Method, device, electronic device and medium for generating risk warning information
By building a multi-source heterogeneous enterprise association knowledge map, and identifying and flow prediction of risk structures, the accuracy and stability issues in enterprise risk warning are solved, and efficient risk detection and early warning are achieved.
Patent Information
- Application Number
- CN202410680361.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-29
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-05-29
AI Technical Summary
In the existing technology, in enterprise risk warning, the construction of a knowledge graph and single relational reasoning of a single data source leads to low accuracy, low accuracy of abnormal detection, and the classification of risk flow prediction of infectious disease models is not fine enough, resulting in early warning accuracy and system stability. Lower.
By obtaining the enterprise multi-source heterogeneous business data set, building the initial enterprise association knowledge graph, conducting link prediction and association relationship screening, identifying risk structures, generating node embedding feature vectors, performing abnormal risk detection and risk flow prediction, combining with the terminal display and Acoustic and optical warning equipment provides graded early warning.
It improves the accuracy of risk detection and early warning reliability, enhances the security and stability of the enterprise system, blocks risk flow in a timely manner, and reduces waste of communication resources.
Smart Images

Figure CN118446526B_ABST
Abstract
Description
Technical Field
[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to a method, device, electronic device, and medium for generating risk warning information. Background Art
[0002] In the current economic environment, the survival environment of enterprises is more difficult. Preventing abnormal risks caused by the relationship between enterprises has become one of the current development directions and has attracted more and more attention. For early warning of enterprise risks, the usual method is to obtain enterprise business data; secondly, construct a knowledge graph for enterprise business data to obtain an enterprise knowledge graph. Then, a tensor decomposition model is used to perform relational reasoning on the enterprise knowledge graph to obtain the inferred enterprise knowledge graph. Finally, a random walk graph embedding model is used to perform anomaly detection on the inferred enterprise knowledge graph to obtain an abnormal enterprise information set, and control the associated early warning equipment for early warning notification.
[0003] However, it has been found in practice that when the above method is used to warn of enterprise risks, the following technical problems often occur: Since the knowledge graph is constructed through data from a single data source and only relational reasoning is performed on the constructed knowledge graph, the deep embedded expressions between entities and entity relationships cannot be captured, resulting in low accuracy and narrow coverage of the generated enterprise knowledge graph; since the random walk graph embedding model is difficult to capture the multiple associations between different types of enterprise nodes, it can only extract local or global features, resulting in low accuracy of enterprise anomaly detection, and a large amount of erroneous and redundant abnormal enterprise information, resulting in missed detection and false detection of enterprise warnings, reducing the timeliness and reliability of enterprise warnings, and prolonging the processing time of warning information and the waste of warning information communication resources.
[0004] In the process of adopting technical solutions to solve the above technical problem 1, the following technical problem 2 is often accompanied: how to accurately and comprehensively detect abnormal risks on the risk enterprise associated network subgraph based on the node embedding feature vector set, so as to improve the efficiency of abnormal detection, shorten the duration of abnormal detection and reduce the waste of communication resources for sending warning information. For the above technical problem 2, the conventional solution is generally: extract the semantic features and spatial topological features of each node in the enterprise associated knowledge graph through a convolutional neural network for abnormal detection, obtain the enterprise risk information set, and send the enterprise risk information set and the preset warning information to each enterprise terminal for risk warning. However, the above conventional solution still has the following problems: because the semantic features and spatial topological features of the enterprise nodes of the entire enterprise associated knowledge graph are extracted based on the enterprise associated knowledge graph, the extracted features are relatively single and the amount of data for abnormal detection is large, and the convolutional neural network needs to be retrained when the enterprise information is updated, which requires a large amount of computing resources, resulting in low accuracy of risk abnormal detection and enterprise risk warning accuracy, resulting in low accuracy and processing efficiency of warning information, and waste of communication resources and system computing resources for sending warning information.
[0005] In the process of adopting technical solutions to solve the above technical problem one, the following technical problem three is often accompanied: how to accurately quantify and predict the risk flow of risky enterprises in the enterprise-related knowledge graph, and to accurately and timely warn related enterprises in advance and timely block the risk flow to improve the stability and security of each enterprise system. For the above technical problem three, the conventional solution is generally: using the infectious disease model to predict the risk flow of abnormal risk information sets, and obtain the enterprise information set to be circulated. However, the above conventional solution still has the following problems: because the infectious disease model divides the enterprise risk infection status into infection state, susceptible state and immune state, the division is not fine enough, and the consideration of the risk flow path is too single, resulting in low risk flow prediction accuracy and low warning efficiency and reliability. The flow of enterprise risks in various enterprise systems leads to low security and stability of enterprise systems, reducing the timeliness of warning information and wasting communication resources for sending warning information.
