Knowledge graph surrounding bidding risk identification method based on graph network and community discovery
By constructing an entity-relationship graph, the associations and feature vectors of bidders are obtained. Graph networks and community discovery methods are used for classification to identify risky objects, thus solving the problem of bid rigging and collusion in the bidding process and reducing regulatory costs and project performance risks.
Patent Information
- Application Number
- CN202511413251.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-29
- Publication Date
- 2026-01-20
AI Technical Summary
During the bidding process, collusion and bid-rigging occur, leading to unfairness in the bidding process and making it difficult to accurately identify risk information.
By constructing an entity-relationship graph, the relationships between bidding users and bidders are obtained, the feature vectors of bidders are extracted, and graph network and community discovery methods are used for classification, risk coefficient assessment, and identification of target risk objects.
It enables the identification of bid rigging, collusion, and shell alliances at a glance during the bid evaluation process, significantly reducing regulatory costs and project performance risks.
Smart Images
Figure CN121366028A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data processing, and particularly relates to a knowledge graph surrounding bid risk identification method based on a graph network and community discovery. BACKGROUND
[0002] Bidding refers to a process in which a tenderer selects an optimal supplier or contractor by publishing procurement information and inviting bidders to participate in competition in a market economy.
[0003] In the bidding or tendering process, both the bidding party and the tendering party need to perform the principle of fairness to give all potential bidders equal opportunities. However, at present, there are still many surrounding bid behaviors in the bidding process, and there are too many irregular interventions and collusion between bidders.
[0004] Therefore, it is currently urgent to accurately identify the risk information (for example, surrounding bid information) in the bidding information. SUMMARY
[0005] Therefore, it is necessary to provide a knowledge graph surrounding bid risk identification method based on a graph network and community discovery, which can accurately identify the risk information in the bidding information.
[0006] In a first aspect, the present application provides a knowledge graph surrounding bid risk identification method based on a graph network and community discovery, comprising:
[0007] Obtaining an entity-relation graph corresponding to a tendering user and at least two bidders; the entity-relation graph comprises an association relationship between the tendering user and each bidder;
[0008] According to the entity-relation graph and attribute information corresponding to each bidder, performing feature extraction on the bidders to obtain a feature vector of each bidder;
[0009] Classifying each bidder in the entity-relation graph to obtain category information corresponding to each bidder;
[0010] According to the feature vector of each bidder, evaluating a risk coefficient corresponding to each category, and determining a bidder in a category whose risk coefficient meets a preset requirement as a target risk object.
[0011] In one embodiment, the above-mentioned feature extraction on the bidders according to the entity-relation graph and the attribute information corresponding to each bidder to obtain the feature vector of each bidder comprises:
[0012] According to the entity-relation graph, the relationship of the bidder and a target user is expanded, to obtain an association relationship between the bidder and the target user; the target user includes a bidding user and other bidders except the bidder in the at least two bidders;
[0013] According to the association relationship between the bidder and the target user and attribute information of each bidder, feature extraction is performed on the bidder to obtain a feature vector of each bidder.
[0014] In one embodiment, the feature extraction on the bidder according to the association relationship between the bidder and the target user and the attribute information of each bidder to obtain the feature vector of each bidder includes:
[0015] According to the association relationship between the bidder and the target user, a plurality of adjacent bidders adjacent to the bidder are determined;
[0016] The attribute information of each adjacent bidder is aggregated to obtain an initial adjacent feature sum;
[0017] According to the attribute information of the bidder, an initial feature vector of the bidder is determined;
[0018] The initial adjacent feature sum and the initial feature vector of the bidder are input into a multi-layer perception machine for fusion to obtain the feature vector of the bidder.
[0019] In one embodiment, the classification of each bidder in the entity-relation graph to obtain the category information corresponding to each bidder includes:
[0020] For any bidder in the entity-relation graph, gain data between the bidder and a plurality of adjacent bidders is determined;
[0021] According to the gain data, the bidder and the adjacent bidder are merged into a community, the target bidder is determined as a new adjacent bidder, the community is determined as a new bidder, and the determination of the gain data between the bidder and the plurality of adjacent bidders is returned to be executed until each bidder in the entity-relation graph is traversed to obtain the category information corresponding to each bidder;
[0022] The target bidder refers to a bidder adjacent to the bidder or the adjacent bidder.
[0023] In one embodiment, the merging of the bidder and the adjacent bidder into a community according to the gain data includes:
[0024] If the gain data between the bidder and the adjacent bidder is the maximum data in a plurality of gain data, the bidder and the adjacent bidder are merged into a community;
[0025] If the gain data between the bidder and each adjacent bidder is negative, the bidder and each adjacent bidder are not merged into one community.
[0026] In one of the embodiments, the risk coefficients corresponding to each category are evaluated according to the feature vectors of the bidders, including:
[0027] For a category, the similarity between the feature vectors of the bidders in the category is determined.
[0028] For any bidder in the category, the weight between the bidder and an adjacent bidder is determined.
