An intelligent bidding risk control early warning method

Abstract

The application discloses a kind of intelligent bidding risk control early warning method, belong to artificial intelligence and big data analysis technical field, including the multidimensional correlation degree between rules by constructing rule grid engine calculation, form the dynamic expected correlation table of rule;Standardized metadata model and abnormal mapping visualization mechanism are established simultaneously, can focus on associated risk link after abnormal triggering quickly, and with mapping table form is shown, solve the technical problem that rule is not isolated in bidding risk control system, improve detection efficiency and abnormal result is directly visualized and shown, the correlation degree between rules is calculated by the application, realize the grid coupling between rules, improve the systematicness and accuracy of risk control detection, compare relevant rules using expected correlation table, significantly reduce redundant matching, improve early warning response speed, by mapping relationship, abnormal data is visualized into two-dimensional table structure, it is convenient for manager to determine risk source and influence scope intuitively.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and big data analysis technology, and in particular relates to an intelligent bidding risk control and early warning method. Background Technology

[0002] In the fields of bidding management and corporate credit risk control, existing risk identification methods mostly rely on static rule matching or post-audit analysis. Their core mechanism typically involves setting several risk rules in bidding data or corporate information tables to perform batch screening of input data to detect anomalies. However, these methods have the following shortcomings:

[0003] (1) The rules are independent of each other and lack dynamic correlation calculation between the rules, resulting in one-sided, repetitive or omission of risk warning results;

[0004] (2) The rule triggering mechanism is singular and cannot dynamically adjust the detection path according to the logical or historical coupling relationship between rules, resulting in low computational efficiency;

[0005] (3) When multi-dimensional anomalies occur, existing systems usually only output textual anomaly items, lacking clear visual mapping, which makes it difficult to assist human decision-making and tracking. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent bidding risk control and early warning method, which solves the technical problems of rules not being isolated in the bidding risk control system, improving detection efficiency, and displaying abnormal results intuitively.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A smart bidding risk control and early warning method includes the following steps:

[0009] Step 1: In the general template module, pre-define rule G0 and data summary template table G1, extract and construct the key field set and core entity set of each rule in rule G0, construct the rule grid engine RME1, and use the rule grid engine RME1 to calculate the initial correlation degree W' between any two rules. ij And by setting constraints, the final correlation degree W is obtained. ij This leads to the expected association table of the rules being obtained and output;

[0010] Step 2: The data acquisition module obtains the table data input by the user from the client, preprocesses the acquired table data to obtain initial metadata, extracts the data type, timestamp, and key fields from the initial metadata, and reorganizes it according to the field structure of the data summary template table G1 to obtain the standardized metadata record Meta. i The initial metadata is compared with the corresponding metadata record. iBind the metadata to obtain complete metadata and construct a metadata set, MetaSet.

[0011] Step 3: The anomaly warning module retrieves the metadata set MetaSet and compares any piece of metadata in it with each rule in rule G0. During the comparison, if an anomaly is detected in a certain node rule, the expected association table is called, and other rules associated with that rule are compared. Comparisons of unrelated rules are stopped, and the metadata record Meta is updated. i The exception marker generates an exception metadata record nMeta. i ;

[0012] Iterate through the MetaSet metadata collection. Once all metadata in the collection has undergone rule comparison, record the Meta data based on the metadata. i Extract exception metadata from the exception marker field and construct an exception metadata set MetaSet. anomaly ;

[0013] Step 4: The visualization module retrieves the exception metadata set MetaSet. anomaly ;

[0014] Based on the metadata record nMeta for each abnormal metadata record i and its corresponding initial metadata record D i Establish mapping relationship Map(nMeta) i →D i Based on the mapping relationship Map, an exception mapping table is constructed, and the exception metadata collection MetaSet is added. anomaly Each piece of abnormal metadata is filled into the corresponding position in the abnormal mapping table according to the mapping relationship, and a complete abnormal mapping result table is generated. The generated abnormal mapping table is then sent to the client for visualization.

[0015] Preferably, when performing step 1, the general template module specifically performs the following steps:

[0016] Step 1-1: The user inputs rule G0 through the client, and the client sends rule G0 to the general template module for storage. Rule G0 includes rules for bid rigging and collusion nodes, bidding behavior nodes, and bid awarding behavior early warning nodes.

[0017] Steps 1-2: Construct a general template for the data summary template table G1, including unique ID, type enumeration, timestamp, key fields, anomaly markers, and text descriptions;

[0018] The general template of data summary template table G1 is for three nodes: bid rigging and collusion node, bidding behavior node, and bid award behavior early warning node. The general fields include unique ID, type enumeration, timestamp, key fields, and text description.

[0019] For each node, the differentiated item is marked as an anomaly, specifically:

[0020] The specific rule subset for bid-rigging nodes includes: signature information, registration information, file author information, and abnormal judgment rules for not winning bids for six months and winning bids not being covered by profits;

[0021] The specific rule subset for bidding behavior nodes includes: anomaly judgment rules for winning bids exceeding failed bids, modification of project approval documents, modification of bidding details, and changes in technical weights;

[0022] The subset of rules specific to the benchmarking behavior node includes: anomaly judgment rules for various winning bid prices;

[0023] Steps 1-3: Construct the rule grid engine RME1, use the rule grid engine RME1 to parse rule G0, and calculate the key field sharing ratio, entity sharing ratio, and correlation degree W among the rules. ij Generate a list of expected associations for the rules, including the associated rule nodes and the association degree W. ij And description of association rules;

[0024] Steps 1-4: Output the fixed expected association table.