[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention
[0007] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.
[0008] Some embodiments of the present disclosure propose risk warning information generation methods, devices, electronic devices and media to solve one or more of the technical problems mentioned in the above background technology section.
[0009] In a first aspect, some embodiments of the present disclosure provide a method for generating risk warning information, comprising: in response to detecting business data query request information, obtaining a multi-source heterogeneous business data set of an enterprise from a distributed database; constructing an initial enterprise-related knowledge graph based on the above-mentioned multi-source heterogeneous business data set; performing link prediction processing on the above-mentioned initial enterprise-related knowledge graph to obtain an enterprise-related knowledge graph; performing association relationship screening processing on the above-mentioned enterprise-related knowledge graph to obtain a screened enterprise-related knowledge graph; performing risk structure identification on the above-mentioned screened enterprise-related knowledge graph to obtain a risk enterprise-related network subgraph set; performing adjacent node embedding representation on each enterprise node included in each risk enterprise-related network subgraph in the above-mentioned risk enterprise-related network subgraph set to generate a node embedding feature vector , obtain a node embedding feature vector set; based on the above node embedding feature vector set, perform abnormal risk detection on the above risk enterprise associated network subgraph set to obtain an enterprise risk information set; based on the above enterprise risk information set, perform enterprise risk flow prediction on the enterprise node set included in the above enterprise associated knowledge graph to obtain an enterprise information set to be transferred; use the associated terminal display screen to perform graded display for each enterprise information to be transferred in the above enterprise information set to be transferred, and control the sound and light warning device to perform graded warning through sounds of different frequencies and flashing lights of different colors that represent the warning level, so as to generate risk warning information for each enterprise information to be transferred in the above enterprise information set to be transferred, and send the generated risk warning information set to the enterprise terminal corresponding to the above enterprise information set to be transferred.
[0010] In a second aspect, some embodiments of the present disclosure provide a risk warning information generating device, comprising: an acquisition unit, configured to acquire an enterprise multi-source heterogeneous business data set from a distributed database in response to detecting business data query request information; a construction unit, configured to construct an initial enterprise association knowledge graph based on the above-mentioned enterprise multi-source heterogeneous business data set; a link prediction processing unit, configured to perform link prediction processing on the above-mentioned initial enterprise association knowledge graph to obtain an enterprise association knowledge graph; a screening unit, configured to perform association relationship screening processing on the above-mentioned enterprise association knowledge graph to obtain a screened enterprise association knowledge graph; a risk structure identification unit, configured to perform risk structure identification on the above-mentioned screened enterprise association knowledge graph to obtain a risk enterprise association network subgraph set; an adjacent node embedding representation unit, configured to perform adjacent node embedding on each enterprise node included in each risk enterprise association network subgraph in the above-mentioned risk enterprise association network subgraph set The input is represented to generate a node embedding feature vector to obtain a node embedding feature vector set; the abnormal risk detection unit is configured to perform abnormal risk detection on the risk enterprise associated network subgraph set according to the above node embedding feature vector set to obtain an enterprise risk information set; the risk flow prediction unit is configured to perform enterprise risk flow prediction on the enterprise node set included in the above enterprise associated knowledge graph according to the above enterprise risk information set to obtain an enterprise information set to be transferred; the level warning unit is configured to use the associated terminal display screen to perform graded display for each enterprise information to be transferred in the above enterprise information set to be transferred, and to control the sound and light warning device to perform level warning through sounds of different frequencies and flashes of different colors that represent the warning level, so as to generate risk warning information for each enterprise information to be transferred in the above enterprise information set to be transferred, and send the generated risk warning information set to the enterprise terminal corresponding to the above enterprise information set to be transferred.
[0011] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.
[0012] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method described in any implementation manner in the first aspect is implemented.
