Collection bid evaluation risk prevention and control system and method integrated with multi-dimensional intelligent analysis

Abstract

The invention relates to the technical field of bid evaluation risk prevention and control, in particular to a bid evaluation risk prevention and control system and method integrated with multi-dimensional intelligent analysis, and the method comprises the steps: carrying out the semantic vector mapping and fine-grained similarity calculation of a bid text through a pre-training language model, recognizing hidden similar contents, and carrying out the recognition of the hidden similar contents; generating a semantic-structure two-dimensional similar segment set; on the basis, calculating a text similarity matrix between bidding subjects and a key term matching offset, and screening an abnormal associated subject cluster; further extracting industry and commerce information, a behavior time sequence and network topology data, and constructing a subject association evolution trajectory chain; and finally, in combination with a risk threshold reference, measuring a semantic space distance and a behavior correlation critical distance, and screening a high-risk bidding path. The full-chain prevention and control from text detection to risk deduction is realized, and the recognition accuracy and early warning capability of hidden bid surrounding and stringing behaviors are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of bid evaluation risk prevention and control technology, and in particular to a bidding and procurement bid evaluation risk prevention and control system and method that integrates multi-dimensional intelligent analysis. Background Technology

[0002] In today's large-scale, high-frequency bidding and procurement activities, unfair competition behaviors such as bid rigging and collusion are becoming increasingly covert, organized, and sophisticated, seriously disrupting the fair competition market order and affecting the rational allocation of public resources. Traditional risk control methods in bidding and procurement evaluation mainly rely on manual review and simple keyword comparison techniques, which are insufficient to effectively identify implicit similarities achieved through synonym substitution, sentence restructuring, and paragraph reordering, let alone detect collaborative behavior patterns between multiple entities. With the continuous increase in the number and complexity of bid documents, rule-based or shallow text matching methods are no longer sufficient to meet the needs for accurate and efficient risk identification.

[0003] In recent years, Natural Language Processing (NLP) technology has made significant progress in semantic understanding and text similarity calculation. Pre-trained language models (such as BERT and RoBERTa) can capture deep semantic information in text, providing a technological possibility for identifying plagiarism behavior that circumvents keyword substitution. Meanwhile, Graph Neural Networks (GNNs) and complex network analysis methods have demonstrated powerful capabilities in mining potential relationships between entities. They can be used to construct relationship graphs between bidding entities, revealing abnormal collaborative patterns in areas such as business registration, bidding time series, and geographical location. However, existing research largely focuses on single-dimensional risk detection, lacking a systematic method that organically combines text semantic analysis with the mining of relationships between entities' behaviors. There are still significant shortcomings, particularly in constructing chains from local text similarity to global bid-rigging path deduction.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a bidding and tendering risk prevention and control system and method that integrates multi-dimensional intelligent analysis. It aims to solve the technical problems that existing bidding and tendering risk prevention and control technologies are unable to effectively identify hidden text similarities and complex subject collaboration behaviors, and lack a systematic method that integrates deep semantic analysis and correlation network mining to achieve full-chain intelligent prevention and control of bid rigging and collusion.

[0006] To achieve the above objectives, this invention provides a method for risk prevention and control in procurement and evaluation that integrates multi-dimensional intelligent analysis, the method comprising:

[0007] Based on multi-source heterogeneous text data in the procurement document library and the tender document library, semantic vector mapping is performed through a pre-trained language model, fine-grained text similarity calculation is performed, suspected plagiarized paragraphs and implicit similar content that avoids keyword replacement are identified, and a set of semantic-structural dual-dimensional similar paragraphs is generated.

[0008] Based on the information of each segment in the semantic-structural dual-dimensional similar segment set, calculate the text similarity matrix between bidding entities and the matching offset of key clauses, filter bidding entity clusters that meet the multi-dimensional abnormal association conditions, and generate multi-entity collaborative bid-rigging node clusters.

[0009] Based on each node cluster in the multi-entity collaborative bid-rigging node cluster, extract the business registration information of the associated entities, the time series of bidding behavior and the associated network topology, construct the entity association evolution trajectory chain, and generate a set of information on the advancement path of bid-rigging behavior.

[0010] Based on the path end node information in the information set of the bid-rigging and collusion behavior promotion path, the corresponding bidding and procurement risk threshold benchmark set is extracted, the semantic space distance from the path end node to the critical point of bid-rigging behavior is measured and compared with the behavior association critical distance, the path set that meets the bid-rigging penetration condition is screened, and a list of high-risk bid-rigging and collusion paths is generated.

[0011] Optionally, the semantic-structural dual-dimensional similarity segment set includes semantic similarity values, structural similarity indicators, and the duration of implicit similarity segments within the similarity segments; the multi-entity collaborative bid-rigging node cluster includes the bid entity number, inter-entity association strength index, and bid space continuity identifier that satisfy the multi-dimensional abnormal association conditions; the bid-rigging and collusion behavior advancement path information set includes the business information topological distance value, behavior collaboration time delay value, and association propagation direction sequence of each node in the trajectory chain; the high-risk bid-rigging and collusion path list includes the entity business registration coordinate information, risk threshold distance measurement value, and corresponding procurement project level code of the path terminal node.

[0012] Optionally, the steps for obtaining the semantic-structural dual-dimensional similar segment set are as follows:

[0013] Based on the text content of technical specifications and business terms in the procurement document library, hierarchical semantic encoding is performed through the BERT pre-trained model. Dynamic sliding window semantic similarity calculation is performed on each text segment to judge the similarity change fluctuation characteristics, identify abnormal similar segments whose fluctuation amplitude exceeds the set threshold, and generate a set of semantically similar text segments.

[0014] Based on each segment in the set of semantically similar segments, extract the keyword replacement pattern data in the corresponding bidding documents, calculate the keyword distribution entropy value for each segment of text, evaluate the degree of implicit similarity in avoiding detection, filter key abnormal segments by setting an implicit similarity threshold, and generate a structural avoidance feature set.

[0015] Based on the set of semantically similar segments and the set of structural avoidance features, spatiotemporal correlation analysis is performed on the similar segments. By judging the coupling relationship between semantic similarity and structural avoidance, segments that meet the characteristics of bid-rigging behavior are selected, and a set of semantically-structurally similar segments is generated.

[0016] Optionally, the steps for obtaining the multi-entity collaborative bid-rigging node cluster are as follows:

[0017] Based on the information of each segment in the semantic-structural dual-dimensional identical segment set, the corresponding bidder identifier, bid submission time and start and end positions of the identical segment are extracted. The distribution density of the identical segment in the bid document is calculated. Based on the spatial topological relationship of the business registration information between bidders, a topological combination list between any related entities is established. The identical distribution density values ​​of the entities in the combination are called in turn, the density difference is calculated, and a sequence of identical density difference values ​​of related entities is generated.

[0018] Based on the starting position of the similar segments of the associated subjects, obtain the text matching starting position offset between each pair of subjects, construct the position offset sequence, call the similarity density difference sequence of the associated subjects and the text matching starting position offset sequence, and compare them with the set similarity density tolerance threshold and position synchronization tolerance threshold respectively, filter the subject combinations that are simultaneously less than the two tolerance thresholds, and obtain the multi-dimensional abnormal association satisfying combination index set.

[0019] The multi-dimensional abnormal association satisfies the number identifier of each subject combination in the combined index set, extracts the corresponding business information code of the subject in the original similar segment information, constructs the subject topology cluster that meets the multi-dimensional abnormal association conditions, calculates and obtains the association cluster strength value, and filters according to whether the cluster strength value falls within the procurement risk intensity range, obtains the subject combination that passes the filter, and establishes a multi-subject collaborative bid-rigging node cluster.

[0020] Optionally, the steps for obtaining the information set of the path to advance the bid-rigging behavior are as follows:

[0021] Based on each node cluster in the multi-subject collaborative bid-rigging node cluster, the orientation vector, spatial coordinates of associated information and behavior start time of the subject in the business registration network are extracted. Pairwise combinations between subjects are constructed and the order of the associated network is determined. The spatial Euclidean distance and behavior collaboration start time difference of adjacent subjects in the combination are calculated to generate a set of associated behavior extension distance-time difference values.

[0022] The main body combination in the aforementioned associated behavior extension distance-time difference set is called, and it is determined whether the angle between the associated network direction difference vector and the Euclidean connection direction is less than the associated path consistency angle threshold. At the same time, it is determined whether the behavior coordination time strictly meets the network incremental trend. The main body sequence combination that meets the dual constraints is selected to obtain the network incremental extension combination index set.

[0023] Based on the main combination index in the network incremental expansion combination index set, a list of network-connected behavioral trajectory segments is established, a path chain for promoting bid-rigging and collusion behavior is constructed, and the cumulative association propagation distance and behavioral time-domain window width of the trajectory segments are recorded. The maximum association propagation length and the minimum behavioral time-domain window value are extracted to obtain the information set of the path for promoting bid-rigging and collusion behavior.

[0024] Optionally, the steps for obtaining the list of high-risk bid-rigging and collusion paths are as follows:

[0025] Based on the trajectory end node information of the information set of the bid-rigging and collusion behavior, the corresponding main business information code and the spatial coordinates of the end node are extracted. According to the information code, the risk threshold benchmark set of the relevant bidding and procurement project is retrieved. The coordinate values ​​are called to calculate the spatial Euclidean distance between the end node and all critical points in the risk threshold benchmark set, and the distance information from the end of the path to the risk threshold point is generated.