[0025] Preferably, when performing steps 1-3, the construction and application of the rule grid engine RME1 specifically includes the following steps:

[0026] Step 1-3-1: Construct the regular mesh engine RME1, the specific model is as follows:

[0027] W' xy =α×FieldOverlap(x,y)+γ×EntityShare(x,y);

[0028] FieldOverlap(x, y) = {F x ∩F y}÷{F x ∪F y};

[0029] EntityShare(x, y) = {E x ∩E y}÷{E x ∪E y};

[0030] Where FieldOverlap is the proportion of key fields shared between rules; EntityShare is the proportion of entities shared between rules; W is the degree of relevance; F x F represents the set of key fields for rule x. y E represents the set of key fields for rule y; x E represents the set of entities that rule x is expected to or historically associated with. y Represents the set of entities that rule y is expected to be associated with, or that it is historically associated with.

[0031] FieldOverlap takes values ​​of [0,1]. If rule x and rule y have no overlap in fields, then FieldOverlap=0; if the key field sets are exactly the same, then FieldOverlap=1. EntityShare takes values ​​of [0,1]. If the entities involved in rule x and rule y have no overlap, then EntityShare=0; if rule x and rule y have the same set of entities, then EntityShare=1.

[0032] α and γ are both adjustment parameters, taking values ​​α, γ∈[0,1], and α+γ=1;

[0033] Step 1-3-2: Read rule G0, and using the rule grid engine RME1, calculate the original correlation degree W' between each rule in rule G0 and other rules. ij :

[0034] W' ij =α×FieldOverlap(i,j)+γ×EntityShare(i,j);

[0035] Where FieldOverlap(i,j) represents the proportion of key fields shared between rule i and rule j; EntityShare(i,j) represents the proportion of the same core entity involved in rule i and rule j, i.e., the entity sharing proportion.

[0036] Construct shared constraints for key fields:

[0037] Shared scale factor ϕ ij calculate:

[0038] ϕ ij ={∣F i ∩F j |}÷{(|F i ∣+∣F j |)÷2+ε};

[0039] Calculate the final constrained correlation degree W ij :

[0040] W ij =clip(W') ij ×max(ϕ ij ,η),0,1).

[0041] Among them, F i Represents the set of key fields for rule i (e.g., key fields {MAC, IP, legal representative}); |F i | represents the size of the collection (e.g., the number of fields); |F i ∩F j | represents the number of fields shared by rules i and j; W' ij Indicates the original correlation degree; ε represents a small constant (e.g., 10). −6 ), used to prevent division by zero; η represents the lower limit coefficient, used to prevent ϕ from pulling all edges to 0, the default value is 0.25; clip (•) means to clip W' ij ×max(ϕ ij η) is limited to between 0 and 1;

[0042] Step 1-3-3: Based on the calculation results of Step 1-3-2, obtain the associations between all rules and generate a rule network. The structure of the rule network includes nodes, edges, and edge attributes: a node represents the rule identifier (Rule-id); an edge represents the degree of association (W) between it and other rules. ij The edge attribute represents the field sharing ratio and the entity sharing ratio; the content of the association rule description is filled in with the content of the edge attribute;

[0043] Step 1-3-4: Send the rule network to the client to display to the user, allowing the user to make manual adjustments. Based on the adjustment results and the methods from Step 1-3-2 to Step 1-3-3, finally generate a fixed expected association table.

[0044] Preferably, when performing step 2, the data acquisition module specifically performs the following steps:

[0045] Step 2-1: The data acquisition module obtains risk control-related table data entered by the user from the client via the Internet, including hardware information, registration information, bidding and winning records, document data, bidding entrustment information, and related relationship data;

[0046] Step 2-2: After preprocessing the table data, the initial metadata is obtained. The preprocessing includes: handling missing values, standardizing field formats, deduplication, and mapping company names and personnel names to unique IDs, which are then used as the IDs of the initial metadata.

[0047] Steps 2-3: After extracting the data type, timestamp, and key fields from the initial metadata, the initial metadata record D is obtained. iThe initial metadata records are reorganized according to the format of the data digest template table G1 to obtain the metadata records Meta. i The specific formula is as follows:

[0048] Meta i =f(D i )={nodeid i ,type i timestamp i ,fields i anomaly i ,severity i description i};

[0049] Among them, Meta i D represents the i-th metadata record; i Indicates the i-th initial metadata record; nodeid i Indicates the ID of the i-th initial metadata; type i Indicates the data type in the i-th initial metadata; timestamp i Represents the timestamp of the collection of the i-th initial metadata; fields i Anomaly represents the set of key fields for the i-th initial metadata. i This represents the set of Rule-ids for the i-th initial metadata triggering rule; description i A text description representing the i-th initial metadata;

[0050] Steps 2-4: Combine the initial metadata with the metadata record. i Bind the metadata to obtain complete metadata and construct a metadata set, MetaSet.

[0051] The specific data structure of MetaSet is as follows:

[0052] MetaSet={a-data:data1, data2,..., data i ;a-Meta: Meta1, Meta2,..., Meta i};

[0053] Where i represents the index number of the i-th metadata record, Meta i It is the metadata record of the i-th metadata record, data i This represents the data entity of the i-th metadata record; both a-data and a-Meta are data segmentation markers.

[0054] Preferably, when executing step 3, the anomaly warning module specifically performs the following steps:

[0055] Step 3-1: Retrieve the metadata set MetaSet and compare any piece of metadata in the metadata set MetaSet with the bid rigging node rules of rule G0: If no exception is triggered, continue to compare the bidding behavior node rules; if an exception occurs, proceed to step 3-2.

[0056] If no abnormalities are found when comparing the bidding behavior node rules, continue to compare the awarding behavior warning node rules; if an abnormality is found, proceed to step 3-2.

[0057] When comparing the rules of the calibration behavior warning node, if an anomaly is found, proceed to step 3-2; if no anomaly is found, mark the metadata as normal.

