detection.

[0040] In an optional implementation, based on the knowledge graph, the steps of calculating the similarity between the node features and the normal mode library using a spatio-temporal mode encoder at a local level to obtain an anomaly score, and using a graph attention network to optimize the context of the anomaly score at a global level to obtain an anomaly detection result include: Obtain node features and neighborhood information in the knowledge graph, wherein the node features include temporal features and spatial features, and the neighborhood information includes the dependency relationship between nodes; Encode the node features and the neighborhood information using a spatio-temporal mode encoder to obtain a spatio-temporal encoding representation of the node, construct a normal mode feature library based on the spatio-temporal encoding representation, and the normal mode feature library includes a plurality of normal mode center representations; calculate the similarity between the spatio-temporal encoding representation of the node and the normal mode center representations to obtain a local anomaly score of the node; The multi-head graph attention network is constructed, attention weights between nodes are calculated based on a query matrix, a key matrix and a value matrix, node features are weighted and aggregated according to the attention weights, and a context representation of the node is obtained; The local anomaly score of the node is fused with the context representation of the node to obtain a global anomaly score of the node; A time sequence smoothing constraint is constructed based on a time sequence dependency relationship, the time sequence smoothing constraint is realized by minimizing the difference between global anomaly scores of time sequence related nodes, the global anomaly scores are optimized by propagation according to the time sequence dependency strength, and a final anomaly score of the node is obtained; a detection threshold is determined according to the final anomaly score and abnormal marking is performed to obtain an anomaly detection result.

[0041] For example, in combination with Figure 2 Based on the knowledge graph, the spatial-temporal anomaly detection flowchart is described: the node features and neighborhood information in the knowledge graph are obtained. The node features are represented by a 64-dimensional vector, including time sequence features and spatial features. The time sequence features occupy the first 38 dimensions of the vector, mainly describing the time pattern features of the bidding event, such as bidding time point distribution, bidding frequency change, etc. The spatial features occupy the last 26 dimensions of the vector, mainly describing the distribution characteristics of the bidding geographical location, such as location aggregation degree, spatial coverage, etc. The neighborhood information represents the dependency relationship between nodes, which is stored in an adjacency matrix. The matrix element value represents the dependency strength, which ranges from 0 to 1. For example, for a node set in a bidding project knowledge graph, a feature matrix (500x64) containing 500 nodes and a corresponding adjacency matrix (500x500) are extracted.

[0042] The spatial-temporal pattern encoder adopts a graph convolution network structure, which includes three layers of graph convolution layers, and the output dimensions of each layer are 128, 96 and 64 respectively. In each layer of graph convolution operation, the node features are updated by weighted combination of the node features and the neighborhood node features. The weight coefficient is determined by the dependency strength between nodes. The graph convolution operation is followed by a nonlinear activation function. The encoder outputs a 64-dimensional node spatial-temporal encoding representation. For example, 300 bidding projects in a certain region are encoded to obtain a 300x64-dimensional encoding matrix.

[0043] A normal pattern feature library is constructed based on the node spatial-temporal encoding representation. The feature library is constructed by a clustering algorithm. First, 5000 confirmed normal bidding event samples are selected from historical data. The spatial-temporal encoding representation of these samples is applied to the clustering algorithm, and the number of clusters is set to 20 to obtain 20 cluster centers. Each cluster center is a 64-dimensional vector representing a normal bidding behavior pattern. For example, one of the cluster centers represents the normal pattern of "multiple bidders uniformly distributing the submission of bidding documents before the bidding deadline" in large infrastructure projects. The 20 cluster center vectors form the normal pattern feature library.

[0044] The similarity calculation between the spatio-temporal encoding of a node and the normal pattern center representation adopts the cosine similarity method. The cosine similarity between the 64-dimensional encoding vector of a node and the 20 normal pattern center vectors in the feature library is calculated respectively, and the maximum similarity value is taken as the matching degree of the node and the normal pattern. The local anomaly score is defined as 1 minus the matching degree, so the higher the score value, the more abnormal the node. For example, the local anomaly scores of 100 test nodes are calculated, and the scores are distributed between 0.05 and 0.92, of which 80% of the nodes have scores less than 0.4, indicating that most of the nodes conform to the normal pattern.

[0045] A multi-head graph attention network is constructed for context optimization of the anomaly score. The network includes 4 attention heads, each of which includes a set of query matrix, key matrix and value matrix. The dimensions of the query matrix and the key matrix are 64x32, and the dimension of the value matrix is 64x16. For each node, the feature vector thereof is transformed by the query matrix to obtain a query vector, and the feature vectors of the neighborhood nodes thereof are transformed by the key matrix to obtain a set of key vectors. The dot product operation of the query vector and the set of key vectors is performed to calculate the attention score, and the attention weight is obtained after normalization processing. The attention weight reflects the attention degree of the center node to the neighborhood nodes. The feature vectors of the neighborhood nodes are transformed by the value matrix to obtain a set of value vectors, and the set of value vectors is weighted and summed according to the attention weight to obtain the output of the attention head. The outputs of the 4 attention heads are spliced and then linearly transformed to obtain the context representation of the node, with a dimension of 64. In actual application, it is found that the attention weights of an abnormal node in a certain bidding network to 3 neighborhood nodes are 0.42, 0.35 and 0.23 respectively, indicating that the abnormal state of the node is associated with these neighborhood nodes.

[0046] The local anomaly score of a node is fused with the context representation to obtain the global anomaly score of the node. The fusion process is realized through a gating mechanism, which calculates the correlation between the local anomaly score and the context representation to generate a fusion weight. The global anomaly score is the weighted sum of the local anomaly score and the context adjustment term. The context adjustment term is generated by the context representation passing through two fully connected networks, representing the influence of neighborhood information on anomaly judgment. In a certain case, a node with a local anomaly score of 0.75, after considering that most of its neighborhood nodes are normal nodes, the global anomaly score is adjusted to 0.62, reflecting the smoothing effect of context information.