[0026] Based on the distance information from the end of the path to the risk threshold point, the minimum distance value corresponding to each end node of the path is extracted, the risk distance sequence of the end node is constructed, the behavior-related critical distance benchmark value is called to compare each value in the distance sequence, the path combination index that is less than or equal to the benchmark value is filtered, and the risk path index set of bid rigging and penetration is obtained.

[0027] Based on the path number identified in the risk path index set for bid rigging, the corresponding trajectory segment information is extracted from the information set on the path advancement of bid rigging and collusion behavior. The risk path rigging risk is reconstructed and the shortest penetration distance between the end of the path and the risk threshold point is marked to obtain the list of high-risk bid rigging and collusion paths.

[0028] Optionally, after obtaining the list of high-risk bid-rigging and collusion paths, the method further includes:

[0029] For each path in the list of high-risk bid rigging and collusion paths, the maximum semantic similarity change amplitude within the identical segment at the starting point of the path, the total time span of the path evolution, and the number of abnormally associated entities in the path are extracted. The corresponding procurement risk index is calculated, and the handling priority is marked according to the level of the procurement project to which it belongs, generating multi-dimensional procurement risk labeling results.

[0030] The multi-dimensional procurement risk assessment results include risk level labels, procurement project identifiers corresponding to the levels, and procurement risk index codes.

[0031] Optionally, the steps for obtaining the multi-dimensional procurement risk assessment results are as follows:

[0032] Based on each path in the list of high-risk bid rigging and collusion paths, the semantic similarity sequence of the similar segments of the subjects corresponding to the starting point of the path is extracted, the difference between the maximum value in the sequence and the minimum value of the historical baseline is identified, and normalization is performed based on the complexity of the bidding and procurement project to obtain the normalized risk amplitude of each path and generate a path risk amplitude sequence.

[0033] Call the path information corresponding to the path risk magnitude sequence, extract the start and end times of each path and calculate the time domain span, count the number of bidding entities marked as abnormal association status within the path, combine the three indicators to construct a multi-parameter risk assessment set, calculate and obtain the procurement risk index value, and generate a procurement risk index value set.

[0034] Based on the index corresponding to each path in the procurement risk index value set, the procurement project level identifier to which the path belongs is retrieved. According to the preset risk classification benchmark value range within the procurement project level, the procurement risk index is classified and judged, and the corresponding disposal priority level is marked to obtain multi-dimensional procurement risk labeling results.

[0035] Furthermore, to achieve the above objectives, the present invention also provides a procurement and evaluation risk prevention and control system integrating multi-dimensional intelligent analysis. The system includes: a memory, a processor, and a procurement and evaluation risk prevention and control program integrating multi-dimensional intelligent analysis stored in the memory and executable on the processor. The procurement and evaluation risk prevention and control program integrating multi-dimensional intelligent analysis is configured to implement the steps of the procurement and evaluation risk prevention and control method integrating multi-dimensional intelligent analysis as described above.

[0036] In addition, to achieve the above objectives, the present invention also provides a medium storing a procurement and evaluation risk control program integrating multi-dimensional intelligent analysis, wherein when the procurement and evaluation risk control program integrating multi-dimensional intelligent analysis is executed by a processor, it implements the steps of the procurement and evaluation risk control method integrating multi-dimensional intelligent analysis as described above.

[0037] This invention provides a method for risk prevention and control in bidding and evaluation that integrates multi-dimensional intelligent analysis. The method overcomes the limitations of traditional keyword comparison by using semantic vector mapping and fine-grained similarity calculation through a pre-trained language model. It can effectively identify implicit similarities that are evaded detection through keyword replacement, sentence reconstruction, and other means, achieving semantic-structural dual-dimensional text similarity analysis and significantly improving the ability to mine textual evidence of bid rigging and collusion. Based on multi-dimensional data (text similarity, key clause matching offset, business registration information, bidding behavior time series, etc.), a network of associations among bidding entities is constructed. Abnormal clusters of related nodes are filtered through graph neural networks and complex network analysis, enabling precise location of nodes with cooperative relationships. By combining entities with similar characteristics of bid rigging, this approach addresses the difficulty of traditional methods in detecting hidden cross-entity connections. It constructs a layered analysis framework, from textual similarity segments to entity association clusters, and then to the behavioral progression path, forming a full-chain analysis framework encompassing micro-level textual anomalies, meso-level entity collaboration, and macro-level risk paths. This enables dynamic tracking of bid rigging behavior from its inception to its evolution, providing a continuous spatiotemporal evidence chain for risk warning. Through quantitative comparison of semantic spatial distance and behavioral association thresholds, combined with the procurement project level and risk threshold benchmark set, it can objectively screen high-risk penetration paths and label risk levels, avoiding subjective judgment bias, improving the targeting and efficiency of risk prevention and control, and providing intelligent decision-making basis for procurement supervision. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating an embodiment of the procurement and evaluation risk control method integrating multi-dimensional intelligent analysis according to the present invention.

[0039] Figure 2 This is a schematic diagram of the steps for obtaining the semantic-structural dual-dimensional identical segment set in one embodiment of the bidding and evaluation risk prevention and control method integrating multi-dimensional intelligent analysis of the present invention;

[0040] Figure 3 This is a schematic diagram of the steps for obtaining a cluster of collaborative bid-rigging nodes by multiple entities in one embodiment of the bidding and evaluation risk prevention and control method integrating multi-dimensional intelligent analysis of the present invention.

[0041] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0042] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0043] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the procurement and evaluation risk control method integrating multi-dimensional intelligent analysis according to the present invention. An embodiment of the procurement and evaluation risk control method integrating multi-dimensional intelligent analysis according to the present invention is presented.

[0044] In one embodiment, the procurement and evaluation risk prevention and control method integrating multi-dimensional intelligent analysis includes:

[0045] Step S100: Based on the multi-source heterogeneous text data in the procurement document library and the tender document library, semantic vector mapping is performed through a pre-trained language model, fine-grained text similarity calculation is performed, suspected plagiarized paragraphs and implicit similar content that avoids keyword replacement are identified, and a set of semantic-structural dual-dimensional similar paragraphs is generated.

[0046] The semantic-structural dual-dimensional similarity segment set can be a collection of segments in the bidding documents that are highly similar at both the semantic and textual structure levels, identified through semantic vector mapping and structural feature comparison. This set can be used to reveal implicit similarities that circumvent keyword detection through synonym substitution, sentence reconstruction, or other means. In this embodiment, the semantic-structural dual-dimensional similarity segment set can be based on a pre-trained language model to map text to a semantic vector space, combined with paragraph structural features (such as sentence order and paragraph length distribution) for fine-grained similarity calculation. For example, the semantic-structural dual-dimensional similarity segment set may include, but is not limited to, segments that are semantically similar but structurally different, segments that are structurally consistent but semantically offset, and segments with high similarity in both semantics and structure.

[0047] To perform fine-grained text similarity calculations and identify suspected plagiarized paragraphs and implicit similarities that circumvent keyword substitution, the bidding text can be input into a pre-trained language model to obtain semantic vectors. This can then be combined with paragraph structural features to calculate the comprehensive similarity between paragraphs across documents. Furthermore, this operation can be achieved by using BERT embedding + cosine similarity + paragraph edit distance weighted fusion, or by using RoBERTa to generate context-aware vectors and combining them with syntactic tree structure similarity for joint scoring. This overcomes the limitations of traditional keyword matching and effectively identifies implicit similarities that are semantically equivalent but differ in surface form.

[0048] Step S200: Based on the information of each segment in the semantic-structural dual-dimensional similar segment set, calculate the text similarity matrix between bidding entities and the offset of key clause matching, filter bidding entity clusters that meet the multi-dimensional abnormal association conditions, and generate multi-entity collaborative bid-rigging node clusters.

[0049] The multi-entity collaborative bid-rigging node cluster can be a subset of bidding entities clustered within a bidding entity association network that satisfies multi-dimensional abnormal association conditions such as text similarity and key clause matching offsets. This cluster can be used to accurately locate combinations of multiple bidding entities exhibiting characteristics of collaborative bid-rigging behavior. In an exemplary embodiment, the multi-entity collaborative bid-rigging node cluster can construct an inter-entity similarity matrix based on a set of similar segments and integrate key clause matching offsets as edge weights, using a graph clustering algorithm to identify abnormally dense subgraphs. Furthermore, the multi-entity collaborative bid-rigging node cluster can rely on a semantic-structural dual-dimensional set of similar segments to provide textual association evidence and an initial node set for the information set of the bid-rigging and collusion behavior advancement path.

[0050] Selecting bidding entity clusters that meet multi-dimensional abnormal association conditions can be achieved by constructing a weighted graph based on the text similarity matrix and the offset matching of key clauses, and then applying a community detection algorithm to identify abnormally dense subgraphs. In a specific embodiment, this operation can be performed by using the Louvain algorithm to optimize clustering on the multi-dimensional weighted graph, or by using GNN to encode node features and then performing K-means clustering in the embedding space, thereby achieving accurate clustering and identification of cross-entity covert collaborative behavior.

[0051] Step S300: Based on each node cluster in the multi-entity collaborative bid-rigging node cluster, extract the business registration information of the associated entities, the time series of bidding behavior, and the associated network topology, construct the entity association evolution trajectory chain, and generate a set of information on the advancement path of bid-rigging behavior.

[0052] The information set for the advancement path of bid-rigging behavior can be a dynamic set of behavioral trajectories constructed from a cluster of multi-entity collaborative bid-rigging nodes, integrating business registration information, bidding time series, and network topology evolution. This can be used to achieve full-chain tracking from static text similarity to dynamic bid-rigging behavior evolution. In this embodiment, the information set for the advancement path of bid-rigging behavior can perform time-series alignment and path splicing on multi-source data such as historical bidding records, equity structures, and registered addresses of entities within the node cluster. For example, the information set for the advancement path of bid-rigging behavior may include, but is not limited to, business association-driven paths, time-coordination-intensive paths, and network topology-transmission paths.