[0058] Step 3-2: Use the expected association table to directly traverse all relevant association rules and compare them; update the metadata record Meta. i Mark an anomaly;

[0059] Step 3-3: Traverse all metadata records in the MetaSet i Meta records based on metadata i The exception markers are used to collect metadata for all exceptions and construct an exception metadata set, MetaSet. anomaly :

[0060] MetaSet anomaly ={n-data:ndata1,ndata2,...,ndata i ;n-Meta: nMeta1, nMeta2,..., nMeta i};

[0061] Wherein, n-data and n-Meta are both data segmentation markers in the abnormal metadata set, ndata i The data entity representing abnormal metadata; nMeta i Metadata records that represent abnormal metadata.

[0062] Preferably, step 3-2 includes the following steps:

[0063] Step 3-2-1: Retrieve the expected association table and find all association rules corresponding to the rule triggered by the current i-th metadata;

[0064] Step 3-2-2: Select any association rule and compare it with the current i-th metadata. Determine whether the current i-th metadata triggers the association rule: if yes, write the Rule-id of the association rule into the anomaly. i Update the description at the same time. i If the content is correct, then find the next association rule;

[0065] Step 3-2-3: Iterate through and compare all association rules according to the methods in Steps 3-2-1 and 3-2-2, and update the description accordingly. i The content.

[0066] Preferably, when performing step 4, the visualization module specifically performs the following steps:

[0067] Step 4-1: Retrieve the exception metadata set MetaSet anomaly For each piece of abnormal metadata, based on its nMeta i Determine its corresponding initial metadata record D i Establish a mapping relationship Map: Map(nMeta) i →D i );

[0068] Based on the mapping relationship Map, construct the exception mapping table as follows:

[0069] Horizontal dimension: Entities involved, i.e., affected companies;

[0070] Vertical dimension: rule items in rule G0;

[0071] Step 4-2: Traverse the exception metadata collection MetaSet anomaly The abnormal metadata is mapped to an abnormal mapping table and sent to the client for display.

[0072] The intelligent bidding risk control and early warning method described in this invention solves the technical problems of rules not being isolated, improving detection efficiency, and intuitively displaying abnormal results in bidding risk control systems. This invention achieves grid-like coupling between rules by calculating the correlation between them, thereby improving the systematicness and accuracy of risk control detection. By using the expected correlation table to compare only relevant rules, redundant matching is significantly reduced, and the early warning response speed is improved. Through the Map(nMetai→Di) mapping relationship, abnormal data is visualized into a two-dimensional table structure, which makes it easier for managers to intuitively determine the source and scope of risk. Attached Figure Description

[0073] Figure 1 This is the main flowchart of the present invention;

[0074] Figure 2 This is a flowchart illustrating the creation of the rule template and rule grid of this invention;

[0075] Figure 3 This is a flowchart of the data acquisition and metadata standardization process of this invention;

[0076] Figure 4 This is a flowchart of the anomaly detection and rule comparison process of the present invention;

[0077] Figure 5 This is a flowchart of the abnormal data mapping and visualization display of the present invention. Detailed Implementation

[0078] Depend on Figures 1-5 The intelligent bidding risk control and early warning method shown is characterized by the following steps:

[0079] Step 1: In the general template module, pre-define rule G0 and data summary template table G1, extract and construct the key field set and core entity set of each rule in rule G0, construct the rule grid engine RME1, and use the rule grid engine RME1 to calculate the initial correlation degree W' between any two rules. ij And by setting constraints, the final correlation degree W is obtained. ij This leads to the expected association table of the rules being obtained and output;

[0080] When executing step 1, the general template module performs the following steps:

[0081] Step 1-1: The user inputs rule G0 through the client, and the client sends rule G0 to the general template module for storage. Rule G0 includes rules for bid rigging and collusion nodes, bidding behavior nodes, and bid awarding behavior early warning nodes.

[0082] In this embodiment, the default preset rule G0 targets three nodes: bid rigging and collusion nodes, bidding behavior nodes, and bid awarding behavior early warning nodes.

[0083] The rules for bid rigging and collusion nodes include: abnormal signature information, abnormal registration information, abnormal file author information, failure to win bids 30 or more times within six months, and failure to cover the profits from winning bids, etc., which are the abnormal judgment rules for bid rigging and collusion nodes.

[0084] The rules for bidding behavior nodes include: rules for judging abnormalities at bidding behavior nodes such as the winning bid price exceeding the failed bid price by 5%, modification of project approval documents, modification of bidding details, and change of technical weight.

[0085] The rules for early warning nodes of bidding behavior include the following abnormal judgment rules: the winning bidder is not the lowest bidder in the initial round, the winning bidder is not the lowest bidder, the winning bid price is higher than the previous price, the winning bid price is higher than the historical lowest price, the winning bid price is more than 30% lower than the estimated price, the winning bid price is more than 20% higher than the estimated price, and the winning bid price is close to the estimated price ±1%.

[0086] Each rule in rule G0 includes the following attributes:

[0087] Rule-id: A unique identifier for a rule;

[0088] Rule-type: The type of the node to which it belongs;

[0089] Rule-fields i The set of key fields involved in rule triggering;

[0090] Rule-description: The description of the rule.

[0091] Rule G0 is written into the database of the general template module.

[0092] Steps 1-2: Construct a general template for the data summary template table G1, including unique ID, type enumeration, timestamp, key fields, anomaly markers, and text descriptions;

[0093] The general template of data summary template table G1 is for three nodes: bid rigging and collusion node, bidding behavior node, and bid award behavior early warning node. The general fields include unique ID, type enumeration, timestamp, key fields, and text description.