[0013] The above-mentioned embodiments of the present disclosure have the following beneficial effects: the risk warning information generation method of some embodiments of the present disclosure can improve the risk detection accuracy and warning reliability, improve the security and stability of the enterprise system, timely block the risk flow, and reduce the waste of communication resources. Specifically, the reasons for the omission and false detection of relevant enterprise warnings, the reduction of the timeliness and reliability of enterprise warnings, the extension of the warning information processing time and the waste of warning information communication resources are: because the random walk graph embedding model is difficult to capture the multi-association relationship between different types of enterprise nodes, it can only extract local or global features, resulting in low accuracy of enterprise anomaly detection, a large number of erroneous and redundant abnormal enterprise information, resulting in omission and false detection of enterprise warnings, reducing the timeliness and reliability of enterprise warnings, and extending the warning information processing time and the waste of warning information communication resources. Based on this, the risk warning information generation method of some embodiments of the present disclosure can first, in response to the detection of business data query request information, obtain a multi-source heterogeneous business data set of the enterprise from a distributed database. Here, the enterprise multi-source heterogeneous business data set is used for the subsequent construction of the enterprise knowledge graph. Secondly, according to the above-mentioned enterprise multi-source heterogeneous business data set, an initial enterprise association knowledge graph is constructed. Here, since the enterprise multi-source heterogeneous business data set comes from multiple heterogeneous data sources, the data sources are rich, which can improve the coverage of the enterprise-related knowledge graph. Again, the above-mentioned initial enterprise-related knowledge graph is subjected to link prediction processing to obtain the enterprise-related knowledge graph. Here, the accuracy and richness of the enterprise-related knowledge graph can be improved. Subsequently, the above-mentioned enterprise-related knowledge graph is subjected to association relationship screening processing to obtain the screened enterprise-related knowledge graph. Here, it is convenient to perform targeted anomaly detection for different associations in the future, so as to improve the accuracy of subsequent anomaly detection. Next, the risk structure of the above-mentioned screened enterprise-related knowledge graph is identified to obtain a risk enterprise-related network subgraph set. Here, the risk structure of the screened enterprise-related knowledge graph is targeted to improve the accuracy of identification. Afterwards, each enterprise node included in each risk enterprise-related network subgraph in the above-mentioned risk enterprise-related network subgraph set is represented by adjacent node embedding to generate a node embedding feature vector and obtain a node embedding feature vector set. Here, by combining the embedding of spatial features and semantic features of adjacent nodes, deeper and richer feature vectors of nodes can be extracted, improving the quality and accuracy of node embedding feature vectors. Then, based on the above node embedding feature vector set, abnormal risk detection is performed on the above risk enterprise association network subgraph set to obtain an enterprise risk information set. Here, abnormal risk prediction is performed through high-quality node embedding feature vectors to improve the accuracy of detection and obtain more accurate enterprise risk information. Afterwards, based on the above enterprise risk information set, enterprise risk flow prediction is performed on the enterprise node set included in the above enterprise association knowledge graph to obtain an enterprise information set to be circulated.Here, the risk transfer process and risk transfer scope of the enterprise corresponding to the enterprise risk information set can be determined, and more accurate enterprise information to be transferred can be obtained, so as to improve the efficiency of blocking risk transfer. Finally, the associated terminal display screen is used to display each enterprise information to be transferred in the above-mentioned enterprise information set to be transferred, and the sound and light warning device is controlled to perform level warnings through sounds of different frequencies and flashes of different colors that represent the warning level, so as to generate risk warning information for each enterprise information to be transferred in the above-mentioned enterprise information set to be transferred, and the generated risk warning information set is sent to the enterprise terminal corresponding to the above-mentioned enterprise information set to be transferred. Here, the enterprise security and warning accuracy can be improved. It can be obtained that the risk warning information generation method can improve the risk detection accuracy and warning reliability, improve the security and stability of the enterprise system, timely block the risk flow, and reduce the waste of communication resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.
[0015] Figure 1 is a flow chart of some embodiments of the risk warning information generation method according to the present disclosure;
[0016] Figure 2 is a schematic diagram of the structure of some embodiments of the risk warning information generating device according to the present disclosure;
[0017] Figure 3 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION
[0018] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.
[0019] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.
[0020] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.
[0021] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".
[0022] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.
[0023] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.
[0024] Figure 1 The process 100 of some embodiments of the risk warning information generation method according to the present disclosure is shown. The risk warning information generation method comprises the following steps:
[0025] Step 101 , in response to detecting business data query request information, obtaining enterprise multi-source heterogeneous business data sets from a distributed database.
[0026] In some embodiments, the execution subject (e.g., electronic device) of the above-mentioned risk warning information generation method can obtain the enterprise multi-source heterogeneous business data set from the distributed database through a wired connection method or a wireless connection method in response to detecting the business data query request information. Among them, the enterprise multi-source heterogeneous business data in the above-mentioned enterprise multi-source heterogeneous business data set can be data related to enterprise information from a data source. The above-mentioned enterprise multi-source heterogeneous business data set may include but is not limited to at least one of the following: enterprise name, enterprise shareholder information, enterprise investment information, enterprise transaction information, enterprise management personnel information, enterprise guarantee information, enterprise spouse information. The above-mentioned business data query request information may be request information for querying business data in a distributed database through a network protocol.