[0053] Constructing an evolutionary trajectory chain of subject associations can involve aligning and connecting time-series data such as business registration change records, bidding timestamps, and IP addresses of subjects within a node cluster. Furthermore, this operation can be achieved by constructing a sequence of subject interaction events based on a sliding time window, linking them chronologically to form a trajectory, or by utilizing knowledge graph temporal reasoning methods to fuse multi-source heterogeneous events into a unified evolutionary path, thereby forming an evidence chain from static associations to dynamic behavioral evolution.

[0054] Step S400: Based on the path end node information in the information set of the path promotion path of bid rigging and collusion behavior, extract the corresponding bidding and procurement risk threshold benchmark set, measure the semantic space distance from the path end node to the critical point of bid rigging behavior, compare it with the behavior association critical distance, filter the path set that meets the bid rigging penetration condition, and generate a list of high-risk bid rigging and collusion paths.

[0055] The procurement risk threshold benchmark set can be a pre-defined set of risk judgment thresholds based on the procurement project level, historical bid-rigging case statistics, and regulatory rules. This set can provide objective quantitative evidence for screening high-risk paths, avoiding subjective judgment bias. In one specific embodiment, the procurement risk threshold benchmark set can establish a hierarchical threshold system through backtracking analysis of historical high-risk bid-rigging events, combined with dimensions such as project type, amount scale, and industry characteristics. Furthermore, the procurement risk threshold benchmark set can be compared with the end nodes in the bid-rigging and collusion behavior advancement path information set to support the determination of high-risk paths.

[0056] Measuring the semantic space distance from the end node of a path to the critical point of bid-rigging behavior and comparing it with the behavior-related critical distance can be achieved by calculating the distance between the end node and the center of a known bid-rigging pattern in the semantic vector space and comparing it with a preset critical distance threshold. For example, this operation can be implemented by using Euclidean distance to measure the deviation of the semantic vector from the center of the bid-rigging prototype, or by using Mahalanobis distance combined with the covariance matrix to correct the scale differences of multidimensional semantic features, thereby objectively screening bid-rigging paths with penetration risk.

[0057] Taking the bidding evaluation of large-scale infrastructure construction projects as an example, the risk prevention and control method for bidding evaluation integrating multi-dimensional intelligent analysis in this embodiment can be as follows: In a highway construction bidding process, the system uses the BERT model to find that the technical solution paragraphs of three bidding units are highly similar in both semantic and structural dimensions; further, it constructs a subject association network, and combines information such as the time interval between their bids being less than 2 hours and their registered addresses being located in the same office building, and clusters them into a cluster of multi-subject collaborative bid-rigging nodes; tracing their bidding records over the past year, it is found that the subjects in this cluster have alternately appeared as co-bidders in 5 different projects, forming a set of information on the path of bid-rigging and collusion; finally, the semantic features of the terminal node of this path are less than the critical value of the center of historical bid-rigging cases, and the project is of high risk level, so the system automatically marks it as a high-risk bid-rigging and collusion path and issues a warning.

[0058] In one embodiment, the semantic-structural dual-dimensional set of similar segments includes semantic similarity values, structural similarity indices, and the duration of implicit similarity segments within the similar segments.

[0059] The semantic-structural dual-dimensional set of similar segments can be a structured dataset used to identify highly similar content in tender documents. In this embodiment, the set is constructed by simultaneously calculating the semantic vector similarity, paragraph structural feature similarity, and the length of consecutive similar text segments. For example, structural similarity can be calculated using paragraph syntactic tree edit distance or sentence order consistency; the duration of implicit similarity segments can be detected by using a sliding window to measure the maximum span of consecutive highly similar segments, thereby enhancing the ability to identify structural plagiarism behaviors such as template reuse and paragraph reordering, and improving the fine-grainedness and robustness of similarity determination.

[0060] The multi-entity collaborative bid-rigging node cluster includes the bid entity number that meets the multi-dimensional abnormal association conditions, the association strength index between entities, and the continuousness identifier of the bidding space.

[0061] The multi-entity collaborative bid-rigging node cluster can be a set of nodes formed based on clustering algorithms, representing abnormal collaborative relationships among multiple bidding entities. In an exemplary embodiment, this node cluster assigns a unique number to each entity, calculates its comprehensive association strength with other entities, and marks its continuity characteristics in the bidding geographical or temporal space. Furthermore, the association strength index can be calculated by weighting text similarity, business equity penetration depth, and IP address overlap; the bidding spatial continuity identifier can be determined by a bidding time interval threshold or geographical proximity of registered addresses, enabling collaborative bid-rigging combinations to have traceable and verifiable quantitative evidence, supporting cross-project behavior pattern analysis.

[0062] The information set for the path of bid rigging and collusion includes the business information topological distance value, behavior coordination time delay value, and associated propagation direction sequence of each node in the trajectory chain.

[0063] The information set on the path of bid-rigging and collusion can be a time-series structured dataset used to characterize the evolution of bid-rigging and collusion. In a specific embodiment, this information set is constructed by calculating the shortest path distance in the business relationship graph of each pair of adjacent nodes in the trajectory chain, the time difference of the bidding behavior, and inferring the direction of association transmission. For example, the topological distance value of business information can be calculated by constructing a knowledge graph based on corporate equity, cross-appointment of senior executives, etc.; the sequence of association propagation direction can be inferred by jointly inferring the temporal order and control relationship (such as parent company → subsidiary), giving the behavior path a spatiotemporal causal logic, and realizing the evidence upgrade from static co-occurrence to dynamic evolution.

[0064] The list of high-risk bid-rigging and collusion paths includes the main business registration coordinates of the path's terminal nodes, the risk threshold distance measurement value, and the corresponding procurement project level code.

[0065] The high-risk bid-rigging and collusion path list can be an output set containing structured attribute information of bid-rigging and collusion behaviors identified as high-risk after quantitative comparison. This list can provide tiered and manageable decision-making input for procurement supervision, supporting risk warning and subsequent verification. In this embodiment, the list is generated based on the terminal nodes of the bid-rigging and collusion behavior advancement path information set, combined with a procurement risk threshold benchmark set for distance measurement and project level matching. Furthermore, business registration coordinate information can be parsed from structured addresses and converted into latitude and longitude using the enterprise credit information system; project level codes can be pre-classified according to national or industry standards based on project type, budget scale, and strategic importance, transforming abstract risks into concrete, tiered, and locatable regulatory decision-making elements. For example, the high-risk bid-rigging and collusion path list may include, but is not limited to, one or more of the following: high-value project high-risk paths, cross-regional registered entity high-risk paths, and high-frequency collaborative bidding high-risk paths.

[0066] For example, in the scenario of centralized procurement of medical equipment at the provincial level, the risk prevention and control method for bidding and evaluation integrating multi-dimensional intelligent analysis in this embodiment can be as follows: When the system analyzes the bidding documents of a certain CT equipment, it finds that the semantic similarity of the paragraphs describing the technical parameters of three suppliers reaches 0.92, the structural similarity index shows that the sentence order is highly consistent, and the implicit similarity lasts for a period of 800 words; further clustering forms a cluster of multi-entity collaborative bid-rigging nodes, among which the association strength index of entity numbers A, B, and C exceeds the threshold, and the bidding IPs all come from the same commercial building in the same city (the bidding spatial continuity identifier is true); tracing their behavior trajectory, the topological distance value of the business information is 2 (there is a common legal person), the time delay of the last two bids is only 3 days, and the association propagation direction sequence is A→B→C; finally, this path is included in the list of high-risk bid-rigging paths, the terminal node registration coordinates are located in a certain free trade zone, the risk threshold distance measurement value is lower than the critical value of 0.15, the corresponding project level code is "Class I high-value medical consumables", triggering an automatic warning.

[0067] In one embodiment, reference Figure 2 The specific steps for obtaining the set of semantically and structurally identical segments are as follows:

[0068] Step S201: Based on the text content of the technical specifications and business terms in the procurement document library, hierarchical semantic encoding is performed through the BERT pre-trained model. Dynamic sliding window semantic similarity calculation is performed on each text segment to determine the similarity change fluctuation characteristics, identify abnormal similar segments whose fluctuation amplitude exceeds the set threshold, and generate a set of semantically similar text segments.

[0069] The set of semantically similar segments can be a collection of bidding text segments that exhibit abnormally high similarity at the local semantic level, identified through dynamic sliding window semantic similarity calculation. This set can serve as preliminary semantic anomaly detection results to locate potentially similar content regions. In this embodiment, the set of semantically similar segments can be obtained by using BERT to perform hierarchical encoding on the bidding and tendering texts, calculating cross-document segment similarity within a sliding window, and identifying abnormally similar segments with fluctuations exceeding a set threshold. Furthermore, the set of semantically similar segments can include, but is not limited to, one or more of the following: semantically similar segments in technical specifications, semantically similar segments in business terms, and semantically similar segments in mixed content.

[0070] For each text segment, a dynamic sliding window semantic similarity calculation is performed to determine the fluctuation characteristics of similarity changes and identify abnormally similar segments whose fluctuation amplitude exceeds a set threshold. This can be achieved by setting a variable-length sliding window on the semantic vector sequence after BERT encoding, calculating the similarity between the bidding text and the procurement template segment by segment, and analyzing its local fluctuation patterns. In an exemplary embodiment, this operation can be implemented by using an adaptive window size that is dynamically adjusted according to sentence boundaries or paragraph semantic coherence, or by using the standard deviation of similarity or kurtosis within the sliding window as a fluctuation feature indicator. This can effectively capture hidden similarities where the local semantics are highly consistent but the overall structure is scattered, avoiding omissions in fixed-granularity comparisons.