[0094] For each node, the differentiated item is marked as an anomaly, specifically:

[0095] The specific rule subset for bid-rigging nodes includes: signature information, registration information, file author information, and abnormal judgment rules for not winning bids for six months and winning bids not being covered by profits;

[0096] The specific rule subset for bidding behavior nodes includes: anomaly judgment rules for winning bids exceeding failed bids, modification of project approval documents, modification of bidding details, and changes in technical weights;

[0097] The subset of rules specific to the benchmarking behavior node includes: anomaly determination rules for various winning bid prices.

[0098] In this embodiment, the general template of the data summary template table G1 is shown in Table 1 below:

[0099]

[0100] Table 1

[0101] Steps 1-3: Construct the rule grid engine RME1, use the rule grid engine RME1 to parse rule G0, and calculate the key field sharing ratio, entity sharing ratio, and correlation degree W among the rules. ij Generate a list of expected associations for the rules, including the associated rule nodes and the association degree W. ij And description of association rules;

[0102] When performing steps 1-3, the construction and application of the rule grid engine RME1 specifically includes the following steps:

[0103] Step 1-3-1: Construct the regular mesh engine RME1, the specific model is as follows:

[0104] W' xy =α×FieldOverlap(x,y)+γ×EntityShare(x,y);

[0105] FieldOverlap(x, y) = {F x ∩F y}÷{F x ∪F y};

[0106] EntityShare(x, y) = {E x ∩E y}÷{E x ∪E y};

[0107] Where FieldOverlap is the proportion of key fields shared between rules; EntityShare is the proportion of entities shared between rules; W is the degree of relevance; F x F represents the set of key fields for rule x. y Represents the set of key fields for rule y (e.g., {MAC, IP, company_name}); E x E represents the set of entities that are expected to be associated with rule x or historically (e.g., a set of enterprise IDs or personnel IDs that may trigger rule x, such as company ID, legal representative ID, judge ID, etc.). y Represents the set of entities that rule y is expected to be associated with, or that it is historically associated with.

[0108] FieldOverlap takes values ​​of [0,1]. If rule x and rule y have no overlap in fields, then FieldOverlap=0; if the key field sets are exactly the same, then FieldOverlap=1. EntityShare takes values ​​of [0,1]. If the entities involved in rule x and rule y have no overlap, then EntityShare=0; if rule x and rule y have the same set of entities, then EntityShare=1.

[0109] α and γ are both adjustment parameters, taking values ​​α, γ∈[0,1], and α+γ=1;

[0110] In this embodiment, by default, in the initial stage, E is estimated using the "rule field candidate entity set". x or E y For example: the set of candidate entities strongly related to rule i, that is, the set of entities that may be marked when the rule is triggered; the set of entities for the rule "the same MAC is used by multiple companies" is a list of companies with duplicate MACs that have appeared before.

[0111] Step 1-3-2: Read rule G0, and using the rule grid engine RME1, calculate the original correlation degree W' between each rule in rule G0 and other rules. ij :

[0112] W' ij =α×FieldOverlap(i,j)+γ×EntityShare(i,j);

[0113] Where FieldOverlap(i,j) represents the proportion of key fields shared between rule i and rule j; EntityShare(i,j) represents the proportion of the same core entity involved in rule i and rule j, i.e., the entity sharing proportion.

[0114] Construct shared constraints for key fields:

[0115] Shared scale factor ϕ ij calculate:

[0116] ϕ ij ={∣F i ∩F j |}÷{(|F i ∣+∣F j |)÷2+ε};

[0117] Calculate the final constrained correlation degree W ij :

[0118] W ij =clip(W') ij ×max(ϕij ,η),0,1).

[0119] Among them, F i Represents the set of key fields for rule i (e.g., key fields {MAC, IP, legal representative}); |F i | represents the size of the collection (e.g., the number of fields); |F i ∩F j | represents the number of fields shared by rules i and j; W' ij Indicates the original correlation degree; ε represents a small constant (e.g., 10). −6 ), used to prevent division by zero; η represents the lower limit coefficient, used to prevent ϕ from pulling all edges to 0, the default value is 0.25; clip (•) means to clip W' ij ×max(ϕ ij η) is limited to between 0 and 1;

[0120] In this embodiment, the number of shared fields only accounts for a certain percentage (ϕ) relative to the field size of the two rules. ij Only when it is relatively large can the original W' be maintained. ij If only one field is shared occasionally while many other fields are shared, ϕ ij It will be very small, thus weakening W; max(ϕ) ij ,η) is a function that takes the maximum value; η only needs to be set once. Key field sharing constraints are used to reduce false alarm noise.

[0121] Step 1-3-3: Based on the calculation results of Step 1-3-2, obtain the associations between all rules and generate a rule network. The structure of the rule network includes nodes, edges, and edge attributes: a node represents the rule identifier (Rule-id); an edge represents the degree of association (W) between it and other rules. ij The edge attribute indicates the field sharing ratio and the entity sharing ratio.

[0122] The description of the association rule should be filled in with the content of the edge attribute;

[0123] Step 1-3-4: Send the rule network to the client to display to the user, allowing the user to make manual adjustments. Based on the adjustment results and the methods from Step 1-3-2 to Step 1-3-3, finally generate a fixed expected association table.

[0124] In this embodiment, manual adjustment includes adding, deleting, and editing rules in rule G0, adjusting the weight parameters of α and γ, and correcting the representative entity set of a certain rule.

[0125] In this embodiment, the rule network can be displayed to the user in the form of an edge list plus a list of shared fields and sample entities for each edge, so as to facilitate viewing and modification.

[0126] Steps 1-4: Output the fixed expected association table.