[0029] The first initial risk value of the bidder and the second initial risk value of each adjacent bidder are determined.
[0030] The risk coefficients corresponding to each category are determined according to the similarity, the weight, the first initial risk value and the second initial risk value.
[0031] In a second aspect, the present application further provides a knowledge graph bid risk identification device based on a graph network and community discovery, including:
[0032] An acquisition module is configured to acquire an entity-relation graph corresponding to a bidding user and at least two bidders; the entity-relation graph includes the association relationship between the bidding user and the bidders.
[0033] An extraction module is configured to perform feature extraction on the bidders according to the entity-relation graph and attribute information corresponding to the bidders, to obtain feature vectors of the bidders.
[0034] A classification module is configured to classify the bidders in the entity-relation graph, to obtain category information corresponding to the bidders.
[0035] A determination module is configured to evaluate risk coefficients corresponding to each category according to the feature vectors of the bidders, and to determine the bidders in the category whose risk coefficients meet preset requirements as target risk objects.
[0036] In a third aspect, the present application further provides a computer device including a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program:
[0037] An acquisition module is configured to acquire an entity-relation graph corresponding to a bidding user and at least two bidders; the entity-relation graph includes the association relationship between the bidding user and the bidders.
[0038] An extraction module is configured to perform feature extraction on the bidders according to the entity-relation graph and attribute information corresponding to the bidders, to obtain feature vectors of the bidders.
[0039] Classify each tenderer in the entity-relation graph to obtain category information corresponding to each tenderer;
[0040] According to the feature vector of each tenderer, a risk coefficient corresponding to each category is evaluated, and a tenderer in a category satisfying a preset requirement is determined as a target risk object.
[0041] In a fourth aspect, the present application further provides a computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the following steps:
[0042] Obtaining an entity-relation graph corresponding to a tendering user and at least two tenderers; the entity-relation graph includes an association relationship between the tendering user and each tenderer;
[0043] According to the entity-relation graph and attribute information corresponding to each tenderer, feature extraction is performed on the tenderer to obtain a feature vector of each tenderer;
[0044] Classifying each tenderer in the entity-relation graph to obtain category information corresponding to each tenderer;
[0045] According to the feature vector of each tenderer, a risk coefficient corresponding to each category is evaluated, and a tenderer in a category satisfying a preset requirement is determined as a target risk object.
[0046] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, the computer program being executed by a processor to implement the following steps:
[0047] Obtaining an entity-relation graph corresponding to a tendering user and at least two tenderers; the entity-relation graph includes an association relationship between the tendering user and each tenderer;
[0048] According to the entity-relation graph and attribute information corresponding to each tenderer, feature extraction is performed on the tenderer to obtain a feature vector of each tenderer;
[0049] Classifying each tenderer in the entity-relation graph to obtain category information corresponding to each tenderer;
[0050] According to the feature vector of each tenderer, a risk coefficient corresponding to each category is evaluated, and a tenderer in a category satisfying a preset requirement is determined as a target risk object.
[0051] The knowledge graph surrounding bid risk identification method based on the graph network and community discovery acquires an entity-relation graph of a bid invitation party and all bid invitation parties, combines background attributes of each bid invitation party to determine a characteristic vector of each bid invitation party, automatically groups the bid invitation parties according to the relationship distance, determines a risk value of each group by using the characteristic vector of each bid invitation party, and determines a target risk object according to the risk value of each group, so as to identify surrounding bid and serial bid, and empty shell alliance at a glance during bid evaluation, and significantly reduce the supervision cost and project performance risk. BRIEF DESCRIPTION OF DRAWINGS
[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other related drawings can be obtained by those skilled in the art without creative labor.
[0053] Figure 1 An application environment diagram of the knowledge graph surrounding bid risk identification method based on the graph network and community discovery in one embodiment;
[0054] Figure 2 A flowchart of the knowledge graph surrounding bid risk identification method based on the graph network and community discovery in one embodiment;
[0055] Figure 3 An acquisition flowchart of the entity-relation graph in one embodiment;
[0056] Figure 4 A flowchart of the knowledge graph surrounding bid risk identification method based on the graph network and community discovery in another embodiment;
[0057] Figure 5 A flowchart of the knowledge graph surrounding bid risk identification method based on the graph network and community discovery in another embodiment;
[0058] Figure 6 A flowchart of the knowledge graph surrounding bid risk identification method based on the graph network and community discovery in another embodiment;
[0059] Figure 7 A flowchart of the knowledge graph surrounding bid risk identification method based on the graph network and community discovery in another embodiment;
[0060] Figure 8 A flowchart of the knowledge graph surrounding bid risk identification method based on the graph network and community discovery in another embodiment;
[0061] Figure 9A flowchart of a knowledge graph surrounding risk identification method based on a graph network and community discovery in another embodiment is shown in the figure;
[0062] Figure 10 A structural block diagram of a knowledge graph surrounding risk identification device based on a graph network and community discovery in an embodiment is shown in the figure;
[0063] Figure 11 An internal structure diagram of a computer device in an embodiment is shown in the figure. DETAILED DESCRIPTION
[0064] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.