[0047] Identify node pairs with time sequence correlation in the knowledge graph, such as bidding behaviors of the same bidding subject at different time points or adjacent bidding events in time. Build a time sequence dependency graph for these node pairs, and the weight of the edge in the graph represents the time sequence dependency strength. Time sequence smoothing constraint is realized by minimizing the difference of global anomaly scores between time sequence related nodes. For strongly dependent node pairs, their anomaly scores should be closer; for weakly dependent node pairs, a larger difference in anomaly scores is allowed. In a certain bidding sequence, 5 time sequence correlation nodes are identified, and the time sequence dependency strength matrix shows that the dependency strength between node 1 and node 2 is 0.85, and the dependency strength between node 2 and node 3 is 0.72, indicating that these nodes are closely related in time sequence.

[0048] According to the time sequence dependency strength, the global anomaly score is optimized by propagation. The propagation process adopts an iterative method, and the initial value is the global anomaly score of the node. In each iteration, the anomaly score of the node is updated by the weighted sum of its own score and the score of the neighbor node, and the weight coefficient is the time sequence dependency strength. The number of iterations is set to 10 to ensure convergence. After each iteration, a residual connection is introduced to retain the original anomaly score information. After propagation optimization, the final anomaly score of the node is obtained. In a certain bidding sequence case, the global anomaly scores of the two adjacent bidding event nodes are 0.82 and 0.35 respectively, and after propagation optimization, they are adjusted to 0.78 and 0.45, which reflects the effect of time sequence smoothing.

[0049] According to the final anomaly score, a detection threshold is determined and an anomaly is marked. The threshold is determined based on historical data statistics and business requirements, and is usually set to the 95th percentile of the anomaly score distribution. On a certain regional bidding data set, the final anomaly score distribution of 2000 nodes is analyzed, and the threshold is determined to be 0.68. Nodes with scores exceeding the threshold are marked as anomalies, generating an anomaly detection result report. The report includes anomaly node identification, anomaly score, associated node information and preliminary anomaly type judgment. In one review, 27 abnormal events were detected from 500 bidding events, of which 23 were confirmed as real irregular behaviors in subsequent verification, and 4 were false positives.

[0050] The multi-level anomaly detection method of the application combines local features and global context, and realizes accurate identification of bidding anomalies through spatio-temporal pattern coding and graph attention mechanism. This method not only can detect single abnormal events, but also can find abnormal propagation paths and correlation patterns, significantly improving the efficiency and accuracy of bidding compliance review.

[0051] In an optional implementation, the step of constructing a time sequence smoothing constraint based on a time sequence dependency relationship, the time sequence smoothing constraint being realized by minimizing the difference of global anomaly scores between time sequence related nodes, and optimizing the global anomaly scores according to time sequence dependency strength to obtain final anomaly scores of nodes comprises: The time sequence dependency is constructed, and a dynamic graph neural network is used to calculate the time sequence dependency strength of the time sequence related node pair, the dynamic graph neural network comprising a time sequence query matrix and a time sequence key matrix, the node feature vector being transformed through the time sequence query matrix and the time sequence key matrix, and the time sequence dependency strength of the node pair being calculated in combination with a time interval decay function; A smooth constraint objective function is constructed based on the time sequence dependency strength, the smooth constraint objective function comprising a smooth loss term and a causal loss term, the smooth loss term being based on the time sequence dependency strength to calculate the square difference of the global anomaly score of the related node, and the causal loss term being based on the intervention operation to calculate the deviation of the global anomaly score of the node from the expected value; An anomaly score propagation function is designed, the global anomaly score of the node is iteratively updated according to the time sequence dependency strength, and the smooth constraint objective function is taken as an optimization objective, the global anomaly score is optimized through a gradient descent method, and a penalty term is introduced in each iteration to constrain the change amplitude of the global anomaly score; The optimized global anomaly score is fused in multiple scales, and the final anomaly score of the node is calculated based on the global anomaly score and the corresponding weight of different scales.

[0052] For example, the time sequence dependency is constructed, and the time sequence related node pair in the time dimension is identified. For a given knowledge graph, all nodes with time attributes are extracted, including the bidding event nodes and the related time label nodes. For each time sequence related node pair, a dynamic graph neural network is used to calculate the time sequence dependency strength. The dynamic graph neural network has special time sequence perception capability, and the core components include a time sequence query matrix and a time sequence key matrix. The time sequence query matrix has a dimension of 64x32 and is used to transform the feature vector of the source node into a query vector; the time sequence key matrix also has a dimension of 64x32 and is used to transform the feature vector of the target node into a key vector. In actual application, for two time points A and B of a bidding project, the node feature vectors are 64-dimensional vectors VA and VB, respectively, VA is converted into a 32-dimensional query vector QA through the time sequence query matrix, and VB is converted into a 32-dimensional key vector KB through the time sequence key matrix.

[0053] The specific implementation of the time interval decay function adopts an exponential decay form. For two nodes with timestamps tA and tB, the time interval is calculated as |tB-tA|, in days. The decay function is defined as D(tA,tB)=alpha^(|tB-tA| / beta), where alpha is the basic decay rate, set to 0.9, and beta is the time scale parameter, set to 10 days. This means that when the time interval is 10 days, the decay value is 0.9; when the interval is 20 days, the decay value is 0.81 (i.e. 0.9 2). The parameters alpha and beta can be adjusted according to specific business scenarios. When two bidding events occur on the same day, the decay value is 1; when the interval exceeds 90 days, the decay value approaches 0.1, indicating that the long-term dependence is significantly weakened. When calculating the timing dependence strength, first calculate the dot product of the query vector QA and the key vector KB to obtain the similarity score S, then multiply S by the time decay value D, and finally map the result to 0 to 1 through the Sigmoid function to obtain the final timing dependence strength value. In a certain bidding data set, the bidding events within three consecutive months were analyzed, and a complete dependence strength matrix was generated, in which the dependence strength of consecutive bidding events of the same bidder within a week was 0.85 on average, while the dependence strength of events adjacent in time between different bidders was only about 0.32.