[0027] Step 102, construct an initial enterprise-related knowledge graph based on the enterprise's multi-source heterogeneous business data sets.
[0028] In some embodiments, the execution entity may construct an initial enterprise association knowledge graph based on the enterprise multi-source heterogeneous business data set. The initial enterprise association knowledge graph may be a graph structure data that displays the association relationship between enterprises in the form of an association network. The nodes in the initial enterprise association knowledge graph may be enterprise entities. The edges in the initial enterprise association knowledge graph may represent the association relationship between enterprises. For example, the association relationship may be, but is not limited to, at least one of the following: a transaction relationship, a guarantee relationship, an investment relationship, and a branch relationship.
[0029] As an example, the execution entity may first perform data fusion processing on the enterprise's multi-source heterogeneous business data set to obtain a fused business data set. Secondly, the entity association extraction model is used to extract entity relationships from the fused business data set to obtain an enterprise triple set. Among them, the entity association extraction model may be a model that extracts the enterprise entity set, entity attribute set, and entity association relationship from the input fused business data set and outputs them in the form of triples. The entity association extraction model may be a recurrent neural network model based on an attention mechanism. Then, the enterprise triple set is input into a graph database to obtain an initial enterprise association knowledge graph.
[0030] In some optional implementations of some embodiments, the construction of the initial enterprise-related knowledge graph based on the enterprise multi-source heterogeneous business data set may include the following steps:
[0031] The first step is to perform data fusion processing on the above-mentioned enterprise multi-source heterogeneous business data set to obtain a fused enterprise multi-source heterogeneous business data set. Among them, the fused enterprise multi-source heterogeneous business data in the above-mentioned fused enterprise multi-source heterogeneous business data set can be data of data structure and data type that fuses data from multiple data sources. In practice, the above-mentioned execution subject can first perform data preprocessing on the above-mentioned enterprise multi-source heterogeneous business data set to obtain a preprocessed multi-source heterogeneous business data set. Secondly, perform multi-dimensional heterogeneous data feature extraction on the above-mentioned preprocessed multi-source heterogeneous business data set to obtain a heterogeneous multi-dimensional tensor. Among them, the above-mentioned heterogeneous multi-dimensional tensor can represent the high-dimensional feature information of the preprocessed multi-source heterogeneous business data set. Then, the above-mentioned heterogeneous multi-dimensional tensor and each multi-dimensional feature vector of the above-mentioned heterogeneous multi-dimensional tensor are dimensionally multiplied to obtain a fused feature vector set. Finally, the above-mentioned fused feature vector set is input into the fully connected layer for feature decision to obtain a fused enterprise multi-source heterogeneous business data set.
[0032] The second step is to construct a business ontology data set based on the above-mentioned integrated enterprise multi-source heterogeneous business data set. The business ontology data in the above-mentioned business ontology data set can be one of the following: business entity class, business object attribute and business object association relationship. The business ontology data in the above-mentioned business ontology data set can be a formal description of enterprise entities, attribute concepts and associations, and a conceptual template for defining concepts and relationships between concepts in the knowledge graph. The above-mentioned business entity class can be the name of the enterprise. The above-mentioned business object attributes can be the attributes included in the enterprise. For example, the above-mentioned business object attributes can include but are not limited to at least one of the following: management personnel, shareholders, spouse, investment.
[0033] As an example, the execution subject may first use an ontology construction tool to construct an ontology for the merged enterprise multi-source heterogeneous business data set to obtain a business ontology data set. The ontology construction tool may be a protégé tool.
[0034] The third step is to perform the following mapping relationship determination steps for each business field in the business field set included in the above-mentioned integrated enterprise multi-source heterogeneous business data set:
[0035] Sub-step 1, determining the field similarity value between the business field and each business ontology data in the business ontology data set, and obtaining a field similarity value group. The business field may be an attribute field in a data table included in the fused enterprise multi-source heterogeneous business data set. The field similarity value may represent the character distance between the business field and the business ontology data.
[0036] Sub-step 2: Filter out at least one field similarity value greater than or equal to a preset similarity threshold from the field similarity value group as the target field similarity value, and obtain a target field similarity value group. The preset similarity threshold may be a predefined value. For example, the preset similarity threshold may be 0.8.
[0037] Sub-step 3, determining the business ontology data group corresponding to the target field similarity value group as the business ontology data group corresponding to the business field, and obtaining a field ontology mapping relationship group. The field ontology mapping relationship in the field ontology mapping relationship group can be a relationship of at least one business field corresponding to each business ontology data determined by the target field similarity value.