[0065] Tendering refers to a process in which a tenderer selects the best supplier or contractor by publishing procurement information and inviting bidders to participate in competition in a market economy.
[0066] In the tendering or bidding process, both the tenderer and the bidder need to perform the principle of fairness to give all potential bidders equal opportunities. However, at present, there are still many surrounding risk behaviors in the tendering process, such as too many irregular interventions by people, collusion between bidders, etc.
[0067] Therefore, it is an urgent problem to be solved to accurately identify the risk information (for example, surrounding risk information) in the tendering information. The present application provides a knowledge graph surrounding risk identification method based on a graph network and community discovery, which aims to solve the above problems.
[0068] After the background of the knowledge graph surrounding risk identification method based on a graph network and community discovery provided by the present application is introduced above, the implementation environment of the knowledge graph surrounding risk identification method based on a graph network and community discovery provided by the present application can also be introduced. The knowledge graph surrounding risk identification method based on a graph network and community discovery provided by the present embodiment can be applied to an implementation environment as shown in the figure. Figure 1 The implementation environment includes a server 104, which can be implemented by an independent server 104 or a server cluster composed of multiple servers 104, and a data storage system 102 that can store data required by the server 104 for processing. The data storage system 102 can be integrated on the server 104, or placed on a cloud or other network server. The server 104 can identify surrounding risk information from tendering information.
[0069] In other possible implementation manners, the knowledge graph surrounding risk identification method based on a graph network and community discovery provided in the embodiments of the present application can also be applied to a terminal. The terminal can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things device can be a smart speaker, a smart television, a smart air conditioner, a smart vehicle-mounted device, and the like. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, and the like.
[0070] In one embodiment, as shown in Figure 2 , a knowledge graph surrounding risk identification method based on a graph network and community discovery is provided. The method is applied to a server in Figure 1 , for example, and includes the following steps.
[0071] S201, acquiring an entity-relation graph corresponding to a bidding user and at least two bidders; the entity-relation graph includes the association relationship between the bidding user and each bidder.
[0072] In this embodiment, the construction of the entity-relation graph includes four steps of knowledge modeling, data acquisition, knowledge extraction, and graph storage, as shown in Figure 3 , wherein:
[0073] (1) Knowledge modeling: according to the content of the bid announcement and the relationship between the persons in the relevant procurement network, a plurality of entities and the relationship between each entity are determined; it should be noted that the entity-knowledge relation graph involved in this embodiment includes a total of 7 entities, including: core entities (suppliers, bidding projects) and extended entities (legal persons, senior managers, shareholders, proxy agencies, bid evaluation experts);
[0074] (2) Data acquisition: in this embodiment, the crawling method is used to acquire the procurement information involved in this embodiment from the relevant procurement network, and the procurement information is data cleaned to filter out the information in the procurement information that does not meet the requirements, and the procurement information after filtering out the information that does not meet the requirements is stored in the form of text for subsequent data labeling;
[0075] (3) Knowledge extraction: The embodiment mainly includes two parts for the step of knowledge extraction, one is entity extraction, and the other is relation extraction; wherein, the entity extraction is to utilize the BERT-BILSTM-Attention-CRF model, to obtain the word vector of the input text through BERT, and then according to the BILSTM which can capture the context correlation of the word sequence, the subsequent Attention mechanism is to process the feature sequence, aiming to prevent the semantic information loss that may occur at the front end of the long sequence, after a series of processing steps, the global feature representation is finally obtained, and the classification is performed by CRF classification label, CRF is an undirected graph model, which shows good effect in the task of sequence labeling; and the relation extraction is completed by using the BERT-TextCNN model. The model, first, the input sentence is input into BERT to obtain the corresponding feature vector, and the finally formed wordembeddings is input into TextCNN, second, into the TextCNN convolution layer, the TextCNN model generally uses multiple convolution kernels of different sizes, and different sizes of convolution kernels will result in different sizes of the feature set obtained, third, into the pooling layer, the pooling layer can reduce the output of the convolution layer and obtain deeper features, and finally, according to the output of the pooling layer and the number of classification categories, a full connection layer is constructed, and then a SoftMax is performed to obtain the final classification result;
[0076] (4) Graph construction: the entity and relation obtained by knowledge extraction are constructed into an entity-relation graph in the form of a triple, and the entity-relation graph is stored in a Neo4j graph database, and then visualized by the Neo4j graph database.
[0077] In the embodiment, for a bidding project, a bidding user and at least two bidders can be included, and according to the identification information of the bidding user and the identification information of the at least two bidders, the entity-relation graph corresponding to the bidding user and the at least two bidders is obtained from the entity-relation graph.
[0078] S202, according to the entity-relation graph and the attribute information corresponding to each bidder, the feature of the bidder is extracted to obtain the feature vector of each bidder.