[0054] The smoothing constraint objective function includes two key components: the smoothing loss term and the causal loss term. The specific calculation method of the smoothing loss term is: for all node pairs (i, j) with timing dependence in the knowledge graph, calculate the weighted square difference of their global anomaly scores Si and Sj. The calculation formula of the weighted square difference is: the timing dependence strength Wij of the node pair (i, j) multiplied by the square of the anomaly score difference (Si-Sj) 2 . The sum of the weighted square differences of all node pairs constitutes the smoothing loss term. For example, the smoothing loss is calculated for 5 associated bidding events in a certain bidding project, where the dependence strength of node pair (1, 2) is 0.9, and the anomaly scores are 0.78 and 0.65, respectively, contributing a smoothing loss of 0.9 x (0.78-0.65) 2 =0.015; while the node pair (3, 4) with a dependence strength of only 0.3, even with the same anomaly score difference, contributes only 0.005, which is only one-third of the strong dependence pair. The causal loss term is implemented using the counterfactual analysis method. First, build an anomaly propagation prediction model that accepts node features and anomaly scores as input to predict the anomaly score changes of timing-related nodes. For node i, by replacing its features with the average features of normal samples, simulate the scenario where it becomes normal, and then predict the expected anomaly scores E[Sj] of all timing-related nodes j. The causal loss term calculates the squared difference between the actual anomaly score Sj and the expected anomaly score E[Sj], and is weighted according to the timing dependence strength of the node pair (i, j). This design enables the differentiation between true causal relationships and accidental correlations. In a certain bidding anomaly detection, an intervention was made on a node with an anomaly score of 0.86, and it was found that the expected anomaly scores of the three subsequent nodes directly affected by it decreased by 0.32, 0.28, and 0.15, respectively, while the anomaly scores of the nodes with no direct timing association with it remained almost unchanged, clearly showing the causal path of anomaly propagation.

[0055] The specific implementation of the anomaly score propagation function adopts an iterative message passing mechanism. For node i, the update formula of its anomaly score in the t+1th iteration is: Si(t+1) = lambda * Si(t) + (1-lambda) * weighted average value. Where Si(t) is the anomaly score of node i in the tth iteration, lambda is the retention coefficient, set to 0.6, representing retaining 60% of the original score. The weighted average value is calculated by weighting the anomaly scores Sj(t) of all nodes j that have a temporal dependency with node i according to the dependency strength Wij. In specific calculation, first normalize the dependency strength Wij of all related nodes j so that their sum is 1, then calculate the sum of the product of the normalized weight and the corresponding node anomaly score. Calculate the objective function value after each iteration, and adjust the anomaly score according to the gradient direction. To prevent over-smoothing, introduce a penalty term delta * |Si(t+1) - Si(0)| 2 , where Si(0) is the initial anomaly score, and delta is the penalty coefficient, with an initial value of 0.1, which is reduced by half every 5 iterations until it drops to 0.01. In the anomaly analysis of a certain bidding project, 120 nodes were optimized for 25 iterations, and the maximum anomaly score change was observed to decrease from the initial 0.23 to 0.02 in the 25th iteration, indicating that the iteration process has basically converged.

[0056] The specific implementation of multi-scale fusion involves three time scales: short-term (7 days), medium-term (30 days), and long-term (90 days). At each time scale, a temporal dependency network is independently constructed and the anomaly score is optimized. The short-term scale focuses on recent bidding behavior patterns, using a higher time decay rate a = 0.8; the medium-term scale focuses on monthly bidding patterns, with a decay rate a = 0.9; the long-term scale focuses on quarterly bidding trends, with a decay rate a = 0.95. For each node, obtain the optimized anomaly scores Ss, Sm, and Sl at the three time scales, and the final anomaly score is calculated as: S = ws * Ss + wm * Sm + wl * Sl, where ws, wm, and wl are the weight coefficients of each scale, determined by grid search on the validation set, with typical values of ws = 0.5, wm = 0.3, and wl = 0.2. For example, in the bidding analysis of a certain construction project, a bidding enterprise was identified as highly abnormal in the short term (short-term score 0.88), but with lower medium-term and long-term anomaly scores (0.42 and 0.35 respectively), the final score after multi-scale fusion was 0.62, lower than the original detection threshold 0.7, avoiding possible false positives. It also supports adaptive weight adjustment, automatically adjusting the weights of the three scales according to the temporal characteristics of the node. For example, for a newly established bidding subject, the short-term and medium-term weights are automatically increased due to the lack of long-term data, while for a long-term active bidding subject, the weights of the three scales are more balanced.

[0057] For example, the analysis of all bidding records for a certain bidding project. There are 52 bidding event nodes in this project, forming a complex time-dependent network. The complete time-dependent intensity matrix is calculated by dynamic graph neural network, and the intensity value is distributed between 0.05 and 0.92. Based on the dependence matrix, a smooth constraint objective function is constructed, and the global anomaly score of the node is adjusted through iterative optimization. Before optimization, there are many cases where the anomaly scores of bidding events at adjacent time points differ greatly, such as two events with an interval of only one day, with scores of 0.78 and 0.31 respectively; After optimization, the scores of these nodes are adjusted to 0.72 and 0.45, which reflects the effect of time smoothing. After multi-scale fusion of the final output anomaly score, 7 highly suspicious bidding events are detected, which show a clear serial pattern in time, suggesting possible bid-rigging behavior.

[0058] The present application realizes accurate identification of bidding and tendering abnormal behavior through time-dependent relationship and smooth constraint, overcoming the defects of traditional methods that are difficult to capture time patterns and associated anomalies. This technology effectively reduces the fluctuation of anomaly scores in the time dimension, improves the stability and reliability of the detection results. The multi-scale fusion mechanism enables comprehensive consideration of abnormal features in different time ranges, balancing short-term fluctuations and long-term trends, and providing more comprehensive and accurate anomaly detection results for bidding and tendering supervision.

[0059] In an optional implementation, based on deep reinforcement learning, the anomaly verification is modeled as a sequential decision-making process, a bidding and tendering compliance verification path is dynamically generated based on the anomaly detection results and the auxiliary data, and a compliance evaluation result is obtained by optimizing the verification path. The step of generating the review report comprises: Based on the anomaly detection results, a verification feature vector is constructed, which combines the auxiliary data to construct a verification state space and a verification action space. The verification state space includes the verification feature vector and the historical verification decision sequence; Based on the verification state space and the verification action space, a verification policy generation network is constructed. The verification policy generation network uses a variational autoencoder to encode the verification feature vector to obtain a latent representation of the anomaly pattern. Based on the latent representation, a verification policy distribution is constructed, and a verification action sequence is generated according to the verification policy distribution; A verification evaluation function is constructed, which includes a verification accuracy term, a verification cost term and a decision consistency term. The verification evaluation function is optimized by Thompson sampling method to obtain the optimal verification action sequence, and the verification path is determined according to the optimal verification action sequence; The verification process is performed along the verification path, a time sequence compliance index, a spatial consistency index and a text semantic similarity index are calculated for each verification node, a multi-dimensional risk dependence matrix is constructed based on the time sequence compliance index, the spatial consistency index and the text semantic similarity index, a multi-dimensional risk joint distribution is calculated using a Copula function, and a compliance evaluation result is output; based on the compliance evaluation result, abnormal spatio-temporal pattern information and its propagation path are extracted, and an inspection report is generated.