[0038] The fourth step is to extract the enterprise business data corresponding to each business ontology data in the above-mentioned business ontology data set from the above-mentioned enterprise multi-source heterogeneous business data set through the obtained field ontology mapping relationship group set, and obtain a business entity relationship triple set. Among them, the business entity relationship triples in the above-mentioned business entity relationship triple set can be triples composed of entities, entity relationships, and entities. Since the association relationships between business entities, between business entities and business object attributes, and between business object attributes have been determined in the second ontology construction step, it is only necessary to map the business fields in the fused enterprise multi-source heterogeneous business data set to the business ontology to obtain the business entity relationship triple set.
[0039] The fifth step is to perform instance deduplication processing on the above business entity relationship triple set to obtain a deduplicated business entity relationship triple set.
[0040] In the sixth step, the above-mentioned triplet set of business entity relationships after deduplication is input into a graph database to obtain an initial enterprise-related knowledge graph. The above-mentioned graph database may be a database that stores data in the form of a graph structure. For example, the above-mentioned graph database may be a Neo4j graph database.
[0041] Step 103, perform link prediction processing on the initial enterprise-related knowledge graph to obtain an enterprise-related knowledge graph.
[0042] In some embodiments, the execution subject may perform link prediction processing on the initial enterprise-related knowledge graph to obtain an enterprise-related knowledge graph. The enterprise-related knowledge graph may be a knowledge graph obtained by expanding the nodes and association relationships in the initial enterprise-related knowledge graph. In practice, the execution subject may input the initial enterprise-related knowledge graph into the graph attention network included in the knowledge graph reasoning model, and extract the entity semantic space topological features and association relationship features in the neighborhood of each enterprise entity by the feature embedding method to obtain a node neighborhood embedding feature vector set. Then, the node neighborhood embedding feature vector set is input into the first layer capsule network and the second layer capsule network included in the knowledge graph reasoning model to decode the triples corresponding to the initial enterprise-related knowledge graph and extract the hidden relationships between the triples to obtain the enterprise-related knowledge graph.
[0043] Step 104, performing association screening processing on the enterprise-related knowledge graph to obtain a screened enterprise-related knowledge graph.
[0044] In some embodiments, the execution subject may perform association screening on the enterprise association knowledge graph to obtain a screened enterprise association knowledge graph. The association screening process may be screening of guarantee relationships, shareholder relationships, investment relationships, transaction relationships, management relationships, and spouse relationships in the enterprise association knowledge graph. The screened enterprise association knowledge graph may be a knowledge graph including guarantee relationships, shareholder relationships, investment relationships, transaction relationships, management relationships, and spouse relationships.
[0045] Step 105, performing risk structure identification on the screened enterprise-related knowledge graph to obtain a risk enterprise-related network subgraph set.
[0046] In some embodiments, the execution subject may identify the risk structure of the enterprise-related knowledge graph after the screening to obtain a risk enterprise-related network subgraph set. Among them, the risk enterprise-related network subgraph in the risk enterprise-related network subgraph set may be a relationship network with a specific risk structure. The specific risk structure may be, but is not limited to, at least one of the following: a star structure in which the node out-degree of each node included in the risk enterprise-related network subgraph is greater than or equal to 4, a chain structure in which the length of the longest path included in the risk enterprise-related network subgraph is greater than or equal to 4, a mutual insurance structure in which the strong connection component corresponding to the risk enterprise-related network subgraph is 2, and a ring structure in which the strong connection component corresponding to the risk enterprise-related network subgraph is greater than or equal to 2.
[0047] In some optional implementations of some embodiments, the risk structure identification of the above-mentioned enterprise-related knowledge graph after screening to obtain the risk enterprise-related network subgraph set may include the following steps:
[0048] In the first step, the self-correlation edge removal process is performed on the above-mentioned enterprise-related knowledge graph after screening to obtain the enterprise-related knowledge graph after removal. The above-mentioned self-correlation edge removal process can be a process of removing the edges that associate the nodes with themselves in the enterprise-related knowledge graph after screening.
[0049] The second step is to perform sub-graph segmentation processing on the above-mentioned enterprise-related knowledge graph after removal to obtain an enterprise-related network sub-graph set. Among them, the enterprise-related network sub-graph in the above-mentioned enterprise-related network sub-graph set can be a sub-graph obtained by dividing nodes with high connectivity. In practice, the above-mentioned execution subject can use the Louvain algorithm to perform sub-graph segmentation processing on the above-mentioned enterprise-related knowledge graph after removal to obtain an enterprise-related network sub-graph set.