[0071] Step S202: Based on each segment in the set of semantically similar segments, extract the keyword replacement pattern data in the corresponding bidding documents, calculate the keyword distribution entropy value for each segment of text, evaluate the degree of implicit similarity in avoiding detection, filter key abnormal segments by setting an implicit similarity threshold, and generate a structural avoidance feature set.

[0072] The keyword replacement pattern data can be a set of word mapping relationships in the tender document and the corresponding position in the procurement document that have lexical differences but similar semantics. It can be used to support the calculation of keyword distribution entropy value and quantify the strength of text avoidance detection strategies. In a specific embodiment, the keyword replacement pattern data can be obtained by comparing the lexical differences in semantically similar segments through word alignment or synonym mapping tools, and extracting categorizable replacement rules or instances.

[0073] The structural evasion feature set can be a collection of abnormal text segments with obvious keyword substitution or distribution perturbation characteristics, selected based on keyword distribution entropy value evaluation. It can be used to reveal implicit text camouflage behaviors implemented by bidders to evade keyword detection. For example, the structural evasion feature set can be formed by extracting keyword substitution pattern data from semantically similar text segments, calculating the entropy value of each keyword distribution segment, and including segments with entropy values ​​exceeding the implicit similarity threshold. Further, the structural evasion feature set can include, but is not limited to, high-entropy keyword substitution segments, low-frequency word cluster substitution segments, and terminology system shift segments. The implicit similarity threshold can be a preset critical value used to determine whether the keyword distribution entropy value reaches a significant level of evasion detection. It can be used as a decision boundary for selecting structural evasion feature segments, controlling the balance between false positives and false negatives. In this embodiment, the implicit similarity threshold can be determined empirically or optimized through machine learning based on the statistical distribution of keyword substitution behavior in historical bid-rigging cases.

[0074] For each text segment, keyword distribution entropy is calculated to assess the degree of implicit similarity in evasion detection. This can be achieved by statistically analyzing the frequency distribution of keywords (such as technical parameters and qualification requirements) in semantically similar segments and calculating their information entropy to measure the degree of confusion or deliberate substitution in word usage. Furthermore, this operation can be implemented by calculating conditional entropy only for the key fields marked in the procurement documents, or by combining TF-IDF weighted word frequency distribution to calculate weighted entropy values. This allows for the quantification of the strength of text evasion behavior and the identification of implicit collusion that appears identical on the surface but is actually highly similar.

[0075] Step S203: Based on the set of semantically similar segments and the set of structural avoidance features, perform spatiotemporal correlation analysis on the similar segments. By judging the coupling relationship between semantic similarity and structural avoidance, select segments that meet the characteristics of bid-rigging behavior and generate a set of semantic-structural dual-dimensional similar segments.

[0076] By determining the coupling relationship between semantic similarity and structural avoidance, segments that meet the characteristics of bid-rigging behavior can be screened. This can be achieved by jointly analyzing the semantic similarity score in the set of semantically similar segments and the entropy value in the set of structural avoidance features, and then selecting segments where both are in the high-risk range. In an exemplary embodiment, this operation can be implemented by constructing a two-dimensional risk plane and setting a semantic-structural joint threshold region for screening, or by training a lightweight classifier to predict whether a segment belongs to bid-rigging similarity based on semantic similarity and entropy values. This allows for dual verification of semantic consistency and structural camouflage, improving the accuracy and adversarial nature of similar segment identification.

[0077] For example, in the scenario of bidding for a smart city software platform, the risk prevention and control method for bidding and evaluation integrating multi-dimensional intelligent analysis in this embodiment can be as follows: The system analyzes the bidding documents of a municipal government cloud platform. First, through the BERT dynamic sliding window, it discovers that multiple units have a sudden increase in local semantic similarity (fluctuation exceeding the threshold) in the "Data Security Architecture" section, generating a set of semantically similar text segments. Further, it extracts keywords (such as "Level 3 Network Security Protection" and "Two-Factor Authentication") from these segments, calculates their distribution entropy value, and finds that some bidders use terms such as "commercial cryptographic algorithms approved by the State Cryptography Administration" and "multi-factor authentication" to replace the original words, with entropy values ​​significantly higher than normal levels, forming a structural avoidance feature set. Finally, the system selects segments with semantic similarity > 0.85 and entropy value > 1.2, confirming that they simultaneously possess high semantic consistency and strong avoidance features, and includes them in the semantic-structural dual-dimensional similar segment set as the starting point of the evidence chain for bid rigging.

[0078] In one embodiment, reference Figure 3 The specific steps for obtaining a cluster of nodes for multi-entity collaborative bid-rigging are as follows:

[0079] Step S301: Based on the information of each segment in the semantic-structural dual-dimensional set of identical segments, extract the corresponding bidder identifier, bid submission time and start and end positions of the identical segment, calculate the distribution density of the identical segment in the bid document, establish a topological combination list between any related entities based on the spatial topological relationship of business registration information between bidders, sequentially call the identical distribution density value of the entities within the combination, calculate the density difference, and generate a sequence of identical density difference values ​​of related entities;

[0080] The related subject similarity density difference sequence can be an ordered numerical sequence composed of the difference in the distribution density of similar segments in the respective bid documents of bidding subject pairs with industrial and commercial topological relationships. This sequence can be used to quantify the consistency of the density of similar content among different related subjects and identify abnormal convergence behavior. In this embodiment, the related subject similarity density difference sequence can extract the start and end positions of similar segments for each subject based on a semantic-structural dual-dimensional set of similar segments, calculate the number of similar segments per unit text length as the distribution density, and then calculate the density difference for each pair of related subjects. Furthermore, the related subject similarity density difference sequence can include, but is not limited to, one or more of the following: high-density synchronous difference sequence, low-density asynchronous difference sequence, and locally concentrated difference sequence.

[0081] The calculation of the distribution density of identical segments in the tender document involves establishing a topological combination list based on the spatial topological relationships of business registration information, and generating a sequence of identical density differences among related entities. This can be achieved by: statistically analyzing the distribution density of identical segments for each bidding entity throughout the entire text; constructing a topological relationship graph between entities based on business registration information (such as equity, address, and legal representative), enumerating all related pairs, and calculating their density differences. For example, this operation can be implemented by calculating the distribution density based on the number of identical segments per thousand characters and combining it with a sliding window for smoothing, or by determining the business topological relationships through the shortest path or common control nodes in the enterprise knowledge graph. This couples the spatial distribution characteristics of textual similarity with the entity registration relevance, identifying anomalous entity pairs with highly consistent density.

[0082] Step S302: Based on the starting position of the similar segments of the associated subjects, obtain the text matching starting position offset between each pair of subjects, construct the position offset sequence, call the similarity density difference sequence of associated subjects and the text matching starting position offset sequence, and compare them with the set similarity density tolerance threshold and position synchronization tolerance threshold respectively, filter the subject combinations that are both less than the two tolerance thresholds, and obtain the multi-dimensional abnormal association satisfying combination index set.

[0083] The text matching start position offset sequence can be a sequence of character or paragraph-level offsets of the starting positions of corresponding identical segments in the respective tender documents of associated subject pairs. This sequence can reflect the synchronization of multiple subjects at the document structure level, revealing template reuse or collaborative writing traces. In an exemplary embodiment, the text matching start position offset sequence can record the starting position index of the matching identical segments of each subject pair in the source file and calculate the absolute or relative offset. For example, the text matching start position offset sequence can be a paragraph-level start offset sequence, a chapter-level start offset sequence, a character-level start offset sequence, etc.

[0084] The text matching starting position offset between each pair of subjects is obtained, and a position offset sequence is constructed. This can be achieved by extracting the starting position of matched identical segment pairs in their respective bid documents, calculating the offset, and organizing them into a sequence by subject pair. Furthermore, this operation can be implemented by using the starting position as the paragraph number and the offset as the absolute difference, or by using dynamic time warping to align multiple identical segments and calculating the average offset. This enables the capture of synchronous insertion behavior at the document structure level and enhances the ability to identify template collusion.

[0085] The sequence of similar density differences and the sequence of positional offsets are compared with preset tolerance thresholds. Subject combinations that are simultaneously less than both tolerance thresholds are filtered out. This can be achieved by setting a similar density tolerance threshold T1 and a positional synchronization tolerance threshold T2, retaining only subject pairs where the density difference is less than T1 and the offset is less than T2. ​​In one specific embodiment, this operation can be implemented by adaptively adjusting the tolerance thresholds according to the project type, or by using fuzzy logic to fuse the two differences instead of a hard threshold, thereby achieving joint filtering of dual anomalies of "content density convergence + structural layout synchronization," improving the accuracy of collaborative behavior recognition.

[0086] The multi-dimensional anomaly association satisfying combined index set can be a set of combined indexes of bidding entities that simultaneously satisfy both the similarity density difference and the position offset being below their respective tolerance thresholds. This set can be used as a high-confidence collaborative bid-rigging candidate set to support subsequent cluster strength calculation and risk screening. In this embodiment, the multi-dimensional anomaly association satisfying combined index set can be obtained by comparing the similarity density difference sequence of associated entities and the text matching start position offset sequence with preset tolerance thresholds, retaining the combined indexes that satisfy both conditions.