[0060] For example, a verification feature vector is constructed based on the abnormal detection result. The verification feature vector includes three parts: abnormal score feature, time sequence pattern feature and entity relationship feature, with a total of 128 dimensions. The abnormal score feature occupies 32 dimensions, including local abnormal score, global abnormal score and final abnormal score of the node; the time sequence pattern feature occupies 48 dimensions, describing the time distribution characteristics and change trend of the abnormal node; and the entity relationship feature occupies 48 dimensions, describing the interaction mode of the abnormal node and other entities. For example, in a certain bidding project, 7 highly suspicious events are detected from 52 bidding events, and a 128-dimensional verification feature vector is constructed for each suspicious event. In addition, relevant auxiliary data is collected, including bidding document information, enterprise registration information, qualification certificate information and historical bid-winning records, etc. These auxiliary data are stored in a structured manner to enrich the verification features.

[0061] Based on the verification feature vector and the auxiliary data, a verification state space and a verification action space are constructed. The verification state space is composed of the current verification feature vector and the historical verification decision sequence. The historical verification decision sequence records the executed verification actions and their results, and the upper limit of the length is set to 10. In the abnormal verification process of a certain bidding project, the dimension of the verification state space is 256, of which 128 dimensions are the verification feature vector and 128 dimensions are the encoded representation of the historical verification decision sequence. The verification action space includes multiple verification operation types, such as document consistency check, qualification authenticity verification, bidding timeline analysis, enterprise association relationship query, etc., and a total of 15 basic verification actions are designed. Each verification action has different execution cost and expected accuracy. For example, the execution cost of bidding file similarity analysis is 2 units, and the expected accuracy is 0.8; while the execution cost of enterprise on-site visit is 10 units, and the expected accuracy is 0.95. In actual application, the priority of the verification action will be dynamically adjusted according to the abnormal type and severity.

[0062] A verification strategy generation network is constructed based on the verification state space and the verification action space. The network uses a variational autoencoder to encode the verification feature vectors to obtain the latent representation of the abnormal pattern. The variational autoencoder consists of an encoder and a decoder. The encoder includes three fully connected networks, with the input layer dimension of 128, the hidden layer dimensions of 96 and 64, and the output layer dimension of 32, generating a mean vector and a variance vector each with 16 dimensions. The decoder also includes three fully connected networks, with the structure symmetric to the encoder. Through the reparameterization trick, a 16-dimensional latent representation vector is sampled from the distribution defined by the mean and variance. For example, in a certain bidding abnormality verification case, the verification feature vectors of 7 suspicious events are encoded to obtain 7 latent representation vectors. The clustering of these vectors in the latent space reflects the similarity of the abnormal patterns. Based on the latent representation, a verification strategy distribution is constructed. The distribution is realized through a strategy network consisting of two fully connected networks, with the input layer dimension of 16, the hidden layer dimension of 32, and the output layer dimension of 15, corresponding to the probability distribution of 15 verification actions. In each verification step, the next verification action is sampled from the strategy distribution to form a verification action sequence. For a suspicious bidding record of a certain bidding project, a sequence of 5 verification actions is generated: bidding document similarity analysis, bidder qualification inspection, bidding timeline comparison, enterprise association relationship analysis, and historical bidding pattern comparison.

[0063] A verification evaluation function is constructed, including three parts: verification accuracy term, verification cost term, and decision consistency term. The verification accuracy term reflects the expected accuracy of the verification action sequence, and the calculation method is the weighted average of the accuracy of each verification action, and the weight is determined by the action importance. The verification cost term considers the total cost of executing the verification action sequence, including time cost, labor cost and resource cost. The decision consistency term evaluates the internal consistency of the verification action sequence, which is realized by calculating the semantic correlation of adjacent verification actions. In a certain bidding verification case, two candidate verification sequences are evaluated: the accuracy of sequence A is 0.85, the cost is 12, and the consistency is 0.78; the accuracy of sequence B is 0.88, the cost is 15, and the consistency is 0.82. The Thompson sampling method is used to optimize the verification evaluation function. This method maintains the evaluation distribution of each verification action sequence, and in each optimization iteration, samples are taken from these distributions to select the sequence with the highest evaluation value. By continuously updating the evaluation distribution parameters, a balance between exploration and exploitation is achieved. After 50 rounds of iterative optimization, the optimal verification action sequence is obtained, and the verification path is determined accordingly. For the above bidding case, the optimal verification path is selected as a compromise solution with an accuracy of 0.87, a cost of 13, and a consistency of 0.81.

[0064] Along the determined verification path, the verification process is performed, and three core indicators are calculated for each verification node. The timing compliance indicator assesses the compliance degree of bidding behavior in the time dimension, including bidding time point distribution, modification frequency, and submission timing, etc. Timing compliance is calculated by comparing the deviation degree of the actual timing mode and the normal baseline mode. The larger the deviation degree, the lower the compliance. For example, in a certain bidding project, the submission time distribution of the bidding document was analyzed, and it was found that the submission time of 3 suspicious bidders was highly concentrated within 15 minutes before the deadline, and the timing compliance indicator was less than 0.4, while the timing compliance indicator of normal bidding was usually higher than 0.7. The spatial consistency indicator assesses the rationality of bidding behavior in the spatial dimension, including bidding location distribution, relationship between office address and registered address, etc. The indicator is calculated through consistency analysis of geographic location information. In the verification process, it was found that although two bidding companies had different registered addresses, they actually used the same network environment to submit bidding documents, and the spatial consistency indicator was only 0.32. The text semantic similarity indicator assesses the independence of the bidding document in content, and calculates the semantic similarity by analyzing the document structure, wording characteristics and format style. Analysis of 5 bidding documents found that the semantic similarity of 3 of them in the specific technical solution part was as high as 0.88, much higher than the industry average level of 0.45, indicating that there may be bidding collusion behavior.