[0050] The third step is to classify the association relationships of the above-mentioned enterprise association network subgraph set to obtain an association relationship network subgraph group set. The above-mentioned association classification processing can be classified according to guarantee relationship, shareholder relationship, investment relationship, transaction relationship, executive relationship and spouse relationship. Each association relationship network subgraph group in the above-mentioned association relationship network subgraph group set can be a subgraph group corresponding to one of the above relationships.
[0051] The fourth step is to identify the risk structure of each association relationship network subgraph group in the above association relationship network subgraph group set to obtain a risk enterprise association network subgraph group. Among them, the risk enterprise association network subgraph in the above risk enterprise association network subgraph group may be a subgraph with a preset risk structure. The above preset risk structure may be a predefined risk structure. For example, the above preset risk structure may be but is not limited to at least one of the following: a star structure in which the out-degree of the nodes in the association relationship network subgraph corresponding to the guarantee relationship is greater than or equal to 4, and a ring structure in which the number of strongly connected components in the association relationship network subgraph corresponding to the guarantee relationship is greater than or equal to 2.
[0052] The fifth step is to determine the obtained venture enterprise association network subgraph group set as the venture enterprise association network subgraph set.
[0053] Step 106, performing adjacent node embedding representation on each enterprise node included in each venture enterprise association network subgraph in the venture enterprise association network subgraph set to generate a node embedding feature vector, thereby obtaining a node embedding feature vector set.
[0054] In some embodiments, the execution subject may perform adjacent node embedding representation on each enterprise node included in each venture enterprise association network subgraph in the venture enterprise association network subgraph set to generate a node embedding feature vector and obtain a node embedding feature vector set. The node embedding feature vector may characterize the semantic information, spatial information and association relationship information of the enterprise node and the adjacent nodes. In practice, the execution subject may use a graph neural network model to perform adjacent node embedding representation on each enterprise node included in each venture enterprise association network subgraph in the venture enterprise association network subgraph set to generate a node embedding feature vector and obtain a node embedding feature vector set.
[0055] In some optional implementations of some embodiments, the above-mentioned performing adjacent node embedding representation on each enterprise node included in each venture enterprise association network subgraph in the venture enterprise association network subgraph set to generate a node embedding feature vector to obtain a node embedding feature vector set may include the following steps:
[0056] In the first step, for each venture enterprise association network subgraph in the venture enterprise association network subgraph set, the following feature embedding steps are performed:
[0057] Sub-step 1: determine a venture enterprise node pair set corresponding to the venture enterprise node group included in the venture enterprise association network subgraph, wherein the venture enterprise node pair in the venture enterprise node pair set may be a node pair consisting of any two venture enterprise nodes having an association relationship and an association path connection.
[0058] Sub-step 2, determining the node meta-path group of each risk enterprise node pair in the risk enterprise node pair set, and obtaining a node meta-path group set. The node meta-path in the node meta-path group set may be a path that captures the associated semantic relationship between the risk enterprise node pairs. For example, the node meta-path of the first enterprise node-guarantee company-second enterprise node shows that the first enterprise node and the second enterprise node have a common semantic relationship of the guarantee company.
[0059] Sub-step 3: classify the node meta-path group set to obtain a category node meta-path group. The classification process may be performed according to different associated semantic relationships displayed by the node meta-path.
[0060] Sub-step 4: for each category node meta-path in the above category node meta-path group, perform the following feature vector determination step:
[0061] In the first sub-step, each category node included in the category node meta-path is encoded in node space to generate a node space feature vector and obtain a node space feature vector group. The node space feature vector can represent the spatial information of the category node. In practice, the execution subject can use a graph convolutional network to perform node space encoding on each category node included in the category node meta-path to generate a node space feature vector and obtain a node space feature vector group.
[0062] The second sub-step is to determine the starting spatial feature vector of the starting venture enterprise node and the ending spatial feature vector of the ending venture enterprise node according to the node spatial feature vector group. The starting venture enterprise node is the risk node at the starting position included in the risk enterprise node pair corresponding to the category node meta-path, and the ending venture enterprise node is the risk node at the ending position included in the risk enterprise node pair.
[0063] As an example, the execution subject may select the starting spatial feature vector of the starting venture enterprise node and the ending spatial feature vector of the ending venture enterprise node from the node spatial feature vector group.
[0064] The third sub-step is to perform node feature mapping processing on the above-mentioned starting space feature vector and the above-mentioned ending space feature vector to obtain a starting mapping feature vector and a ending mapping feature vector. The above-mentioned starting mapping feature vector and the above-mentioned ending mapping feature vector may be feature vectors mapped to the same feature space. The above-mentioned node feature mapping processing may be a mapping processing performed on the above-mentioned starting space feature vector and the above-mentioned ending space feature vector by a feature conversion matrix generated by the node types of the starting venture enterprise node and the ending venture enterprise node.