[0087] Step S303: Call the number identifier of each subject combination in the multi-dimensional abnormal association set to extract the corresponding business information code of the subject in the original identical segment information, construct the subject topology cluster that meets the multi-dimensional abnormal association conditions, calculate and obtain the association cluster strength value, filter according to whether the cluster strength value falls within the procurement risk intensity range, obtain the subject combination that passes the filter, and establish a multi-subject collaborative bid-rigging node cluster.

[0088] The procurement risk intensity range can be an effective range of cluster strength values ​​set based on historical bid-rigging case statistics and project characteristics. It is used to determine whether a cluster of entities has actual bid-rigging risk and can be used to filter weakly related or coincidentally similar combinations of entities, ensuring that the cluster of entities has regulatory and disposal value. In an exemplary embodiment, the procurement risk intensity range can be determined by clustering the intensity distribution of entity clusters in historical high-risk bid-rigging events and combining it with project level weighting to determine the upper and lower bounds.

[0089] The process involves filtering based on whether the cluster strength value falls within the procurement risk intensity range, obtaining the selected entity combinations, and establishing multi-entity collaborative bid-rigging node clusters. This can be achieved by calculating the cluster strength value (such as average density consistency, average offset reciprocal weighting, etc.) for entity combinations that meet multi-dimensional abnormal associations, and determining whether they fall within a preset risk intensity range. Furthermore, this operation can be implemented by comprehensively weighting the cluster strength value based on multiple factors such as the depth of business association, text synchronization, and bidding frequency, or by setting risk intensity ranges hierarchically according to project budget size. This can eliminate low-intensity accidental associations and ensure that the output node clusters possess actual bid-rigging risk characteristics.

[0090] Taking the EPC general contracting bidding of a large-scale water conservancy project as an example, the risk prevention and control method for bidding and evaluation integrating multi-dimensional intelligent analysis in this embodiment can be as follows: The system analysis found that the bid documents of companies A, B, and C had semantic-structural dual-dimensional similarities in the construction organization design chapter. Further calculation showed that the distribution density of their similar segments was 0.82, 0.79, and 0.81 (segments / thousand words), and the difference was less than the tolerance threshold of 0.05. At the same time, the starting position offset of the similar segments of the three companies was 3 segments, 2 segments, and 1 segment, respectively, all less than the position synchronization tolerance threshold of 5 segments. Business registration information showed that the registered addresses of the three companies were located in the same park and there were overlapping senior executives, forming a topological association. The system generated a similarity density difference sequence [0.03, 0.02, 0.01] and a position offset sequence [3, 2, 1] of the associated subjects, and screened out this combination to enter the multi-dimensional abnormal association satisfying combination index set. Its cluster strength value was calculated to be 0.88, which falls within the procurement risk intensity range [0.75, 0.95] corresponding to the project. It was eventually included in the cluster of multi-entity collaborative bid-rigging nodes, triggering a key investigation.

[0091] In one embodiment, the steps for obtaining the information set of the bid-rigging and collusion behavior advancement path are as follows:

[0092] Based on each node cluster in the multi-subject collaborative bid-rigging node cluster, the orientation vector, spatial coordinates of associated information and behavior start time of the subject in the business registration network are extracted. Pairwise combinations between subjects are constructed and the order of the associated network is determined. The spatial Euclidean distance and behavior collaboration start time difference of adjacent subjects in the combination are calculated to generate a set of associated behavior extension distance-time difference values.

[0093] The associated behavior extension distance-time difference set can be a joint metric set composed of the spatial Euclidean distance between pairs of entities in a multi-entity collaborative bidding node cluster and the time difference of behavior collaboration start time. This set can be used to quantify the spatiotemporal cost of inter-entity collaboration and support the judgment of path directionality and temporal rationality. In this embodiment, the associated behavior extension distance-time difference set can be calculated and structured for each pair of adjacent entities based on spatial coordinates and behavior start time in the business registration network. For example, the associated behavior extension distance-time difference set can include, but is not limited to, one or more of the following: short-distance high-synchronization combinations, long-distance low-latency combinations, and spatiotemporal asynchronous abnormal combinations.

[0094] Call the subject combination in the set of related behavior extension distance-time difference, determine whether the angle between the direction difference vector of the related network and the direction of the Euclidean line is less than the related path consistency angle threshold, and at the same time determine whether the behavior coordination time strictly meets the network incremental trend, filter the subject sequence combination that meets the dual constraints, and obtain the network incremental extension combination index set;

[0095] The associated path consistency angle threshold can be a preset upper limit for determining whether the angle between the spatial orientation vector of business registration and the direction of behavior propagation is acceptable. It can be used to ensure that the propagation direction of abnormal bidding behavior is consistent with the business geography or equity topology direction, thereby enhancing the physical interpretability of the path. In an exemplary embodiment, the associated path consistency angle threshold may include, but is not limited to, an acute angle consistency threshold (e.g., <30°), a right angle tolerance threshold (e.g., <60°), and a dynamically adaptive angle threshold.

[0096] The network incremental expansion combined index set can be a set of subject sequence combined indexes that satisfy the dual constraints of consistency in the direction of associated paths and strict incrementality in behavior time. It can be used as a legitimate candidate sequence for constructing effective bid-rigging behavior advancement paths. Furthermore, the network incremental expansion combined index set can filter combinations that simultaneously satisfy angle thresholds and temporal monotonicity from the set of associated behavior expansion distance-time differences. In a specific embodiment, the network incremental expansion combined index set may include, but is not limited to, linearly increasing path indexes and tree-like branching path indexes, while cyclic feedback path indexes are excluded.

[0097] Determining whether the angle between the direction difference vector of the associated network and the direction of the Euclidean connection is less than the consistency angle threshold of the associated path, and simultaneously determining whether the behavioral coordination time strictly satisfies the network's increasing trend, can be achieved by calculating the angle between the commercial and industrial spatial orientation vector of each subject sequence and the actual behavioral propagation direction, and verifying whether the timestamp is strictly monotonically increasing. Further, this operation can be achieved by using the vector dot product formula to calculate the cosine value of the angle, converting it to an angle and comparing it against the threshold, or by converting the time series into a directed acyclic graph and verifying the increasing property through topological sorting. This can eliminate spurious paths with contradictory directions or inverted times, ensuring that the generated behavioral chain possesses causal logic and spatiotemporal consistency.

[0098] Based on the main combination index in the network incremental expansion combination index set, a list of network-connected behavioral trajectory segments is established, a path chain for the advancement of bid rigging and collusion behaviors is constructed, and the cumulative association propagation distance and behavioral time domain window width of the trajectory segments are recorded. The maximum association propagation length and the minimum behavioral time domain window value are extracted to obtain the information set of the advancement path for bid rigging and collusion behaviors.

[0099] Constructing a path chain for bid-rigging and collusion, and recording the cumulative propagation distance and temporal window width of each trajectory segment, can be achieved by sequentially connecting the subject sequences subject to dual constraints into a path chain, accumulating the spatial distance of each segment, and calculating the time difference between the first and last nodes as the window width. Furthermore, this operation can be implemented by storing the path chain as a list of directed graph edges with attached distance and time window attributes, or by encoding the path using a spatiotemporal trajectory data model, thereby forming a structured risk evolution trajectory with directionality, cumulative propagation cost, and time span.

[0100] The behavioral time-domain window width can be the time span from the first node to the last node in the path chain of bid-rigging behavior. It can be used to reflect the duration of abnormal bidding organizations and assess the speed and concealment of risk evolution. For example, the behavioral time-domain window width can include, but is not limited to, short-term intensive windows (<7 days), medium-term planning windows (7–30 days), and long-term penetration windows (>30 days).

[0101] For example, in the scenario of abnormal bidding path simulation for cross-provincial power equipment procurement, the risk prevention and control method for bidding and evaluation integrating multi-dimensional intelligent analysis in this embodiment can be as follows: In a bidding process for equipment in a 500kV substation, the system identifies a cluster of multi-entity collaborative bid-rigging nodes consisting of companies M, N, and O. Their business registration coordinates are extracted as (116.4, 39.9), (116.5, 40.0), and (116.6, 40.1), with the behavior start times being 2024-03-01, 2024-03-03, and 2024-03-05, respectively. The angle between the Euclidean connection direction of M→N→O and the business network topology direction (based on equity control flow) is calculated to be 22°, which is less than the set threshold of 30° for the consistency angle of the associated path; the time series is strictly increasing. This sequence is included in the network incremental expansion combined index set. The system constructs a propagation path chain, with a cumulative propagation distance of 28.3 kilometers and a behavior time domain window width of 4 days. Ultimately, this path, with a maximum propagation length of 28.3 kilometers and a minimum window width of 4 days, was included in the information set of the path for bid rigging and collusion, and marked as a "short-term targeted penetration" high-risk path.

[0102] In one embodiment, the steps for obtaining the list of high-risk bid-rigging and collusion paths are as follows:

[0103] Based on the trajectory end node information of the information set of the path of bid rigging and collusion behavior, the corresponding main business information code and the spatial coordinates of the end node are extracted. According to the information code, the risk threshold benchmark set of the relevant bidding and procurement project is retrieved. The coordinate values ​​are called to calculate the spatial Euclidean distance between the end node and all critical points in the risk threshold benchmark set, and the distance information from the end of the path to the risk threshold point is generated.