[0065] Based on the above three indicators, a multi-dimensional risk dependence matrix is constructed to describe the correlation between different risk indicators. The matrix is a 3x3 symmetric matrix, the diagonal elements are 1, and the non-diagonal elements represent the dependence strength between the corresponding risk indicators. In the verification of a certain bidding project, the dependence strength of timing compliance and spatial consistency is 0.65, the dependence strength of timing compliance and text semantic similarity is 0.42, and the dependence strength of spatial consistency and text semantic similarity is 0.58. The Copula function is used to calculate the multi-dimensional risk joint distribution. Copula function can capture the complex dependence structure between multiple risk indicators, even if the marginal distribution of each indicator is different. According to the historical verification data, the parameters of the Copula function are estimated, and then the probability value of the current verification node under the joint distribution is calculated, which is converted into the final compliance score. In the verification of a certain bidding project, 7 suspicious events were evaluated, and the compliance score distribution was between 0.25 and 0.78, of which 3 events scored less than 0.3, and were judged as high-risk violation events.

[0066] Based on the compliance evaluation results, abnormal spatio-temporal pattern information and its propagation path are extracted, and an audit report is generated. The audit report contains five main parts: abnormal event overview, abnormal spatio-temporal pattern description, verification path and key evidence, compliance evaluation results and recommended treatment scheme. In the abnormal spatio-temporal pattern description part, the time distribution and spatial correlation of abnormal events are displayed through visualization technology, and the abnormal propagation path is highlighted. For example, in the audit report of a certain bidding project, a "turnaround bid" pattern is identified, and three associated enterprises show obvious turnaround bid characteristics in 8 projects within six months. The report details the formation process and characteristics of this pattern. The verification path and key evidence part lists the sequence of verification actions performed and the key evidence found, such as hidden association between enterprises, abnormal similarity data of bid documents, etc. The compliance evaluation results part gives the final compliance score and risk level, and explains the contribution of each risk indicator. The recommended treatment scheme part proposes appropriate treatment suggestions according to the type and severity of the violation, such as canceling the bid-winning qualification, suspending the bid qualification, or including in the key monitoring list, etc.

[0067] The present application models abnormal verification as a sequential decision-making process, dynamically generates the optimal verification strategy through deep reinforcement learning, and realizes accurate verification of bidding abnormal behavior. This method integrates multi-dimensional risk index to evaluate compliance, not only effectively identifies single violation behavior, but also reveals complex violation patterns and propagation paths. By optimizing the verification path, the verification efficiency and accuracy are significantly improved, and the verification cost is reduced, providing an interpretable and operable decision support tool for bidding supervision.

[0068] In an optional implementation, the step of constructing a verification evaluation function including a verification accuracy term, a verification cost term and a decision consistency term, and using the Thompson sampling method to sample and optimize the verification evaluation function to obtain the optimal verification action sequence includes: The verification accuracy term is calculated based on the abnormal detection rate corresponding to the verification action sequence, the verification cost term is calculated based on the resource consumption of the verification depth and the verification range, and the decision consistency term is calculated based on the strategy difference of adjacent verification actions; The Bayesian neural network is used to implement the Thompson sampling method, the posterior distribution of the verification evaluation function is established, the weight distribution of the Bayesian neural network is trained based on the historical verification decision sequence, and the probability representation of the verification evaluation function is obtained; In each round of verification decision, an instance of the current verification evaluation function is sampled from the posterior distribution based on the Thompson sampling principle, the candidate verification action sequence is scored based on the instance of the current verification evaluation function, and the verification action sequence with the highest score is selected as the optimal verification action sequence; The posterior distribution is updated according to the verification feedback information, and exploration and utilization are balanced by adaptively adjusting a sampling temperature parameter, which is reduced with an increase in verification rounds.

[0069] In the process of implementing the present application, a verification evaluation function including a verification accuracy term, a verification cost term and a decision consistency term can be constructed, and the Thompson sampling method is used to sample and optimize the verification evaluation function, so as to obtain an optimal verification action sequence.

[0070] For example, the verification accuracy term is calculated based on the anomaly detection rate corresponding to the verification action sequence, and is evaluated by combining the detection ability of each verification action with the matching degree of the anomaly type. An anomaly type-verification action mapping matrix is maintained, and each element in the matrix represents the detection effect of a specific verification action on a specific anomaly type. For example, for the "collusion" anomaly, the detection rate of the "bid file similarity analysis" action is 0.85, the detection rate of the "bid time correlation analysis" action is 0.72, and the detection rate of the "enterprise association relationship query" action is 0.9. According to the current anomaly type distribution of the verification, the weighted detection rate of each verification action sequence is calculated. For a sequence containing five verification actions, the comprehensive detection rate is calculated as the weighted geometric mean of the detection rates of each action, and the weight is determined according to the severity of the anomaly type and the prior probability. The verification accuracy term is normalized to convert the comprehensive detection rate to a score between 0 and 1, and the higher the detection rate, the higher the score. In specific calculation, if the detection rates of the verification actions in the sequence are r1, r2,..., rn, and the corresponding weights are w1, w2,..., wn (the sum of the weights is 1), then the comprehensive detection rate r = r1 w1 × r2 w2 ×... × rn wn , and the final verification accuracy term score AccScore = r.

[0071] The verification cost item is based on the resource consumption calculation of the verification depth and the verification range. The verification depth refers to the length of the verification action sequence, and the verification range refers to the amount of data and the number of associated entities involved in the verification. A basic cost value is assigned to each verification action, and then adjusted according to the verification depth and range. For example, in a certain bid exception verification, the basic cost of "bid file similarity analysis" is 2 units, but when the verification range is expanded to involve 8 bid files, the adjusted cost is 2xlog(8+1)=4.16 units. The influence of verification depth on cost is realized through an exponential decay factor, which makes the cost of verification actions in the later stage of the sequence relatively lower, reflecting the decreasing marginal effect in the verification process. For a verification sequence of length n, the decay factor of the i-th action is 0.9^(i-1). The verification cost item is converted into a score between 0 and 1 by normalization and inversion processing, and the lower the cost, the higher the score. In specific calculation, if the basic costs of the verification actions in the sequence are c1, c2,..., cn respectively, and the verification range adjustment coefficients are s1, s2,..., sn, then the actual cost of the i-th action is ci×si×0.9^(i-1), and the total cost C is the sum of the actual costs of all actions. The final verification cost item score CostScore=max(0,1-C / Cmax), where Cmax is the preset maximum acceptable cost, usually set to 20 units.