[0127] Step 2: The data acquisition module obtains the table data input by the user from the client, preprocesses the acquired table data to obtain initial metadata, extracts the data type, timestamp, and key fields from the initial metadata, and reorganizes it according to the field structure of the data summary template table G1 to obtain the standardized metadata record Meta. i The initial metadata is compared with the corresponding metadata record. i Bind the metadata to obtain complete metadata and construct a metadata set, MetaSet.

[0128] When performing step 2, the data acquisition module specifically performs the following steps:

[0129] Step 2-1: The data acquisition module obtains risk control-related table data entered by the user from the client via the Internet, including hardware information, registration information, bidding and winning records, document data, bidding entrustment information, and related relationship data;

[0130] In this embodiment, the table data specifically includes:

[0131] Hardware information includes the MAC address, motherboard number, IP address, and other hardware information of the bidding client reported by the bidding client.

[0132] Registration information includes company information such as business registration information, legal representative information, contact person information, email address, and bank account number;

[0133] Bidding and winning records include bidding information such as the number of bids submitted, bid opening results, bid document fees, and winning bid amounts;

[0134] The document data includes tender document author information, qualification document author information, text content, and other tender information;

[0135] The tendering entrustment form information includes entrustment information such as entrustment form information and tender announcement information;

[0136] The data on relationships includes information on personnel connections such as investments between companies, positions held, and cross-legal relationships between legal entities.

[0137] Step 2-2: After preprocessing the table data, the initial metadata is obtained. The preprocessing includes: handling missing values, standardizing field formats, deduplication, and mapping company names and personnel names to unique IDs, which are then used as the IDs of the initial metadata.

[0138] In this embodiment, missing value handling includes marking table data with missing critical fields as errors, and filling missing non-critical fields with default values ​​(such as empty strings or 0).

[0139] The standardized processing of field formats includes standardizing the company name and personnel name (such as removing spaces, unifying capitalization, and converting full-width and half-width characters), and standardizing the format of date, amount, and number fields.

[0140] Deduplication involves removing duplicate table data while retaining the latest timestamp.

[0141] Mapping to a unique ID involves mapping each enterprise to an individual and generating a unique ID.

[0142] Steps 2-3: After extracting the data type, timestamp, and key fields from the initial metadata, the initial metadata record D is obtained. i The initial metadata records are reorganized according to the format of the data digest template table G1 to obtain the metadata records Meta. i The specific formula is as follows:

[0143] Meta i =f(D i )={nodeid i ,type i timestamp i ,fields i anomaly i ,severity i description i};

[0144] Among them, Meta i D represents the i-th metadata record; i Indicates the i-th initial metadata record; nodeid i Indicates the ID of the i-th initial metadata; type i This indicates the data type in the i-th initial metadata; such as hardware information, registration information, bidding and winning records, file data, bidding entrustment information, and related relationship data; timestamp i Represents the timestamp of the collection of the i-th initial metadata; fields i Anomaly represents the set of key fields for the i-th initial metadata. i This represents the set of Rule-ids for the i-th initial metadata triggering rule; description i A text description representing the i-th initial metadata;

[0145] Steps 2-4: Combine the initial metadata with the metadata record. i Bind the metadata to obtain complete metadata and construct a metadata set, MetaSet.

[0146] The specific data structure of MetaSet is as follows:

[0147] MetaSet={a-data:data1, data2,..., data i ;a-Meta: Meta1, Meta2,..., Meta i};

[0148] Where i represents the index number of the i-th metadata record, Meta i It is the metadata record of the i-th metadata record, data i This represents the data entity of the i-th metadata record; both a-data and a-Meta are data segmentation markers.

[0149] In this embodiment, a table may include multiple table data. After preprocessing, each table data may be put into the database or data pool as initial metadata. The complete metadata is the structured data that binds the data entity to the data record.

[0150] Step 3: The anomaly warning module retrieves the metadata set MetaSet and compares any piece of metadata in it with each rule in rule G0. During the comparison, if an anomaly is detected in a certain node rule, the expected association table is called, and other rules associated with that rule are compared. Comparisons of unrelated rules are stopped, and the metadata record Meta is updated. i The exception marker generates an exception metadata record nMeta. i ;

[0151] Iterate through the MetaSet metadata collection. Once all metadata in the collection has undergone rule comparison, record the Meta data based on the metadata. i Extract exception metadata from the exception marker field and construct an exception metadata set MetaSet. anomaly ;

[0152] When executing step 3, the anomaly warning module performs the following steps:

[0153] Step 3-1: Retrieve the metadata set MetaSet and compare any piece of metadata in the metadata set MetaSet with the bid rigging node rules of rule G0: If no exception is triggered, continue to compare the bidding behavior node rules; if an exception occurs, proceed to step 3-2.

[0154] If no abnormalities are found when comparing the bidding behavior node rules, continue to compare the awarding behavior warning node rules; if an abnormality is found, proceed to step 3-2.

[0155] When comparing the rules of the calibration behavior warning node, if an anomaly is found, proceed to step 3-2; if no anomaly is found, mark the metadata as normal.

[0156] Step 3-2: Directly traverse all relevant association rules using the expected association table and compare them, instead of traversing the remaining rules one by one; update the metadata record Meta. i Mark an anomaly;

[0157] Step 3-2 includes the following steps:

[0158] Step 3-2-1: Retrieve the expected association table and find all association rules corresponding to the rule triggered by the current i-th metadata;

[0159] Step 3-2-2: Select any association rule and compare it with the current i-th metadata. Determine whether the current i-th metadata triggers the association rule: if yes, write the Rule-id of the association rule into the anomaly. i Update the description at the same time. i If the content is correct, then find the next association rule;

[0160] Step 3-2-3: Iterate through and compare all association rules according to the methods in Steps 3-2-1 and 3-2-2, and update the description accordingly. i The content.