[0104] The risk threshold benchmark set can be a set of spatial coordinate points representing the critical state of bid-rigging behavior, pre-defined based on the project level, historical bid-rigging cases, and regulatory rules. It can serve as an objective reference system for measuring the degree of risk approach at the end of a path. In this embodiment, the risk threshold benchmark set can be constructed by clustering the main business coordinates and behavioral characteristics of historical high-risk bid-rigging events, and hierarchically constructing critical points according to project level. For example, the risk threshold benchmark set may include, but is not limited to, one or more of the following: critical point sets for national-level major projects, local livelihood projects, and routine procurement.

[0105] The distance information from the end of the path to the risk threshold point can be a set of spatial Euclidean distances between the end node of the bid-rigging path and each critical point in the risk threshold benchmark set. This information can be used to quantify the proximity of the end entity to known high-risk patterns. In an exemplary embodiment, the distance information from the end of the path to the risk threshold point can be obtained by calculating the Euclidean distance based on the business registration coordinates of the end node and the coordinates of each critical point, and then storing it in a structured manner. Furthermore, the distance information from the end of the path to the risk threshold point can include, but is not limited to, minimum distance records, full distance vectors, weighted distance sequences, etc.

[0106] The spatial Euclidean distance between the endpoint and all critical points in the risk threshold benchmark set is calculated by calling coordinate values. This can be done using the latitude and longitude of the endpoint's business registration or planar coordinates, and then calculating the Euclidean distance with the coordinates of each critical point in the risk threshold benchmark set. In a specific embodiment, this operation can be implemented more efficiently by calculating the spherical distance using the Haversine formula in the WGS84 geographic coordinate system, or by directly calculating the planar Euclidean distance in the projected coordinate system. This allows abstract risks to be transformed into measurable spatial proximity indicators.

[0107] Based on the distance information from the end of the path to the risk threshold point, the minimum distance value corresponding to each end node of the path is extracted, the risk distance sequence of the end node is constructed, the behavior-related critical distance benchmark value is called to compare each value in the distance sequence, the path combination index that is less than or equal to the benchmark value is filtered, and the risk path index set of bid rigging and penetration is obtained.

[0108] The behavioral correlation critical distance benchmark value can be a preset spatial distance upper limit used to determine whether the end of a path constitutes a risk of bid rigging penetration. It can provide a unified and objective risk assessment threshold, avoiding subjective bias. For example, the behavioral correlation critical distance benchmark value can include, but is not limited to, a fixed global benchmark value, a project-level adaptive benchmark value, or a dynamic sliding window benchmark value. The bid rigging penetration risk path index set can be a set of path numbers whose minimum distance value at all end nodes is less than or equal to the behavioral correlation critical distance benchmark value. It can be used to identify high-risk path candidates that are actually approaching or exceeding the critical state of bid rigging. In a specific embodiment, the bid rigging penetration risk path index set can be obtained by comparing the risk distance sequence of end nodes item by item with the benchmark value and then filtering for compliant path IDs.

[0109] The process compares each value in the distance sequence with the behavior-related critical distance benchmark, filtering path combinations whose values ​​are less than or equal to the benchmark. This can be achieved by checking if the minimum distance at the end of each path is less than or equal to the behavior-related critical distance benchmark, retaining the paths that meet the condition. Furthermore, this operation can be accelerated by using vectorized comparisons for batch processing, or by introducing fuzzy matching to tolerate boundary disturbances. This allows for the objective identification of bid-rigging paths that have entered high-risk areas, while eliminating low-risk noise at distant points.

[0110] Based on the path number identified in the risk path index set for bid rigging, the corresponding trajectory segment information is extracted from the information set on the path for bid rigging and collusion. The risk path list for bid rigging is then reconstructed and the shortest penetration distance between the end of the path and the risk threshold point is marked to obtain the list of high-risk bid rigging and collusion paths.

[0111] Reconstructing a list of risk penetration paths for bid rigging and labeling the shortest penetration distance between the end of each path and a risk threshold point can be achieved by extracting the complete trajectory corresponding to the risk penetration path index set from the information set of bid rigging and collusion behavior advancement paths, and attaching the shortest distance value as a risk intensity label. In an exemplary embodiment, this operation can be achieved by outputting in JSON-LD format and embedding the distance value as an attribute, or by marking high-risk path nodes in a graph database and associating them with distance metadata, thereby outputting a high-risk decision list that is interpretable and sortable.

[0112] For example, in the scenario of bid-rigging warning for large data center construction projects, the procurement and evaluation risk prevention and control method integrating multi-dimensional intelligent analysis in this embodiment can be as follows: In a server procurement project for a supercomputing center, the system generates a path for bid-rigging behavior, with the terminal node being Company P, whose business registration coordinates are (121.5, 31.2). Based on the fact that this project is a "national-level major project," the corresponding risk threshold benchmark set is retrieved, containing three critical points: (121.48, 31.19), (121.52, 31.21), and (121.60, 31.25). The calculated Euclidean distances from Company P to the three points are 0.18km, 0.15km, and 0.42km, respectively, with a minimum distance of 0.15km. The behavioral association critical distance benchmark value is set to 0.20km (national-level project standard). Since 0.15 ≤ 0.20, this path is included in the bid-rigging penetration risk path index set. The system extracts its complete trajectory segment, marks it with "shortest penetration distance = 0.15km", generates a list of high-risk bid rigging and collusion paths, and pushes it to the procurement supervision platform for priority verification.

[0113] In one embodiment, after obtaining the list of high-risk bid-rigging and collusion paths, the method further includes:

[0114] For each path in the list of high-risk bid rigging and collusion paths, extract the maximum semantic similarity change amplitude within the identical segment of the main body at the starting point of the path, the total time span of the path evolution, and the number of abnormally associated entities in the path, calculate the corresponding procurement risk index, mark the handling priority according to the level of the procurement project, and generate multi-dimensional procurement risk labeling results.

[0115] The multi-dimensional procurement risk assessment results include risk level labels, procurement project identifiers corresponding to the level, and procurement risk index codes.

[0116] The maximum semantic similarity change amplitude can be the difference between the maximum fluctuation peak and the baseline of the similar segments at the path's starting point in the dynamic sliding window semantic similarity calculation. This can reflect the intensity of text plagiarism and the complexity of the avoidance strategies—a larger amplitude indicates more obvious deliberate reconstruction. In an exemplary embodiment, the maximum semantic similarity change amplitude can be obtained by calculating the range of the sliding window similarity sequence. Exemplarily, the maximum semantic similarity change amplitude can include, but is not limited to, one or more of the following: locally abrupt amplitude, multi-peak superposition amplitude, and edge steep rise amplitude.

[0117] The total time span of path evolution can be the time interval from the first node to the last node in the path of bid-rigging and collusion. It can be used to reflect the duration and planning maturity of the bid-rigging organization. A longer span usually indicates that the group's operation is more covert and systematic. Furthermore, the total time span of path evolution can be calculated directly from the timestamps read from the path chain metadata. In a specific embodiment, the total time span of path evolution can adopt short-term assault (<7 days), medium-term planning (7–30 days), long-term infiltration (>30 days), etc.

[0118] The number of abnormally associated entities can be the total number of bidding entities identified as colluding in a high-risk path. It can be used to characterize the scale and organization of the collusion network; the more entities, the higher the complexity of the collusion. In this embodiment, the number of abnormally associated entities can be obtained by statistically analyzing the entity list in the path. For example, the number of abnormally associated entities can include, but is not limited to, two-entity pairings, three-to-five-entity small groups, and six or more-entity networked groups. The procurement risk index can be a comprehensive risk quantification indicator calculated based on the maximum semantic similarity change amplitude, the total time span of path evolution, and the number of abnormally associated entities. It can be used to uniformly characterize the overall risk intensity of collusion paths and supports cross-project horizontal comparison and priority ranking. Furthermore, the procurement risk index can be normalized from the three original indicators and then linearly combined according to preset weights or mapped to a 0–1 interval score through a lightweight model. In a specific embodiment, the procurement risk index can include, but is not limited to, a high-plagiarism intensity dominant index, a long-cycle organized index, and a large-scale collaborative index.

[0119] The disposal priority can be a regulatory response level jointly determined by the procurement project level and the procurement risk index. This can guide the efficient allocation of regulatory resources according to the risk-value ratio, ensuring that high-value, high-risk projects are disposed of first. In this embodiment, the disposal priority can adopt a high-risk priority for national-level projects, a medium-risk priority for provincial-level livelihood projects, and a low-risk filing level for routine procurement. The multi-dimensional procurement risk labeling results can be a structured risk output set containing risk level labels, the identification of the procurement project, and the procurement risk index code. This can be used to provide traceable, sortable, and business-semantic action basis for regulatory decisions. Furthermore, the multi-dimensional procurement risk labeling results can map the procurement risk index to a preset level range and associate it with project metadata to generate standardized labeling records. For example, the multi-dimensional procurement risk labeling results can include, but are not limited to, Level 1 emergency disposal labeling, Level 2 key verification labeling, and Level 3 filing observation labeling.

[0120] Extracting the maximum semantic similarity change amplitude within the similar segments at the starting point of the path, the total temporal span of the path evolution, and the number of abnormally associated entities in the path can be achieved by parsing the similarity fluctuation curves of the similar segments at the starting point, the timestamps of the first and last nodes, and the entity list from the high-risk path list, extracting the three indicators respectively. Furthermore, this operation can be implemented by calculating the maximum amplitude using the range of the similarity sequence through a sliding window, or by directly reading the time span and entity count from the path chain metadata, thereby obtaining multi-dimensional heterogeneous features to support the comprehensive quantification of the risk index. Calculating the corresponding procurement risk index and assigning disposal priorities according to the level of the procurement project can be achieved by normalizing the three indicators and then weighting and summing them to obtain the risk index, which is then mapped to a preset priority rule based on the project level. Further, this operation can use the Analytic Hierarchy Process (AHP) to determine the indicator weights, or use a piecewise function to cross-map the risk index and project level to priority (e.g., national level + index > 0.8 → first-level priority), thereby achieving a closed-loop transformation from risk identification to classification and then to disposal instructions.