[0079] In the embodiment, after obtaining the entity-relation graph corresponding to the bidding user and the at least two bidders, the attribute information corresponding to each bidder can be obtained, and based on the entity-relation graph and the attribute information corresponding to each bidder, the feature of the bidder is extracted to obtain the feature vector of each bidder.
[0080] Optionally, the attributes of the bidders (scale, qualification, credit, etc.) and the structured relationships (cooperation, competition, supply chain, etc.) of the first-order / second-order neighbors in the entity-relation graph are aggregated first, then the low-dimensional feature vectors are generated by using graph embedding and attention mechanism, the project context information is fused for dynamic weighting, and finally the feature vectors of the bidders containing both static strength and current scene adaptation degree are output.
[0081] In S203, the bidders in the entity-relation graph are classified to obtain the category information corresponding to each bidder.
[0082] In this embodiment, after obtaining the entity-relation graphs corresponding to the bidding user and the at least two bidders, the multiple bidders involved in the entity-relation graph can be classified according to the association relationships between the bidders in the entity-relation graph, to obtain the category information corresponding to each bidder.
[0083] Optionally, after obtaining the entity-relation graphs of the bidding user and the at least two bidders, the bidders are divided into several subgroups by using community discovery or graph clustering algorithm according to the association relationships such as stock ownership, cooperation, competition or supply chain between the bidders in the graph, to obtain the feature vectors of the bidders.
[0084] In S204, the risk coefficients corresponding to each category are evaluated according to the feature vectors of the bidders, and a target risk pair in a category whose risk coefficient meets a preset requirement is determined as a target risk object.
[0085] In this embodiment, after determining the feature vectors of the bidders and the category information corresponding to each bidder, the risk coefficients corresponding to each category can be evaluated according to the feature vectors of the bidders, and a target risk pair in a category whose risk coefficient meets a preset requirement is determined as a target risk object.
[0086] Optionally, after obtaining the feature vectors and category labels of the bidders, the feature vectors of all users in each category are pooled or weighted averaged to obtain a category-level risk representation, which is then input into a lightweight risk scoring model (such as single classification SVM or isolation forest) to output the comprehensive risk coefficient of the category; finally, the categories whose coefficients exceed a threshold are marked as high-risk clusters, and all entities in the clusters are listed as target risk objects for key monitoring. The target risk object is the surrounding string risk object.
[0087] Optionally, after determining the risk coefficients corresponding to each category, the bidder in the category with the largest risk coefficient can be determined as a target risk object.
[0088] In the embodiment, the entity-relation graph of the tenderer and all the bidders is obtained first, and then the feature vectors of the bidders are determined in combination with the background attributes of each bidder. Next, the bidders are automatically grouped according to the relationship distance, and the risk values of each group are determined using the feature vectors of the bidders, and the target risk object is determined according to the risk values of each group, so as to realize the one-eye identification of bid-rigging and shell alliance in bid evaluation, and significantly reduce the supervision cost and project performance risk.
[0089] In one embodiment, the detailed process of feature extraction of the bidders to obtain the feature vectors of the bidders is explained. In an exemplary embodiment, as shown in FIG. 2, S202 includes: Figure 4
[0090] S301, for any bidder in the entity-relation graph, the relationship of the bidder is expanded according to the entity-relation graph to obtain the association relationship between the bidder and the target user; the target user includes the tender user and other bidders except the bidder.
[0091] In the embodiment, for a project , the set of all bidders corresponding thereto can be identified as: , all the sets of bidders are merged to obtain: For all the bidders in the set S, one-degree or two-degree relationship expansion is performed in the entity-relation graph to obtain the association relationship between the bidder and the target user.
[0092] Among them, one-degree expansion obtains all entities (such as the legal person, the senior manager, the shareholder, etc. of the bidder) and relationships directly connected with the bidders in the set S, and two-degree expansion further obtains entities (such as other enterprises of the senior manager, other investment enterprises of the legal person, etc.) directly connected with the one-degree entities.
[0093] In the embodiment, after obtaining the association relationship between any bidder and the target user, the association relationship between each bidder and the target user in the entity-relation graph is obtained based on the same method.
[0094] S302, according to the association relationship between the bidder and the target user and the attribute information of each bidder, the feature of the bidder is extracted to obtain the feature vector of the bidder.
[0095] In the embodiment, the association relationship between each bidder and the target user is obtained, and the attribute information of each bidder is obtained, and the feature of each bidder is extracted based on the association relationship between each bidder and the target user and the attribute information of each bidder to obtain the feature vector of each bidder.
[0096] Optionally, a specific determination manner of feature extraction of the bidders according to the association relationship between the bidders and the target user and the attribute information of each bidder is provided as follows, i.e., S302 described above, comprising: Figure 5
[0097] S401, determining a plurality of adjacent bidders adjacent to the bidder according to the association relationship between the bidder and the target user.
[0098] In this embodiment, for any bidder, a plurality of adjacent bidders adjacent to the bidder are determined according to the association relationship between the bidder and the target user.
[0099] S402, performing aggregation processing on the attribute information of each adjacent bidder to obtain an initial adjacent feature sum.