[0065] The fourth sub-step is to determine the node pair weight feature vector of the above-mentioned starting mapping feature vector and the above-mentioned ending mapping feature vector under the above-mentioned category node meta-path. The above-mentioned node pair weight feature vector may be a feature vector that assigns different weights to the above-mentioned starting mapping feature vector and the above-mentioned ending mapping feature vector. In practice, the above-mentioned execution subject may input the above-mentioned starting mapping feature vector and the above-mentioned ending mapping feature vector into the attention mechanism network to obtain the node pair weight feature vector.
[0066] The fifth sub-step is to normalize the node pair weight feature vector to obtain a normalized node pair weight feature vector. The normalization process may be a normalization process performed by a softmax function.
[0067] The sixth sub-step is to determine the product of the normalized node pair weight feature vector and the termination space feature vector as the node path feature vector of the starting venture enterprise node under the category node meta-path.
[0068] Sub-step 5: Determine the adjacent venture enterprise node group of the starting venture enterprise node from the venture enterprise node group corresponding to the venture enterprise node pair set. The adjacent venture enterprise nodes in the adjacent venture enterprise node group may be venture enterprise nodes that have a layer of connection relationship with the starting venture enterprise node.
[0069] Sub-step 6: performing feature concatenation processing on the node path feature vector group corresponding to the adjacent risk enterprise node group to obtain a spatial embedding feature vector.
[0070] Sub-step 7, determining the semantic embedding feature vector of the above-mentioned starting venture enterprise node under the category node meta-path group. The above-mentioned semantic embedding feature vector can represent the semantic information of the starting venture enterprise node under the category node meta-path. In practice, the above-mentioned execution subject can use graph representation learning to determine the semantic embedding feature vector of the above-mentioned starting venture enterprise node under the category node meta-path group.
[0071] The second step is to generate a node embedding feature vector set according to the obtained spatial embedding feature vector set and the obtained semantic embedding feature vector set. The node embedding feature vector in the above node embedding feature vector set may be a feature vector that integrates the semantic information and spatial information of the node under the meta-path.
[0072] As an example, the execution subject may perform feature fusion on the obtained spatial embedding feature vector set and the obtained semantic embedding feature vector set to obtain a node embedding feature vector set.
[0073] In some optional implementations of some embodiments, the generating of the node embedding feature vector set according to the obtained spatial embedding feature vector set and the obtained semantic embedding feature vector set may include the following steps:
[0074] In the first step, for each spatial embedding feature vector in the above spatial embedding feature vector set, the following embedding feature vector determination steps are performed:
[0075] Sub-step 1: Perform nonlinear transformation processing on the spatial embedding feature vector to obtain a transformed spatial embedding feature vector. The nonlinear transformation processing may be a nonlinear transformation processing performed using a convolutional network including a Tanh activation function.
[0076] Sub-step 2, performing node semantic encoding on the starting risk enterprise node corresponding to the above-mentioned converted spatial embedding feature vector to obtain a risk node semantic feature vector. The above-mentioned risk node semantic feature vector can represent the semantic information of the starting risk enterprise node. In practice, the above-mentioned execution subject can use the Word2Vec algorithm to perform node semantic encoding on the starting risk enterprise node corresponding to the above-mentioned converted spatial embedding feature vector to obtain a risk node semantic feature vector.
[0077] Sub-step 3: determining the path semantic feature vector as the product of the semantic embedding feature vector corresponding to the spatial embedding feature vector and the converted spatial embedding feature vector.
[0078] Sub-step 4: normalize the path semantic feature vector to obtain a normalized path semantic feature vector. The normalization process may be a normalization process using a Softmax function.
[0079] Sub-step 5: determining the product of the normalized path semantic feature vector and the spatial embedding feature vector as the node embedding feature vector.
[0080] Step 107, performing abnormal risk detection on the risk enterprise associated network subgraph set according to the node embedding feature vector set, and obtaining the enterprise risk information set.
[0081] In some embodiments, the execution subject may perform abnormal risk detection on the risk enterprise associated network subgraph set according to the node embedding feature vector set to obtain an enterprise risk information set. The enterprise risk information in the enterprise risk information set may be information on enterprises with risk anomalies whose risk probability values are greater than or equal to a preset risk probability threshold. The preset risk probability threshold may be a pre-set value. For example, the preset risk probability threshold may be 0.75.