[0161] When comparing metadata with the rules of any node in rule G0, the comparison is performed line by line according to the order of the rules in rule G0. Rule G0 is defined by the user. In this embodiment, if the user does not provide a new rule G0, the default rule G0 is used for line-by-line comparison. For example, the following rules are compared first: abnormal signature information, abnormal registration information, abnormal file author information, 30 or more unsuccessful bids within six months, and uncovered winning profits. The specific determination method is as follows:

[0162] Abnormal signature information:

[0163] If the same device fingerprint (MAC address, motherboard serial number, or bidding client fingerprint) is used by multiple different enterprise IDs, it will be triggered.

[0164] (Short-term abnormal changes in multiple devices for a single enterprise): If a single enterprise has recently used a large number of different devices to log in and the results are inconsistent with the registered address / legal representative, it will be marked as suspicious (optional).

[0165] Registration information error:

[0166] The system compares the "similarity" of key registration fields (text similarity calculation is based on existing technology and will not be described in detail). If the similarity exceeds a threshold, a trigger is activated. For example, when comparing the key registration field sets of two companies: {legal representative name, agent name, email, bank_account, address}, if the similarity reaches a preset threshold, such as 0.75, it is considered abnormal.

[0167] The file author information is incorrect.

[0168] An exception is triggered if two or more bid documents have the same or highly similar author identifiers and are submitted by different bidding entities.

[0169] Unsuccessful bids 30 or more times within six months:

[0170] The judgment result is obtained by querying and statistically analyzing historical bidding data.

[0171] Profits from the winning bid were not covered:

[0172] When the expected revenue from a supplier’s winning bid in one or more projects is insufficient to offset all expenses incurred during the bidding process (including bid document fees and winning bid service fees), an “uncovered winning bid profit” anomaly is triggered.

[0173] Secondly, the bidding process node rules are compared: the winning bid price exceeds the failed bid price by 5%, project initiation documents are modified, bidding details are modified, and technical weights are changed. The specific algorithms are as follows:

[0174] The winning bid exceeded the failed bid price by 5%.

[0175] The current winning bid price is compared with the lowest failed bid price of similar projects in history (or previous failed bids for this project). If the difference exceeds 5%, an anomaly is triggered.

[0176] Modify project approval documents, modify tender details, and change technical weights:

[0177] Compare the corresponding fields in the "Entrustment Form" with those in the "Tender Announcement" or "Final Tender Configuration" to see if they match: if they do not match, trigger the corresponding exception.

[0178] Finally, the rules for early warning nodes in the bid evaluation process are as follows: the winning bidder is not the lowest bidder in the initial round, the winning bidder is not the lowest bidder, the winning bid price is higher than the previous price, the winning bid price is higher than the historical lowest price, the winning bid price is more than 30% lower than the estimated price, the winning bid price is more than 20% higher than the estimated price, and the winning bid price is close to the estimated price ±1%. The specific algorithm is as follows:

[0179] The lowest bid from the winning bidder (excluding the initial round):

[0180] An anomaly is triggered if the final winning bid price of the winning bidder does not correspond to the bid price of that bidder in the initial round (i.e., the winning bid price is not the lowest price in the initial round).

[0181] The winning bidder is not the lowest bidder.

[0182] An anomaly is triggered if the final winning bid price is not the lowest price in the last round of bidding.

[0183] The winning bid price is higher than the previous price and the winning bid price is higher than the historical lowest price:

[0184] The current winning bid price will be compared with the previous winning bid price and the lowest winning bid price in history. If the current winning bid price is greater than either the previous winning bid price or the lowest winning bid price in history, an anomaly will be triggered.

[0185] The winning bid price is more than 30% lower than the estimated price, more than 20% higher than the estimated price, or close to the estimated price ±1%.

[0186] The winning bid price is compared with the estimated price entered by the user, and anomalies are triggered when the price is below 30%, above 20%, and close to ±1%.

[0187] Step 3-3: Traverse all metadata records in the MetaSet i Meta records based on metadata i The exception markers are used to collect metadata for all exceptions and construct an exception metadata set, MetaSet. anomaly :

[0188] MetaSet anomaly ={n-data:ndata1,ndata2,...,ndata i ;n-Meta: nMeta1, nMeta2,..., nMeta i};

[0189] Wherein, n-data and n-Meta are both data segmentation markers in the abnormal metadata set, ndata i The data entity representing abnormal metadata; nMeta i Metadata records that represent abnormal metadata.

[0190] Step 4: The visualization module retrieves the exception metadata set MetaSet. anomaly ;

[0191] Based on the metadata record nMeta for each abnormal metadata record i and its corresponding initial metadata record D i Establish mapping relationship Map(nMeta)i →D i Based on the mapping relationship Map, an exception mapping table is constructed, and the exception metadata collection MetaSet is added. anomaly Each piece of abnormal metadata is filled into the corresponding position in the abnormal mapping table according to the mapping relationship, and a complete abnormal mapping result table is generated. The generated abnormal mapping table is then sent to the client for visualization.

[0192] When performing step 4, the visualization module performs the following steps:

[0193] Step 4-1: Retrieve the exception metadata set MetaSet anomaly For each piece of abnormal metadata, based on its nMeta i Determine its corresponding initial metadata record D i Establish a mapping relationship Map: Map(nMeta) i →D i );

[0194] Based on the mapping relationship Map, construct the exception mapping table as follows:

[0195] Horizontal dimension: Entities involved, i.e., affected companies;

[0196] Vertical dimension: rule items in rule G0;

[0197] Step 4-2: Traverse the exception metadata collection MetaSet anomaly The abnormal metadata is mapped to an abnormal mapping table and sent to the client for display.