[0072] The decision consistency item is based on the strategy difference calculation of adjacent verification actions, aiming to ensure the coherence and logic of the verification process. A verification action semantic correlation matrix is constructed, and each element in the matrix represents the correlation degree between two verification actions. For example, the correlation degree between "bid file similarity analysis" and "bid text style comparison" is 0.85, while the correlation degree with "enterprise on-site visit" is only 0.25. Decision consistency is achieved by calculating the average correlation degree of adjacent actions in the verification sequence. For a sequence of length n, the correlation degrees of n-1 pairs of adjacent actions are calculated, and their arithmetic mean is taken as the consistency score. In specific calculation, if the correlation degree of the adjacent action pair (i, i+1) in the sequence is rel(i, i+1), then the decision consistency item score ConsScore=(rel(1,2)+rel(2,3)+...+rel(n-1,n)) / (n-1). In practice, the correlation matrix is constructed based on historical verification experience and expert knowledge, with a value range of 0 to 1, and the higher the value, the more coherent the two verification actions.

[0073] The verification evaluation function is constructed by comprehensively considering the verification accuracy term, the verification cost term and the decision consistency term. The function combines the three terms by weighting, and the weights can be dynamically adjusted according to the priority of the verification task. In the regular verification, the weights of the three terms are set to 0.5, 0.3 and 0.2; in the resource-limited case, the weights of the three terms are set to 0.6, 0.2 and 0.2; and in the high-risk case, the weights of the three terms are set to 0.7, 0.2 and 0.1. Taking a certain bidding abnormality verification task as an example, a sequence of 5 verification actions needs to be evaluated: "document acquisition", "enterprise association analysis", "bid text similarity comparison", "price analysis" and "historical bidding pattern analysis". The detection rates of the sequence of actions on the current abnormal type are 0.75, 0.9, 0.85, 0.65 and 0.8 respectively, and the weights are 0.15, 0.25, 0.3, 0.1 and 0.2 respectively. The comprehensive detection rate is calculated as 0.75 0.15 ×0.9 0.25 ×0.85 0.3 ×0.65 0.1 ×0.8 0.2 =0.813, so AccScore=0.813; the basic costs of the actions are 1, 3, 2, 1.5 and 2.5 respectively, the verification range adjustment coefficients are 1, 1.2, 1.5, 1 and 1.1 respectively, and the total costs considering the depth attenuation are 1×1×1+3×1.2×0.9+2×1.5×0.9 2 +1.5×1×0.9 3 +2.5×1.1×0.9 4 =8.954, assuming Cmax=20, then CostScore=1-8.954 / 20=0.552; the correlation degrees of adjacent action pairs are 0.7, 0.85, 0.6 and 0.75 respectively, and the average consistency score ConsScore=(0.7+0.85+0.6+0.75) / 4=0.725. The final verification evaluation score is 0.5×0.813+0.3×0.552+0.2×0.725=0.713. Similarly, the evaluation scores of other candidate verification sequences can be calculated, and the sequence with the highest score is selected as the optimal verification action sequence.

[0074] The Thompson sampling method is implemented using a Bayesian neural network to establish the posterior distribution of the validation evaluation function. The main difference between a Bayesian neural network and a traditional neural network is that the weights are not deterministic but have a probability distribution. The constructed Bayesian neural network contains three layers: the input layer has a dimension equal to the feature dimension of the validation sequence (usually 128 dimensions), the hidden layer contains 64 neurons, and the output layer has one neuron corresponding to the validation evaluation score. Each weight in the network has a prior distribution, usually a normal distribution with a mean of 0 and a variance of 0.1. The network is trained using historical validation decision sequences and their effect feedback to update the weight distribution. The training data includes validation sequence features, execution results, and human scores. The training uses the variational inference method to minimize the KL divergence between the true posterior distribution and the approximate posterior distribution. After 5000 iterations of training, the network weight distribution gradually converges, with the mean standard deviation decreasing from the initial 0.1 to around 0.03, indicating that the model's expression of the validation evaluation function is becoming more and more certain.

[0075] At each round of validation decision, an instance of the current validation evaluation function is sampled from the posterior distribution based on the Thompson sampling principle. Specifically, for each weight in the Bayesian neural network, a value is randomly sampled from its posterior distribution to form a complete neural network instance, which is an instance of the current validation evaluation function. This sampled function instance is used to score all candidate validation action sequences. In a certain bidding anomaly validation task, 10 candidate validation sequences are generated, and the sampled evaluation function is used to score them, with scores ranging from 0.45 to 0.82. The validation action sequence with the highest score is selected as the optimal sequence, which in this case is a sequence containing "document acquisition", "enterprise association analysis", "bid text similarity comparison", "price analysis", and "historical bidding pattern analysis", with a score of 0.82.

[0076] After executing the selected validation action sequence, the posterior distribution is updated based on the validation feedback information. The validation feedback information includes the actual detection rate, actual cost, and fluency score of the validation process. These feedback information is converted into the actual score of the validation sequence and compared with the predicted score to calculate the prediction error. Through the Bayesian update rule, the weight distribution of the Bayesian neural network is adjusted according to the prediction error. For well-performing parameters, their distribution mean moves in the effective direction and their variance decreases; for poorly performing parameters, their variance may increase, indicating increased uncertainty. In a certain bidding validation case, after executing the validation sequence, it is found that the actual detection rate is 0.75, which is lower than the expected 0.82, and accordingly the validation action effect distribution for the corresponding anomaly type is updated, with the mean adjusted from 0.85 to 0.82 and the variance increased from 0.03 to 0.05, reflecting the increased uncertainty.