[0100] In this embodiment, for any bidder in S401, the attribute information of each adjacent bidder adjacent to the bidder is determined, and the attribute information of each adjacent bidder is aggregated to obtain an initial adjacent feature sum.
[0101] Optionally, for any bidder, the attribute information of each first-order adjacent bidder corresponding to the bidder is collected, and then a sum aggregation algorithm is used to aggregate the attribute information of each adjacent bidder to obtain an initial adjacent feature sum. .
[0102] S403, determining an initial feature vector of the bidder according to the attribute information of the bidder.
[0103] In this embodiment, for any bidder in S401, the attribute information of the bidder can be weighted based on a preset weight to obtain an initial feature vector of the bidder.
[0104] Optionally, the attribute information of the bidder is multiplied by a preset weight to obtain an initial feature vector of the bidder, wherein w is a learnable parameter (default initial value is 0).
[0105] S404, inputting the initial adjacent feature sum and the initial feature vector of the bidder into a multi-layer perception machine for fusion to obtain a feature vector of the bidder.
[0106] In this embodiment, after determining the initial adjacent features and the initial vector of the bidder, the initial adjacent features and the initial feature vector of the bidder can be input into a multi-layer perception machine for fusion to obtain the feature vector of the bidder.
[0107] Optionally, according to the initial adjacent features and the initial feature vector of the bidder, the process of determining the feature vector of the bidder is determined, see formula (1) as follows:
[0108] (1) ;
[0109] wherein, denotes the bidder in the lth layer of attribute information, denotes the bidder adjacent to the bidder, denotes a learnable parameter for adjusting the importance of the node's own features, and + denotes SumAggregation, denotes a multi-layer perception machine for nonlinear transformation.
[0110] It should be noted that the training method of the multi-layer perception machine is also described in the present application: the multi-layer perception machine model is pre-trained on a large historical graph in an unsupervised manner (such as using node classification or link prediction as an auxiliary task), and then fine-tuned or directly used in the downstream task.
[0111] The input and output of the multi-layer perception machine model will be described in detail below:
[0112] Input layer: input , is the original attribute vector of the bidder i (such as the registered capital and establishment time of the supplier after standardization).
[0113] Output layer: output (l=3), that is, the output is the final embedding vector of the bidder .
[0114] In this embodiment, the entity-relation graph of the bidding party and all bidders is first obtained, and then the background attributes of each bidder are combined to determine the feature vector of each bidder. Then, the bidders are automatically grouped according to the relationship distance, the risk value of each group is determined using the feature vector of each bidder, and the target risk object is determined according to the risk value of each group, so as to realize one-eye identification of bid rigging, hollow alliance during bid evaluation, and significantly reduce the supervision cost and project performance risk.
[0115] In one embodiment, the detailed process of classifying each bidder in the entity-relation graph to obtain the category information corresponding to each bidder is explained. In one exemplary embodiment, as shown in FIG. 2A, S203 includes: Figure 6
[0116] S501, for any bidder in the entity-relation graph, determine the gain data between the bidder and the plurality of adjacent bidders.
[0117] In this embodiment, for any bidder in the entity-relation graph , the modularity gain when moving this bidder to each adjacent bidder adjacent thereto is calculated.
[0118] Next, the calculation formula of modularity is provided, see formula (2) below:
[0119] (2);
[0120] wherein m is the sum of the weights of all edges in the graph, denotes the weight (such as relationship confidence, number of cooperation times, etc. between the bidder and the adjacent bidder j, is a resolution parameter, which controls the community size, and is usually set to 1 by default, denotes the degree (sum of all edge weights) of bidder i, k j denotes the degree of adjacent bidder j, denotes 1 if the bidder and the adjacent bidder j belong to the same community, otherwise 0.
[0121] Next, the modularity gain of bidder moving to community is calculated, and the calculation formula of modularity gain is provided below, see formula (3) below:
[0122] (3);
[0123] wherein is the sum of the internal edge weights of community , is the sum of the edge weights between the node bidder and the nodes in community , is the sum of all edge weights of community (including external edges).
[0124] The gain data between the bidder and the plurality of adjacent bidders is calculated .
[0125] S502, according to the gain data, the bidder and the adjacent bidder are merged into a community, and the target bidder is determined as a new adjacent bidder, the community is determined as a new bidder, and the gain data between the bidder and the plurality of adjacent bidders is determined until each bidder in the entity-relation graph is traversed to obtain the category information corresponding to each bidder; wherein the target bidder refers to the bidder adjacent to the bidder or the adjacent bidder.
[0126] In this embodiment, after determining the gain data between the bidder and the plurality of adjacent bidders, the bidder and the adjacent bidder can be merged into a community according to the gain data, and the target bidder can be determined as a new adjacent bidder, the community can be determined as a new bidder, and the gain data between the bidder and the plurality of adjacent bidders can be determined until each bidder in the entity-relation graph is traversed to obtain the category information corresponding to each bidder.