[0198] The intelligent bidding risk control and early warning method described in this invention solves the technical problems of rules not being isolated, improving detection efficiency, and intuitively displaying abnormal results in bidding risk control systems. This invention achieves grid-like coupling between rules by calculating the correlation between them, thereby improving the systematicness and accuracy of risk control detection. By using the expected correlation table to compare only relevant rules, redundant matching is significantly reduced, and the early warning response speed is improved. Through the Map(nMetai→Di) mapping relationship, abnormal data is visualized into a two-dimensional table structure, which makes it easier for managers to intuitively determine the source and scope of risk.

Claims

1. An intelligent bidding risk control and early warning method, characterized in that: Includes the following steps: Step 1: In the general template module, pre-define rule G0 and data summary template table G1, extract and construct the key field set and core entity set of each rule in rule G0, construct the rule grid engine RME1, and use the rule grid engine RME1 to calculate the initial correlation degree W' between any two rules. ij And by setting constraints, the final correlation degree W is obtained. ij This leads to the expected association table of the rules being obtained and output; The specific model of the regular mesh engine RME1 is as follows: W' xy =α×FieldOverlap(x,y)+γ×EntityShare(x,y); FieldOverlap(x,y)={F x ∩F y }÷{F x ∪F y }; EntityShare(x,y)={E x ∩E y }÷{E x ∪E y }; Where FieldOverlap is the proportion of key fields shared between rules; EntityShare is the proportion of entities shared between rules; W is the degree of association; F x F represents the set of key fields for rule x. y E represents the set of key fields for rule y; x E represents the set of entities that rule x is expected to or historically associated with. y Represents the set of entities that rule y is expected to be associated with, or that it is historically associated with. FieldOverlap takes values ​​of [0,1]. If rule x and rule y have no overlap in fields, then FieldOverlap=0; if the key field sets are exactly the same, then FieldOverlap=1. EntityShare takes values ​​of [0,1]. If the entities involved in rule x and rule y have no overlap, then EntityShare=0; if rule x and rule y have the same set of entities, then EntityShare=1. α and γ are both adjustment parameters, taking values ​​α, γ∈[0,1], and α+γ=1; Step 2: The data acquisition module obtains the table data input by the user from the client, preprocesses the acquired table data to obtain initial metadata, extracts the data type, timestamp, and key fields from the initial metadata, and reorganizes it according to the field structure of the data summary template table G1 to obtain the standardized metadata record Meta. i The initial metadata is compared with the corresponding metadata record. i Bind the metadata to obtain complete metadata and construct a metadata set, MetaSet. Step 3: The anomaly warning module retrieves the metadata set MetaSet and compares any piece of metadata in it with each rule in rule G0. During the comparison, if an anomaly is detected in a certain node rule, the expected association table is called, and other rules associated with that rule are compared. Comparisons of unrelated rules are stopped, and the metadata record Meta is updated. i The exception marker generates an exception metadata record nMeta. i ; Iterate through the MetaSet metadata collection. Once all metadata in the collection has undergone rule comparison, record the Meta data based on the metadata. i Extract exception metadata from the exception marker field and construct an exception metadata set MetaSet. anomaly ; When executing step 3, the anomaly warning module performs the following steps: Step 3-1: Retrieve the metadata set MetaSet and compare any piece of metadata in the metadata set MetaSet with the bid rigging node rules of rule G0: If no exception is triggered, continue to compare the bidding behavior node rules; if an exception occurs, proceed to step 3-2. If no abnormalities are found when comparing the bidding behavior node rules, continue to compare the awarding behavior warning node rules; if an abnormality is found, proceed to step 3-2. When comparing the rules of the calibration behavior warning node, if an anomaly is found, proceed to step 3-2; if no anomaly is found, mark the metadata as normal. Step 3-2: Use the expected association table to directly traverse all relevant association rules and compare them; update the metadata record Meta. i Mark an anomaly; Step 3-2 includes the following steps: Step 3-2-1: Retrieve the expected association table and find all association rules corresponding to the rule triggered by the current i-th metadata; Step 3-2-2: Select any association rule and compare it with the current i-th metadata. Determine whether the current i-th metadata triggers the association rule: if yes, write the Rule-id of the association rule into the anomaly. i At the same time, update the description. i If the content is correct, then find the next association rule; Step 3-2-3: Iterate through and compare all association rules according to the methods in Steps 3-2-1 and 3-2-2, and update the description accordingly. i Content; Step 3-3: Traverse all metadata records in the MetaSet i Meta records based on metadata i The exception markers are used to collect metadata for all exceptions and construct an exception metadata set, MetaSet. anomaly : MetaSet anomaly ={n-data:data1,data2,...,data i ;n-Meta:nMeta1,nMeta2,...,nMeta i }{ Wherein, n-data and n-Meta are both data segmentation markers in the abnormal metadata set, ndata i The data entity representing abnormal metadata; nMeta i Metadata records representing abnormal metadata; Step 4: The visualization module retrieves the exception metadata set MetaSet. anomaly ; Based on the metadata record nMeta for each abnormal metadata record i and its corresponding initial metadata record D i Establish mapping relationship Map(nMeta) i →D i Based on the mapping relationship Map, an exception mapping table is constructed, and the exception metadata collection MetaSet is added. anomaly Each piece of abnormal metadata is filled into the corresponding position in the abnormal mapping table according to the mapping relationship, and a complete abnormal mapping result table is generated. The generated abnormal mapping table is then sent to the client for visualization.