[0121] Taking the evaluation of bid-rigging in the centralized procurement of drugs under the national medical insurance scheme as an example, the risk prevention and control method for bidding and evaluation integrating multi-dimensional intelligent analysis in this embodiment can be as follows: In a certain insulin centralized procurement project, the maximum semantic similarity change amplitude of the segment with identical main entities at the starting point of a high-risk path reached 0.42 (background mean 0.15), the total time span of the path evolution was 22 days, and it involved 5 abnormally associated entities. After system normalization, the procurement risk index was calculated to be 0.86 according to the weights (amplitude 0.4, time span 0.3, quantity 0.3). Because this project belongs to the national-level centralized procurement, the system automatically marked it as "Level 1 Emergency Response Priority". The final multi-dimensional procurement risk labeling result includes: risk level label "high risk", project identifier "GY-YB-2024-INSULIN", risk index code "RISK-0.86-G1", which is pushed to the intelligent supervision platform of the medical insurance regulatory agency to initiate a special verification.

[0122] In one embodiment, the steps for obtaining the multi-dimensional procurement risk assessment results are as follows:

[0123] Based on each path in the list of high-risk bid rigging and collusion paths, the semantic similarity sequence of the similar segments of the subjects corresponding to the starting point of the path is extracted, the difference between the maximum value in the sequence and the minimum value of the historical baseline is identified, and the normalization process is performed based on the complexity of the bidding and procurement project to obtain the normalized risk amplitude of each path and generate a path risk amplitude sequence.

[0124] The normalized risk amplitude can be a standardized indicator, calculated by weighting the difference between the maximum semantic similarity of the similarity segment at the starting point of the path and the minimum historical baseline, after adjusting for project complexity. This standardization can eliminate interference factors such as project size and text length, ensuring cross-project comparability of similarity fluctuations. In this embodiment, the normalized risk amplitude can be obtained by extracting the maximum value from the dynamic sliding window similarity sequence of the similarity segment at the starting point, subtracting the minimum similarity baseline of historical normal bidding for that type of project, and then dividing by the project complexity coefficient. For example, the normalized risk amplitude can include, but is not limited to, one or more of the following: correction amplitude for high-complexity projects, correction amplitude for medium-complexity projects, and correction amplitude for low-complexity projects.

[0125] The path risk magnitude sequence can be a numerical sequence organized by path ID from the normalized risk magnitude of each high-risk path. It can be used as a comparable measure to uniformly characterize the intensity of textual similarity anomalies across different projects, supporting cross-project risk assessment. Furthermore, the operational principle of the path risk magnitude sequence can be explained in context: it is generated based on the difference between the maximum value of the semantic similarity sequence of similar segments and the minimum value of the historical baseline, combined with normalization by the procurement project complexity factor. In an exemplary embodiment, the path risk magnitude sequence can adopt technical bid-driven magnitude sequences, commercial clause anomaly magnitude sequences, and full-text mixed magnitude sequences, etc.

[0126] Extracting the semantic similarity sequence of identical segments corresponding to the starting point of the path, identifying the difference between the maximum value within the sequence and the minimum value of the historical baseline, and normalizing it based on the complexity of the procurement project, can be viewed as extracting the maximum value from the dynamic sliding window similarity sequence of identical segments at the starting point, subtracting the minimum similarity baseline value of historical normal bidding for this type of project, and then dividing by the project complexity coefficient. Furthermore, this operation can be achieved by using a similarity lower limit below the 95th percentile of similar projects for the minimum historical baseline value, and by automatically calculating the clause density of the project complexity using NLP parsing technology, thereby generating a standardized anomaly intensity index that eliminates the impact of project heterogeneity.

[0127] Call the path information corresponding to the path risk magnitude sequence, extract the start and end times of each path and calculate the time domain span, count the number of bidding entities marked as abnormal association status within the path, combine the three indicators to construct a multi-parameter risk assessment set, calculate and obtain the procurement risk index value, and generate a procurement risk index value set.

[0128] The procurement risk index set can be a collection of procurement risk index values ​​corresponding to all high-risk paths. Each value is calculated by weighting the normalized risk amplitude, time span, and number of abnormal entities, and can be used to provide a basis for ranking risk intensity under a unified quantitative scale. In a specific embodiment, the procurement risk index set can be obtained by weighted fusion of three indicators from a multi-parameter risk assessment set. For example, the procurement risk index set may include, but is not limited to, a high plagiarism intensity-dominated index set, a long-cycle organizational index set, and a large-scale collaborative index set.

[0129] A multi-parameter risk assessment set is constructed by weighting three indicators, and the procurement risk index value is obtained by calculation. This can be achieved by combining normalized risk amplitude, path time span (normalized), and number of abnormal associated entities (normalized) according to preset weights to calculate the comprehensive risk index for each path. Furthermore, this operation can be achieved by determining the weights through expert scoring (AHP) and using a logistic regression model to learn the indicator contribution of historical bid-rigging cases, thereby realizing the fusion and quantification of multi-dimensional heterogeneous risk characteristics and improving the comprehensiveness of the assessment.

[0130] Based on the index corresponding to each path in the procurement risk index value set, the procurement project level identifier to which the path belongs is retrieved. According to the preset risk classification benchmark value range within the procurement project level, the procurement risk index is classified and judged, and the corresponding handling priority level is marked to obtain multi-dimensional procurement risk labeling results.

[0131] The preset risk grading benchmark range within the procurement project level can be a risk index grading threshold range set separately for each procurement project level (such as national, provincial, and regular). This can be used to achieve dynamic matching between risk level determination and project importance, avoiding misjudgments caused by applying a one-size-fits-all threshold. In an exemplary embodiment, the preset risk grading benchmark range within the procurement project level may include, but is not limited to, high / medium / low risk ranges for national-level projects, high / medium / low risk ranges for provincial-level livelihood projects, and high / medium / low risk ranges for regular procurement projects.

[0132] Based on the preset risk grading benchmark range within the procurement project level, the procurement risk index is graded and the corresponding disposal priority level is marked. This can be achieved by querying the corresponding risk grading benchmark range according to the project level to which the path belongs, mapping the procurement risk index to high / medium / low levels, and assigning corresponding disposal priority labels. Furthermore, this operation can be implemented using interval mapping tables (e.g., national level: [0.8, 1.0] → high risk) or introducing fuzzy membership functions to handle boundary values, thereby achieving dynamic matching between risk levels and project value, ensuring accurate allocation of regulatory resources.

[0133] For example, in the scenario of risk classification for joint procurement of water conservancy project equipment across regions, the risk prevention and control method for bidding and evaluation integrating multi-dimensional intelligent analysis in this embodiment can be as follows: In the bidding for pump station equipment of a cross-provincial water diversion project, the system identifies a high-risk path. The maximum semantic similarity sequence of its starting point segment is 0.93, the minimum historical baseline is 0.60, and the difference is 0.33; the project complexity coefficient is 1.2 (due to the inclusion of 27 technical parameters), and the normalized risk amplitude is 0.275. The path's time span is 18 days (normalized value 0.6), and the number of abnormal associated entities is 4 (normalized value 0.8). The procurement risk index is calculated to be 0.65 based on weights (amplitude 0.4, time 0.3, quantity 0.3). This project belongs to the "National Key Water Conservancy" category, and its risk classification benchmark range is: high risk ≥ 0.6. The system determines the risk to be high, marks it as "Level 1 Priority", generates a multi-dimensional procurement risk assessment result, including risk level "high risk", project identifier "SL-GJ-2024-PUMP", index code "RISK-0.65-G1", and pushes it to the relevant competent authority's electronic procurement supervision platform.

[0134] Furthermore, to achieve the above objectives, the present invention also provides a procurement and evaluation risk prevention and control system integrating multi-dimensional intelligent analysis. The system includes: a memory, a processor, and a procurement and evaluation risk prevention and control program integrating multi-dimensional intelligent analysis stored in the memory and executable on the processor. The procurement and evaluation risk prevention and control program integrating multi-dimensional intelligent analysis is configured to implement the steps of the procurement and evaluation risk prevention and control method integrating multi-dimensional intelligent analysis as described above.

[0135] In addition, to achieve the above objectives, the present invention also provides a medium storing a procurement and evaluation risk control program integrating multi-dimensional intelligent analysis, wherein when the procurement and evaluation risk control program integrating multi-dimensional intelligent analysis is executed by a processor, it implements the steps of the procurement and evaluation risk control method integrating multi-dimensional intelligent analysis as described above.

[0136] Other embodiments or specific implementations of the procurement and evaluation risk prevention and control system integrating multi-dimensional intelligent analysis described in this invention can be referred to the above-mentioned method embodiments, and will not be repeated here.