[0077] The sampling temperature parameter is adaptively adjusted to balance exploration and exploitation. The sampling temperature parameter controls the degree of randomness in sampling from the posterior distribution. Higher temperatures result in more random sampling and greater exploratory potential; lower temperatures result in closer sampling results to the distribution mean and greater exploitation. A mechanism is designed to decrease the sampling temperature over each validation round. The initial temperature is set at 1.0 and gradually decreases with each validation round. The temperature decay formula is: the current temperature equals the initial temperature multiplied by the decay factor raised to the power of the validation round. The decay factor is typically set to 0.95. In the early stages of validation, a high temperature encourages the experimentation of various validation strategies. As validation progresses and experience accumulates, the temperature decreases, favoring strategies known to be effective. During the continuous validation process for a bidding and tendering supervision project, various validation combinations were tried in the initial rounds, such as "document analysis first," "enterprise association first," and "historical pattern first." By the 20th validation round, the temperature had dropped to 0.36, and the validation strategy stabilized, primarily starting with "enterprise association analysis." Subsequent validation strategies were then determined based on the discovered clues.

[0078] To adapt to different verification scenarios, adaptive adjustment of the initial temperature based on the characteristics of the verification task is also implemented. For high-risk and high-urgency verification tasks, the initial temperature is set lower, such as 0.7, to encourage greater use of known effective strategies; for new anomalies or exploratory verification, the initial temperature is set higher, such as 1.2, to encourage the trial of new verification paths. After a complete verification cycle, the entire verification process will be reviewed and analyzed based on the final verification results to identify efficient verification modes and inefficient verification links, and continuously optimize the posterior distribution and sampling strategy of the verification evaluation function. In the verification of a complex bidding anomaly, after 30 rounds of verification accumulation, an efficient verification path template for the "multi-level affiliated enterprises bidding collusion" type of anomaly was formed. On average, only 4 verification steps are required to complete the verification process that originally required 7 steps, greatly improving verification efficiency.

[0079] The verification evaluation function constructed in this paper comprehensively considers verification accuracy, verification cost, and decision consistency. The Thompson sampling method, implemented through a Bayesian neural network, automatically balances exploration and exploitation during the verification process, dynamically optimizing the verification path. This method not only improves the accuracy and efficiency of exception verification but also continuously learns and improves based on verification feedback. It possesses strong adaptability and scalability, providing intelligent verification decision support for bidding and tendering supervision, and effectively reducing the subjectivity and resource consumption of manual verification.

[0080] According to a second aspect of an embodiment of the present invention, an electronic device is provided, including: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0081] In a third aspect, the present application provides a computer readable storage medium, having stored thereon computer program instructions, which when executed by a processor, implement the method described above.

[0082] The present application can be a method, an apparatus, a system, and / or a computer program product. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for performing various aspects of the present application.

[0083] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. The bidding and tendering full-factor compliance review method based on multimodal model spatiotemporal sequence analysis is characterized by: include: Obtaining spatiotemporal sequence data and auxiliary data of the bidding project to be reviewed, wherein the spatiotemporal sequence data includes bid time data and bid geographic location data, and the auxiliary data includes the bidding text content; Construct a spatiotemporal dual-stream feature network, in which the temporal feature stream uses adaptive time windows to dynamically segment time series and extract temporal feature representations of events, and the spatial feature stream uses density clustering to adaptively partition geographic locations and extract spatial distribution features; Constructing a knowledge graph based on the output of the spatiotemporal dual-stream feature network, the knowledge graph including nodes representing bidding events and geographic locations and edges representing temporal dependencies and spatial constraints, with attributes of the edges supplemented by the auxiliary data; Based on the knowledge graph, a spatiotemporal pattern encoder is used at the local level to calculate the similarity between node features and the normal pattern library to obtain anomaly scores. At the global level, a graph attention network is used to contextually optimize the anomaly scores to obtain anomaly detection results. Based on deep reinforcement learning, anomaly verification is modeled as a sequential decision-making process, and a bidding compliance verification path is dynamically generated based on the anomaly detection results and the auxiliary data. The compliance assessment results are obtained by optimizing the verification path and a review report is generated.

2. The method according to claim 1, characterized in that The steps of constructing a spatiotemporal dual-stream feature network, in which the temporal feature stream uses an adaptive time window to dynamically segment the time series and extract the temporal feature representation of the event, and the spatial feature stream uses density clustering to adaptively partition the geographic location and extract the spatial distribution features, include: Constructing a time series feature stream, wherein the time series feature stream identifies bidding time points by designing a time sensitivity function for the bidding time data, constructing a non-uniform time window sequence based on the time points, wherein the boundaries of the non-uniform time windows are determined by maximizing the difference in bidding behavior patterns between windows, and extracting the temporal dependency features of the bidding events within each time window using an attention-enhanced recurrent network to obtain a hierarchical time series feature representation; Constructing a spatial feature stream, dynamically dividing the bidding area using an adaptive spatial partitioning algorithm on the bidding geographic location data, wherein the adaptive spatial partitioning algorithm is optimized by minimizing the entropy of bidding behavior within the area and maximizing the difference in bidding behavior between areas, constructing a bidding geographic association network based on spatial partitioning, and extracting spatial distribution characteristics; A spatiotemporal collaborative attention mechanism is used to perform feature fusion on the outputs of the temporal feature stream and the spatial feature stream. The spatiotemporal collaborative attention mechanism dynamically calculates the correlation strength between the hierarchical temporal feature representation and the spatial distribution feature, and adaptively adjusts the weights of features of different dimensions to obtain a unified spatiotemporal feature representation.

3. The method according to claim 1, characterized in that A knowledge graph is constructed based on the output of the spatiotemporal dual-stream feature network, wherein the knowledge graph includes nodes representing bidding events and geographic locations and edges representing temporal dependencies and spatial constraints. The steps of supplementing the attributes of the edges with the auxiliary data include: Obtaining fusion features of the spatiotemporal dual-stream feature network, wherein the fusion features include a temporal feature representation of the bidding event and a spatial distribution feature of the geographic location; Constructing a node set of a knowledge graph, the node set including a bidding event node, a geographic location node, and a bidding entity node, the bidding event node including event identifier, occurrence time, and bid amount attributes, the geographic location node including a city area division attribute, and the bidding entity node including entity identifier, registered capital, and establishment time attributes; Mapping the temporal component and spatial component of the fusion feature to the feature space of the corresponding node through a nonlinear mapping function to obtain a node feature vector; Constructing an edge set of the knowledge graph, the edge set includes time-dependent edges, spatially constrained edges, and subject-related edges. The weight of the time-dependent edges is obtained by calculating the inner product of the feature vectors of adjacent nodes, and the weight of the spatially constrained edges is obtained by calculating the distance between geographic location nodes. The bidding text content is converted into a text semantic vector, the edge weight vector is spliced ​​and fused with the text semantic vector to obtain the complete attribute vector of the edge, the edge set is optimized based on the importance score of the complete attribute vector, and the temporal and spatial consistency of the edge attributes is ensured by setting the proportional coefficient of the temporal feature weight and the spatial feature weight.