2. The intelligent bidding risk control and early warning method as described in claim 1, characterized in that: When executing step 1, the general template module performs the following steps: Step 1-1: The user inputs rule G0 through the client, and the client sends rule G0 to the general template module for storage. Rule G0 includes rules for bid rigging and collusion nodes, bidding behavior nodes, and bid awarding behavior early warning nodes. Steps 1-2: Construct a general template for the data summary template table G1, including unique ID, type enumeration, timestamp, key fields, anomaly markers, and text descriptions; The general template of data summary template table G1 is for three nodes: bid rigging and collusion node, bidding behavior node, and bid award behavior early warning node. The general fields include unique ID, type enumeration, timestamp, key fields, and text description. For each node, the differentiated item is marked as an anomaly, specifically: The specific rule subset for bid-rigging nodes includes: signature information, registration information, file author information, and abnormal judgment rules for not winning bids for six months and winning bids not being covered by profits; The specific rule subset for bidding behavior nodes includes: anomaly judgment rules for winning bids exceeding failed bids, modification of project approval documents, modification of bidding details, and changes in technical weights; The specific rule subset for the benchmarking behavior node includes: anomaly judgment rules for various winning bid prices; Steps 1-3: Construct the rule grid engine RME1, use the rule grid engine RME1 to parse rule G0, and calculate the key field sharing ratio, entity sharing ratio, and correlation degree W among the rules. ij Generate a list of expected associations for the rules, including the associated rule nodes and the association degree W. ij And description of association rules; Steps 1-4: Output the fixed expected association table.

3. The intelligent bidding risk control and early warning method as described in claim 2, characterized in that: When performing steps 1-3, the construction and application of the rule grid engine RME1 specifically includes the following steps: Step 1-3-1: Build the rule grid engine RME1; Step 1-3-2: Read rule G0, and using the rule grid engine RME1, calculate the original correlation degree W' between each rule in rule G0 and other rules. ij : W' ij =α×FieldOverlap(i,j)+γ×EntityShare(i,j); Where FieldOverlap(i,j) represents the proportion of key fields shared between rule i and rule j; EntityShare(i,j) represents the proportion of the same core entity involved in rule i and rule j, i.e., the entity sharing proportion. Construct shared constraints for key fields: Shared scale factor ϕ ij calculate: ϕ ij ={∣F i ∩F j ∣}÷{(∣F i ∣+∣F j ∣)÷2+ε]; Calculate the final constrained correlation degree W ij : W ij =clip(W' ij ×max(ϕ) ij ,h),0,1); Among them, F i Represents the set of key fields for rule i; |F i ∩F j | represents the number of fields shared by rules i and j; W' ij ε represents the original correlation degree; ε is a small constant used to prevent division by zero; η is a lower limit coefficient used to prevent ϕ from pulling all edges to 0, with a default value of 0.25; clip(•) indicates that W' ij ×max(ϕ ij η) is limited to between 0 and 1; Step 1-3-3: Based on the calculation results of Step 1-3-2, obtain the associations between all rules and generate a rule network. The structure of the rule network includes nodes, edges, and edge attributes: a node represents the rule identifier (Rule-id); an edge represents the degree of association (W) between it and other rules. ij The edge attribute represents the field sharing ratio and the entity sharing ratio; the content of the association rule description is filled in with the content of the edge attribute; Step 1-3-4: Send the rule network to the client to display to the user, allowing the user to make manual adjustments. Based on the adjustment results and the methods from Step 1-3-2 to Step 1-3-3, finally generate a fixed expected association table.

4. The intelligent bidding risk control and early warning method as described in claim 2, characterized in that: When performing step 2, the data acquisition module specifically performs the following steps: Step 2-1: The data acquisition module obtains risk control-related table data entered by the user from the client via the Internet, including hardware information, registration information, bidding and winning records, document data, bidding entrustment information, and related relationship data; Step 2-2: After preprocessing the table data, the initial metadata is obtained. The preprocessing includes: handling missing values, standardizing field formats, deduplication, and mapping company names and personnel names to unique IDs, which are then used as the IDs of the initial metadata. Steps 2-3: After extracting the data type, timestamp, and key fields from the initial metadata, the initial metadata record D is obtained. i The initial metadata records are reorganized according to the format of the data digest template table G1 to obtain the metadata records Meta. i The specific formula is as follows: Meta i =f(D i )={nodeid i ,type i ,timestamp i ,fields i ,anomaly i ,severity i ,description i }; Among them, Meta i D represents the i-th metadata record; i Indicates the i-th initial metadata record; nodeid i Indicates the ID of the i-th initial metadata; type i Indicates the data type in the i-th initial metadata; timestamp i Represents the timestamp of the collection of the i-th initial metadata; fields i Anomaly represents the set of key fields for the i-th initial metadata. i This represents the set of Rule-ids for the i-th initial metadata triggering rule; description i A text description representing the i-th initial metadata; Steps 2-4: Combine the initial metadata with the metadata record. i Bind the metadata to obtain complete metadata and construct a metadata set, MetaSet. The specific data structure of MetaSet is as follows: MetaSet={a-data:data1,data2,...,data i ;a-Meta:Meta1,Meta2,...,Meta i }; Where i represents the index number of the i-th metadata record, Meta i It is the metadata record of the i-th metadata record, data i This represents the data entity of the i-th metadata record; both a-data and a-Meta are data segmentation markers.

5. The intelligent bidding risk control and early warning method as described in claim 4, characterized in that: When performing step 4, the visualization module performs the following steps: Step 4-1: Retrieve the exception metadata set MetaSet anomaly For each piece of abnormal metadata, based on its nMeta i Determine its corresponding initial metadata record D i Establish a mapping relationship Map: Map(nMeta) i →D i ); Based on the mapping relationship Map, construct the exception mapping table as follows: Horizontal dimension: Entities involved, i.e., affected companies; Vertical dimension: rule items in rule G0; Step 4-2: Traverse the exception metadata collection MetaSet anomaly The abnormal metadata is mapped to an abnormal mapping table and sent to the client for display.

Images