[0127] Optionally, a specific implementation of merging the bidder and the adjacent bidder into a community according to the gain data is provided below, referring to Figure 7 The above S502 comprises:
[0128] S601, if the gain data between the bidder and the adjacent bidder is the maximum data in the plurality of gain data, the bidder and the adjacent bidder are merged into a community.
[0129] In this embodiment, after determining the gain data between the bidder and the plurality of adjacent bidders, if the gain data between the bidder and the adjacent bidder is the maximum data in the plurality of gain data, the bidder and the adjacent bidder are merged into a community.
[0130] S602, if the gain data between the bidder and each adjacent bidder is negative, the bidder and each adjacent bidder are not merged into a community.
[0131] In this embodiment, after determining the gain data between the bidder and the plurality of adjacent bidders, if the gain data between the bidder and each adjacent bidder is negative, the bidder and each adjacent bidder are not merged into a community.
[0132] In this embodiment, by grouping each bidder in the entity-relation graph, data basis is provided for subsequent determination of the target risk object.
[0133] In one embodiment, the detailed process of determining the risk coefficient corresponding to each category according to the feature vector of each bidder can be explained. In one exemplary embodiment, as shown in FIG. 8, S204 includes: Figure 8
[0134] S701, for a category, determining the similarity between the feature vectors of each bidder in the category.
[0135] In this embodiment, for each bidder in a category, the similarity between the feature vectors of each bidder in the category is determined.
[0136] Next, a determination method of the similarity between the feature vectors of each bidder is provided, as shown in the following formula (4):
[0137] (4);
[0138] wherein h i and h j are the feature vector of bidder i and the feature vector of bidder j respectively, and sim(h i , h j ) refers to the similarity between the feature vector of bidder i and the feature vector of bidder j, and the higher the value, the more similar the node attributes between bidder i and bidder j (e.g., the same name of a legal person means the similar vectors).
[0139] S702, for any bidder in the category, determining the weight between the bidder and the adjacent bidder.
[0140] In this embodiment, for any bidder in the category, the weight w ij between the bidder and the adjacent bidder is determined.
[0141] S703, determining the first initial risk value of the bidder and the second initial risk value of each adjacent bidder.
[0142] In this embodiment, the first initial risk value InitialRisk(i) of the bidder and the second initial risk value RiskScore (t) of each adjacent bidder are obtained.
[0143] S704, determining the risk coefficient corresponding to each category according to the similarity, the weight, the first initial risk value and the second initial risk value.
[0144] In this embodiment, after determining the similarity between the feature vectors of each bidder in the category, the weight between the bidder and the adjacent bidder, the first initial risk value of the bidder and the second initial risk value of each adjacent bidder, the risk coefficient corresponding to each category can be determined according to the similarity, the weight, the first initial risk value and the second initial risk value.
[0145] The following provides a determination manner of the risk coefficient corresponding to each category, see formula (5) - formula (6) below:
[0146] (5) ;
[0147] (6) ;
[0148] Wherein, RiskScore (t+1) (i) refers to the first intermediate risk value of the bidder, refers to the damping factor, which represents the probability that the node maintains its initial risk (usually ), refers to the propagation factor, , refers to the neighbor node set of the bidder i, refers to the weight between the bidder and the adjacent bidder, it should be noted that the stronger the relationship between the bidder and the adjacent bidder (such as the holding ratio, the number of cooperation times), the higher the weight, is to normalize all out-edge weights of the adjacent bidder j to avoid excessive risk propagation of high-degree nodes. Optionally, the damping factor retains the initial risk of the seed, and the propagation factor controls the propagation strength.
[0149] Wherein, formula (6) is the risk coefficient AvgRisk(C) corresponding to each category, and C is the category.
[0150] In this embodiment, the risk coefficient corresponding to each category is determined through the similarity, the weight, the first initial risk value and the second initial risk value, which provides a data basis for determining the bidder in the category as the target risk object.
[0151] In one embodiment, as Figure 9 shown, a knowledge graph risk identification method based on graph network and community discovery is also provided, comprising:
[0152] T1, obtaining an entity-relation graph corresponding to the bidding user and at least two bidders; the entity-relation graph includes the association relationship between the bidding user and each bidder;
[0153] T2, for any one bidder in the entity-relation graph, according to the entity-relation graph, the relationship of the bidder is expanded to obtain the association relationship between the bidder and the target user; the target user includes the bidding user and at least two bidders other than the bidder;
[0154] T3, determining a plurality of adjacent bidders adjacent to the bidder according to an association relationship between the bidder and a target user;
[0155] T4, performing aggregation processing on attribute information of each adjacent bidder to obtain an initial adjacent feature sum;
[0156] T5, determining an initial feature vector of the bidder according to attribute information of the bidder;
[0157] T6, inputting the initial adjacent feature sum and the initial feature vector of the bidder into a multi-layer perception machine for fusion to obtain a feature vector of the bidder;
[0158] T7, determining gain data between the bidder and a plurality of adjacent bidders for any bidder in the entity-relation graph;
[0159] T8, merging the bidder and the adjacent bidders into a community according to the gain data, determining a new adjacent bidder as the target bidder, determining a new bidder as the community, and returning to execute the determination of the gain data between the bidder and the adjacent bidders until each bidder in the entity-relation graph is traversed to obtain category information corresponding to each bidder; wherein the target bidder refers to a bidder adjacent to the bidder or the adjacent bidder;
[0160] T9, determining a similarity between feature vectors of each bidder in a category;
[0161] T10, determining a weight between the bidder and the adjacent bidder for any bidder in the category;
[0162] T11, determining a first initial risk value of the bidder and a second initial risk value of each adjacent bidder;
[0163] T12, determining a risk coefficient corresponding to each category according to the similarity, the weight, the first initial risk value and the second initial risk value;
[0164] T13, and determining the bidder in the category whose risk coefficient meets a preset requirement as a target risk object.