[0137] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for risk prevention and control in procurement and evaluation integrating multi-dimensional intelligent analysis, characterized in that, The method includes: Based on multi-source heterogeneous text data in the procurement document library and the tender document library, semantic vector mapping is performed through a pre-trained language model, fine-grained text similarity calculation is performed, suspected plagiarized paragraphs and implicit similar content that avoids keyword replacement are identified, and a set of semantic-structural dual-dimensional similar paragraphs is generated. Based on the information of each segment in the semantic-structural dual-dimensional similar segment set, calculate the text similarity matrix between bidding entities and the matching offset of key clauses, filter bidding entity clusters that meet the multi-dimensional abnormal association conditions, and generate multi-entity collaborative bid-rigging node clusters. Based on each node cluster in the multi-entity collaborative bid-rigging node cluster, extract the business registration information of the associated entities, the time series of bidding behavior and the associated network topology, construct the entity association evolution trajectory chain, and generate a set of information on the advancement path of bid-rigging behavior. Based on the path end node information in the information set of the bid-rigging and collusion behavior promotion path, the corresponding bidding and procurement risk threshold benchmark set is extracted, the semantic space distance from the path end node to the critical point of bid-rigging behavior is measured and compared with the behavior association critical distance, the path set that meets the bid-rigging penetration condition is screened, and a list of high-risk bid-rigging and collusion paths is generated.

2. The method for risk prevention and control in procurement and evaluation integrating multi-dimensional intelligent analysis as described in claim 1, characterized in that, The semantic-structural dual-dimensional similarity segment set includes semantic similarity values, structural similarity indicators, and the duration of implicit similarity segments within the similarity segments; the multi-entity collaborative bid-rigging node cluster includes the bid entity number, inter-entity association strength index, and bid space continuity identifier that meet the multi-dimensional abnormal association conditions; the bid-rigging and collusion behavior advancement path information set includes the business information topological distance value, behavior collaboration time delay value, and association propagation direction sequence of each node in the trajectory chain; the high-risk bid-rigging and collusion path list includes the business registration coordinate information of the entity at the path terminal node, the risk threshold distance measurement value, and the corresponding procurement project level code.

3. The method for risk prevention and control in procurement and evaluation integrating multi-dimensional intelligent analysis as described in claim 2, characterized in that, The specific steps for obtaining the semantic-structural dual-dimensional identical segment set are as follows: Based on the text content of technical specifications and business terms in the procurement document library, hierarchical semantic encoding is performed through the BERT pre-trained model. Dynamic sliding window semantic similarity calculation is performed on each text segment to judge the similarity change fluctuation characteristics, identify abnormal similar segments whose fluctuation amplitude exceeds the set threshold, and generate a set of semantically similar text segments. Based on each segment in the set of semantically similar segments, extract the keyword replacement pattern data in the corresponding bidding documents, calculate the keyword distribution entropy value for each segment of text, evaluate the degree of implicit similarity in avoiding detection, filter key abnormal segments by setting an implicit similarity threshold, and generate a structural avoidance feature set. Based on the set of semantically similar segments and the set of structural avoidance features, spatiotemporal correlation analysis is performed on the similar segments. By judging the coupling relationship between semantic similarity and structural avoidance, segments that meet the characteristics of bid-rigging behavior are selected, and a set of semantically-structurally similar segments is generated.

4. The bidding and tendering risk prevention and control method integrating multi-dimensional intelligent analysis as described in claim 3, characterized in that, The specific steps for obtaining the multi-entity collaborative bid-rigging node cluster are as follows: Based on the information of each segment in the semantic-structural dual-dimensional identical segment set, the corresponding bidder identifier, bid submission time and start and end positions of the identical segment are extracted. The distribution density of the identical segment in the bid document is calculated. Based on the spatial topological relationship of the business registration information between bidders, a topological combination list between any related entities is established. The identical distribution density values ​​of the entities in the combination are called in turn, the density difference is calculated, and a sequence of identical density difference values ​​of related entities is generated. Based on the starting position of the similar segments of the associated subjects, obtain the text matching starting position offset between each pair of subjects, construct the position offset sequence, call the similarity density difference sequence of the associated subjects and the text matching starting position offset sequence, and compare them with the set similarity density tolerance threshold and position synchronization tolerance threshold respectively, filter the subject combinations that are simultaneously less than the two tolerance thresholds, and obtain the multi-dimensional abnormal association satisfying combination index set. The multi-dimensional abnormal association satisfies the number identifier of each subject combination in the combined index set, extracts the corresponding business information code of the subject in the original similar segment information, constructs the subject topology cluster that meets the multi-dimensional abnormal association conditions, calculates and obtains the association cluster strength value, and filters according to whether the cluster strength value falls within the procurement risk intensity range, obtains the subject combination that passes the filter, and establishes a multi-subject collaborative bid-rigging node cluster.

5. The method for risk prevention and control in procurement and evaluation integrating multi-dimensional intelligent analysis as described in claim 4, characterized in that, The specific steps for obtaining the information set of the promotion path of bid-rigging and collusion behavior are as follows: Based on each node cluster in the multi-subject collaborative bid-rigging node cluster, the orientation vector, spatial coordinates of associated information and behavior start time of the subject in the business registration network are extracted. Pairwise combinations between subjects are constructed and the order of the associated network is determined. The spatial Euclidean distance and behavior collaboration start time difference of adjacent subjects in the combination are calculated to generate a set of associated behavior extension distance-time difference values. The main body combination in the aforementioned associated behavior extension distance-time difference set is called, and it is determined whether the angle between the associated network direction difference vector and the Euclidean connection direction is less than the associated path consistency angle threshold. At the same time, it is determined whether the behavior coordination time strictly meets the network incremental trend. The main body sequence combination that meets the dual constraints is selected to obtain the network incremental extension combination index set. Based on the main combination index in the network incremental expansion combination index set, a list of network-connected behavioral trajectory segments is established, a path chain for promoting bid-rigging and collusion behavior is constructed, and the cumulative association propagation distance and behavioral time-domain window width of the trajectory segments are recorded. The maximum association propagation length and the minimum behavioral time-domain window value are extracted to obtain the information set of the path for promoting bid-rigging and collusion behavior.

6. The method for risk prevention and control in procurement and evaluation integrating multi-dimensional intelligent analysis as described in claim 5, characterized in that, The specific steps for obtaining the list of high-risk bid-rigging and collusion paths are as follows: Based on the trajectory end node information of the information set of the bid-rigging and collusion behavior, the corresponding main business information code and the spatial coordinates of the end node are extracted. According to the information code, the risk threshold benchmark set of the relevant bidding and procurement project is retrieved. The coordinate values ​​are called to calculate the spatial Euclidean distance between the end node and all critical points in the risk threshold benchmark set, and the distance information from the end of the path to the risk threshold point is generated. Based on the distance information from the end of the path to the risk threshold point, the minimum distance value corresponding to each end node of the path is extracted, the risk distance sequence of the end node is constructed, the behavior-related critical distance benchmark value is called to compare each value in the distance sequence, the path combination index that is less than or equal to the benchmark value is filtered, and the risk path index set of bid rigging and penetration is obtained. Based on the path number identified in the risk path index set for bid rigging, the corresponding trajectory segment information is extracted from the information set on the path advancement of bid rigging and collusion behavior. The risk path rigging risk is reconstructed and the shortest penetration distance between the end of the path and the risk threshold point is marked to obtain the list of high-risk bid rigging and collusion paths.

7. The method for risk prevention and control in procurement and evaluation integrating multi-dimensional intelligent analysis as described in claim 6, characterized in that, After obtaining the list of high-risk bid-rigging and collusion paths, the method also includes: For each path in the list of high-risk bid rigging and collusion paths, the maximum semantic similarity change amplitude within the identical segment at the starting point of the path, the total time span of the path evolution, and the number of abnormally associated entities in the path are extracted. The corresponding procurement risk index is calculated, and the handling priority is marked according to the level of the procurement project to which it belongs, generating multi-dimensional procurement risk labeling results. The multi-dimensional procurement risk assessment results include risk level labels, procurement project identifiers corresponding to the levels, and procurement risk index codes.

8. The method for risk prevention and control in procurement and evaluation integrating multi-dimensional intelligent analysis as described in claim 7, characterized in that, The specific steps for obtaining the multi-dimensional procurement risk assessment results are as follows: Based on each path in the list of high-risk bid rigging and collusion paths, the semantic similarity sequence of the similar segments of the subjects corresponding to the starting point of the path is extracted, the difference between the maximum value in the sequence and the minimum value of the historical baseline is identified, and normalization is performed based on the complexity of the bidding and procurement project to obtain the normalized risk amplitude of each path and generate a path risk amplitude sequence. Call the path information corresponding to the path risk magnitude sequence, extract the start and end times of each path and calculate the time domain span, count the number of bidding entities marked as abnormal association status within the path, combine the three indicators to construct a multi-parameter risk assessment set, calculate and obtain the procurement risk index value, and generate a procurement risk index value set. Based on the index corresponding to each path in the procurement risk index value set, the procurement project level identifier to which the path belongs is retrieved. According to the preset risk classification benchmark value range within the procurement project level, the procurement risk index is classified and judged, and the corresponding disposal priority level is marked to obtain multi-dimensional procurement risk labeling results.

9. A procurement and evaluation risk prevention and control system integrating multi-dimensional intelligent analysis, characterized in that, The system includes: a memory, a processor, and a procurement and evaluation risk prevention and control program integrating multi-dimensional intelligent analysis stored in the memory and executable on the processor. The procurement and evaluation risk prevention and control program integrating multi-dimensional intelligent analysis is configured to implement the steps of the procurement and evaluation risk prevention and control method integrating multi-dimensional intelligent analysis as described in any one of claims 1 to 8.

10. A medium, characterized in that, The medium stores a procurement and evaluation risk prevention and control program that integrates multi-dimensional intelligent analysis. When the procurement and evaluation risk prevention and control program that integrates multi-dimensional intelligent analysis is executed by the processor, it implements the steps of the procurement and evaluation risk prevention and control method that integrates multi-dimensional intelligent analysis as described in any one of claims 1 to 8.

Images