4. The method according to claim 1, wherein Based on the knowledge graph, a spatiotemporal pattern encoder is used at a local level to calculate the similarity between node features and a normal pattern library to obtain an anomaly score. A graph attention network is used at a global level to perform contextual optimization on the anomaly score. The steps for obtaining an anomaly detection result include: Obtain node features and neighborhood information in the knowledge graph, wherein the node features include temporal features and spatial features, and the neighborhood information includes dependencies between nodes; A spatiotemporal pattern encoder is used to encode the node features and the neighborhood information to obtain a spatiotemporal encoding representation of the node. A normal pattern feature library is constructed based on the spatiotemporal encoding representation, wherein the normal pattern feature library includes multiple normal pattern center representations. The similarity between the spatiotemporal encoding representation of the node and the normal pattern center representation is calculated to obtain a local anomaly score of the node. Construct a multi-head graph attention network, calculate the attention weights between nodes based on the query matrix, key matrix and value matrix, perform weighted aggregation on the node features according to the attention weights, and obtain the contextual representation of the node; Fusing the local anomaly score of the node with the context representation of the node to obtain a global anomaly score of the node; A timing smoothing constraint is constructed based on the timing dependency relationship. The timing smoothing constraint is achieved by minimizing the difference in global anomaly scores between timing-related nodes. The global anomaly score is propagated and optimized according to the strength of the timing dependency to obtain the final anomaly score of the node. The detection threshold is determined based on the final anomaly score and anomaly marking is performed to obtain the anomaly detection result.

5. The method according to claim 4, characterized in that The steps of constructing a timing smoothing constraint based on the timing dependency, which is implemented by minimizing the difference in global anomaly scores between timing-related nodes, and optimizing the propagation of the global anomaly score based on the strength of the timing dependency to obtain the final anomaly score of the node include: Constructing a temporal dependency relationship, and calculating the temporal dependency strength of temporally related node pairs using a dynamic graph neural network. The dynamic graph neural network includes a temporal query matrix and a temporal key matrix. The node feature vectors are transformed using the temporal query matrix and the temporal key matrix, and the temporal dependency strength of the node pairs is calculated in combination with a time interval decay function. Constructing a smoothing constraint objective function based on the temporal dependency strength, the smoothing constraint objective function including a smoothing loss term and a causal loss term, the smoothing loss term calculating the squared difference of the global anomaly scores of the related nodes based on the temporal dependency strength, and the causal loss term calculating the deviation of the global anomaly score of the node from its expected value based on the intervention operation; Design an anomaly score propagation function, iteratively update the global anomaly score of the node according to the temporal dependency strength, and use the smooth constraint objective function as the optimization target to optimize the global anomaly score through the gradient descent method, introducing a penalty term in each iteration to constrain the change range of the global anomaly score; The optimized global anomaly score is fused at multiple scales, and the final anomaly score of the node is calculated based on the global anomaly scores at different scales and the corresponding weights.

6. The method according to claim 1, characterized in that Based on deep reinforcement learning, anomaly verification is modeled as a sequential decision-making process. A bidding compliance verification path is dynamically generated based on the anomaly detection results and the auxiliary data. The compliance assessment results are obtained by optimizing the verification path. The steps of generating the review report include: Constructing a verification feature vector based on the anomaly detection result, constructing a verification state space and a verification action space based on the verification feature vector and the auxiliary data, wherein the verification state space includes the verification feature vector and a historical verification decision sequence; constructing a verification strategy generation network based on the verification state space and the verification action space, wherein the verification strategy generation network uses a variational autoencoder to encode the verification feature vector to obtain a potential representation of the abnormal pattern, constructing a verification strategy distribution based on the potential representation, and generating a verification action sequence according to the verification strategy distribution; Constructing a verification evaluation function including a verification accuracy term, a verification cost term, and a decision consistency term, performing sampling optimization on the verification evaluation function using the Thompson sampling method to obtain an optimal verification action sequence, and determining a verification path based on the optimal verification action sequence; A verification process is performed along the verification path, and the temporal compliance index, spatial consistency index and text semantic similarity index are calculated for each verification node. A multidimensional risk dependency matrix is ​​constructed based on the temporal compliance index, spatial consistency index and text semantic similarity index. The Copula function is used to calculate the multidimensional risk joint distribution, and the compliance assessment result is output. Based on the compliance assessment result, abnormal spatiotemporal pattern information and its propagation path are extracted to generate an audit report.

7. The method according to claim 6, characterized in that Constructing a verification evaluation function including a verification accuracy term, a verification cost term, and a decision consistency term, and optimizing the verification evaluation function using the Thompson sampling method to obtain an optimal verification action sequence includes the following steps: The verification accuracy item is calculated based on the anomaly detection rate corresponding to the verification action sequence, the verification cost item is calculated based on the resource consumption of the verification depth and verification scope, and the decision consistency item is calculated based on the policy difference of adjacent verification actions; A Bayesian neural network is used to implement the Thompson sampling method to establish the posterior distribution of the verification evaluation function, and the weight distribution of the Bayesian neural network is trained based on the historical verification decision sequence to obtain a probabilistic representation of the verification evaluation function; In each round of verification decision, an instance of the current verification evaluation function is sampled from the posterior distribution based on the Thompson sampling principle, candidate verification action sequences are scored based on the instance of the current verification evaluation function, and the verification action sequence with the highest score is selected as the optimal verification action sequence; The posterior distribution is updated according to the verification feedback information, and the exploration and utilization are balanced by adaptively adjusting the sampling temperature parameter, which decreases as the number of verification rounds increases.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images