[0165] It should be noted that the description in T1-T13 above can refer to the description related in the above embodiments, and the effects are similar, and this embodiment will not be repeated here.
[0166] It should be understood that although the steps in the flowcharts involved in the embodiments described above are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be alternately executed with at least part of other steps or steps or stages in other steps.
[0167] Based on the same inventive concept, the embodiments of the present application also provide a graph network and community discovery based knowledge graph risk identification device for implementing the above-mentioned graph network and community discovery based knowledge graph risk identification method. The implementation scheme for solving problems provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more graph network and community discovery based knowledge graph risk identification device embodiments provided below can refer to the limitations of the graph network and community discovery based knowledge graph risk identification method described above, and will not be repeated here.
[0168] In one exemplary embodiment, as shown in Figure 10 A graph network and community discovery based knowledge graph risk identification device is provided, comprising: an acquisition module 10, an extraction module 11, a classification module 12, and a determination module 13, wherein:
[0169] The acquisition module 10 is configured to acquire an entity-relation graph corresponding to a bidding user and at least two bidders. The entity-relation graph includes the association relationship between the bidding user and each bidder.
[0170] The extraction module 11 is configured to perform feature extraction on the bidders according to the entity-relation graph and attribute information corresponding to each bidder, to obtain a feature vector of each bidder.
[0171] The classification module 12 is configured to classify each bidder in the entity-relation graph to obtain category information corresponding to each bidder.
[0172] The determination module 13 is configured to evaluate a risk coefficient corresponding to each category according to the feature vector of each bidder, and determine a bidder in a category whose risk coefficient meets a preset requirement as a target risk object.
[0173] In one exemplary embodiment, the above-mentioned extraction module 11 comprises:
[0174] The expansion unit is specifically configured to perform relationship expansion on the bidders according to the entity-relation graph, to obtain an association relationship between the bidders and a target user, the target user including the bidding user and other bidders except the bidders.
[0175] The extraction unit is specifically configured to perform feature extraction on the bidders according to the association relationship between the bidders and the target user and attribute information of the bidders, to obtain a feature vector of each bidder.
[0176] In an exemplary embodiment, the extraction unit is specifically further configured to determine a plurality of adjacent bidders adjacent to the bidder according to the association relationship between the bidder and the target user; aggregate attribute information of the adjacent bidders to obtain an initial adjacent feature sum; determine an initial feature vector of the bidder according to the attribute information of the bidder; and input the initial adjacent feature sum and the initial feature vector of the bidder into a multi-layer perception machine for fusion, to obtain the feature vector of the bidder.
[0177] In an exemplary embodiment, the classification module 12 includes:
[0178] The determination unit is specifically configured to determine gain data between a bidder and a plurality of adjacent bidders in an entity-relation graph.
[0179] The classification unit is specifically configured to merge the bidder and the adjacent bidders into one community according to the gain data, determine a target bidder as a new adjacent bidder, determine the community as a new bidder, and return to perform determination of the gain data between the bidder and the adjacent bidders until each bidder in the entity-relation graph is traversed to obtain category information corresponding to each bidder; wherein the target bidder refers to a bidder adjacent to the bidder or the adjacent bidder.
[0180] In an exemplary embodiment, the classification unit is specifically further configured to merge the bidder and the adjacent bidders into one community in a case where the gain data between the bidder and the adjacent bidders is maximum data in a plurality of gain data; and not merge the bidder and the adjacent bidders into one community in a case where the gain data between the bidder and each adjacent bidder is negative.
[0181] In an exemplary embodiment, the determination module 13 includes:
[0182] The first determination unit is specifically configured to determine a similarity between feature vectors of bidders in a category.
[0183] The second determining unit is specifically configured to determine a weight between the bidder and a neighboring bidder for any one of the categories;
[0184] The third determining unit is specifically configured to determine a first initial risk value of the bidder and second initial risk values of the neighboring bidders;
[0185] The fourth determining unit is specifically configured to determine a risk coefficient corresponding to each category according to the similarity, the weight, the first initial risk value, and the second initial risk values.
[0186] The above modules in the knowledge graph surrounding bid risk identification device based on a graph network and community discovery can be all or part realized by software, hardware, and combinations thereof. The above modules can be embedded in or independent of a processor in a computer device in a hardware form, or can be stored in a memory in a computer device in a software form, so as to call and execute the operations corresponding to the above modules by the processor.
[0187] In one exemplary embodiment, a computer device, which can be a server, has an internal structure diagram as shown in Figure 11 The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is configured to store attribute data corresponding to bidders. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through a network connection. The computer program is executed by the processor to implement a knowledge graph surrounding bid risk identification method based on a graph network and community discovery.
[0188] Those skilled in the art can understand that Figure 11 The structure shown in the above
[0189] In an embodiment, a computer device is also provided, comprising a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above method embodiments when executing the computer program.
[0190] In an embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program implementing the steps in the above method embodiments when executed by a processor.
[0191] In an embodiment, a computer program product is provided, comprising a computer program, and the computer program implementing the steps in the above method embodiments when executed by a processor.
[0192] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.
[0193] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.
[0194] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.
[0195] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.
Claims
1. A method for identifying risks of knowledge graph encirclement based on graph network and community discovery, characterized in that, The method comprises: obtaining an entity-relation graph corresponding to a bidding user and at least two bidders; the entity-relation graph comprises an association relationship between the bidding user and each bidder; performing feature extraction on the bidders according to the entity-relation graph and attribute information corresponding to each of the bidders to obtain a feature vector of each of the bidders; classifying each of the bidders in the entity-relation graph to obtain category information corresponding to each of the bidders; evaluating a risk coefficient corresponding to each category according to the feature vector of each of the bidders, and determining a bidder in a category satisfying a preset requirement as a target risk object.
2. The method of claim 1, wherein, The feature extraction on the bidders according to the entity-relation graph and the attribute information corresponding to each of the bidders to obtain a feature vector of each of the bidders comprises: for any one of the bidders in the entity-relation graph, performing relationship expansion on the bidder according to the entity-relation graph to obtain an association relationship between the bidder and a target user; the target user comprises the bidding user and other bidders except the bidder among the at least two bidders; performing feature extraction on the bidder according to the association relationship between the bidder and the target user and the attribute information of each of the bidders to obtain a feature vector of each of the bidders.
3. The method of claim 2, wherein, The feature extraction on the bidder according to the association relationship between the bidder and the target user and the attribute information of each of the bidders to obtain a feature vector of each of the bidders comprises: determining a plurality of adjacent bidders adjacent to the bidder according to the association relationship between the bidder and the target user; performing aggregation processing on the attribute information of each of the adjacent bidders to obtain an initial adjacent feature sum; determining an initial feature vector of the bidder according to the attribute information of the bidder; inputting the initial adjacent feature and the initial feature vector of the bidder into a multilayer perception machine for fusion to obtain the feature vector of the bidder.
4. The method of claim 1, wherein, The classification of each of the bidders in the entity-relation graph to obtain category information corresponding to each of the bidders comprises: for any one of the bidders in the entity-relation graph, determining gain data between the bidder and a plurality of adjacent bidders; according to the gain data, merging the bidder and the adjacent bidders into one community, determining a target bidder as a new adjacent bidder, determining the community as a new bidder, and returning to perform the determination of the gain data between the bidder and the plurality of adjacent bidders until each of the bidders in the entity-relation graph is completely traversed to obtain the category information corresponding to each of the bidders; wherein the target bidder refers to a bidder adjacent to the bidder or an adjacent bidder.
5. The method of claim 4, wherein, The merging of the bidder and the adjacent bidders into one community according to the gain data comprises: if the gain data between the bidder and the adjacent bidder is the maximum data in a plurality of gain data, then merging the bidder and the adjacent bidder into one community. If the gain data between the bidder and each adjacent bidder is negative, the bidder and each adjacent bidder are not merged into one community.
6. The method of claim 1, wherein, The risk coefficients corresponding to each category are evaluated according to the feature vectors of the bidders. For a category, the similarity between the feature vectors of the bidders in the category is determined. For any bidder in the category, the weight between the bidder and an adjacent bidder is determined. The first initial risk value of the bidder and the second initial risk values of the adjacent bidders are determined. The risk coefficients corresponding to each category are determined according to the similarity, the weight, the first initial risk value and the second initial risk values.
7. A knowledge graph risk identification device based on graph network and community discovery, characterized in that, The device comprises: An acquisition module configured to acquire an entity-relation graph corresponding to a bidding user and at least two bidders, the entity-relation graph comprising the association relationship between the bidding user and the bidders; An extraction module configured to perform feature extraction on the bidders according to the entity-relation graph and attribute information corresponding to the bidders, to obtain feature vectors of the bidders; A classification module configured to classify the bidders in the entity-relation graph, to obtain category information corresponding to the bidders; A determination module configured to evaluate risk coefficients corresponding to each category according to the feature vectors of the bidders, and to determine the bidders in a category whose risk coefficient meets a preset requirement as target risk objects.
8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 6.
9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.
10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.