[0082] As an example, the execution subject may use an anomaly detection model to perform anomaly risk detection on the risk enterprise association network subgraph set according to the node embedding feature vector set to obtain an enterprise risk information set. The anomaly detection model may be Time-LLM (Time Series Forecasting by Reprogramming Large Language Models, a time series prediction model based on a large language model).
[0083] In some optional implementations of some embodiments, the above-mentioned abnormal risk detection is performed on the above-mentioned risk enterprise association network subgraph set according to the above-mentioned node embedding feature vector set to obtain the enterprise risk information set, including:
[0084] In the first step, for each venture enterprise association network subgraph in the venture enterprise association network subgraph set, the following abnormal risk detection steps are performed:
[0085] Sub-step 1, performing adjacent node sampling processing on each risk node included in the above-mentioned risk enterprise association network subgraph to obtain an adjacent inductive subgraph set. Among them, the adjacent inductive subgraph in the above-mentioned adjacent inductive subgraph set can be a subgraph obtained by a random sampling method with replacement, with each risk node as the sampling center, and the shortest path to the risk node is 1. The above-mentioned adjacent node sampling processing can be sampling performed using a random sampling algorithm with replacement.
[0086] Sub-step 2, performing adjacency information aggregation processing on each adjacency induction node included in each adjacency induction subgraph in the above-mentioned adjacency induction subgraph set to generate a node aggregation feature vector and obtain a node aggregation feature vector group set. Among them, the above-mentioned node aggregation feature vector can characterize the adjacency association information of each node in the adjacency induction subgraph. In practice, the above-mentioned execution entity can use GCN (Graph Convolutional Networks) to perform adjacency information aggregation processing on each adjacency induction node included in each adjacency induction subgraph in the above-mentioned adjacency induction subgraph set to generate a node aggregation feature vector and obtain a node aggregation feature vector group set.
[0087] Sub-step 3, determining the node embedding feature vector corresponding to each node aggregation feature vector in the above node aggregation feature vector set as the target node embedding feature vector, and obtaining the target node embedding feature vector set. The target node embedding feature vector may be a feature vector obtained by mapping the node aggregation feature vector to a low-dimensional vector space. In practice, the execution subject may input the node aggregation feature vector set into a node embedding layer to obtain the target node embedding feature vector set.
[0088] Sub-step 4: performing feature fusion processing on the target node embedding feature vector set and the node aggregation feature vector set to obtain a fused node feature vector set.
[0089] Sub-step 5, performing edge-type multi-layer convolution processing on each fused node feature vector in the fused node feature vector set to obtain a node convolution feature vector set. The node convolution feature vectors in the node convolution feature vector set can represent feature information of the venture enterprise association network subgraph.
[0090] Sub-step 6, input the node convolution feature vector set into the abnormal classification prediction layer to obtain the enterprise abnormal risk probability value set. Among them, the enterprise abnormal risk probability value in the enterprise abnormal risk probability value set can represent the probability value of the corresponding enterprise risk. The larger the enterprise abnormal risk probability value, the greater the possibility that the enterprise node has a risk.
[0091] The second step is to determine at least one enterprise abnormal risk probability value corresponding to each node embedding feature vector in the obtained abnormal risk probability value set of each enterprise, and obtain the abnormal probability value set of each category.
[0092] In the third step, for each category abnormal probability value group in the above-mentioned category abnormal probability value group set, the mean of at least one enterprise abnormal risk probability value corresponding to the node embedded feature vector corresponding to the above-mentioned category abnormal probability value group is determined as the enterprise risk probability value of the node embedded feature vector as the enterprise risk information, and the generated enterprise warning information is sent to the enterprise terminal corresponding to the above-mentioned node embedded feature vector.
[0093] The above technical solution and its related contents, combined with step "step 109", as an inventive point of an embodiment of the present disclosure, solves the second technical problem mentioned in the background technology: "Since the semantic features and spatial topological features of the enterprise nodes of the entire enterprise-related knowledge graph are extracted based on the enterprise-related knowledge graph, the extracted features are relatively simple and the amount of data for anomaly detection is large, and the convolutional neural network needs to be retrained when the enterprise information is updated, which requires a large amount of computing resources, resulting in low accuracy of risk anomaly detection and enterprise risk warning accuracy, low accuracy and processing efficiency of warning information, and waste of communication resources and system computing resources for sending warning information." The factors that lead to low accuracy and processing efficiency of warning information and waste of communication resources and system computing resources for sending warning information are often as follows: Since the semantic features and spatial topological features of the enterprise nodes of the entire enterprise-related knowledge graph are extracted based on the enterprise-related knowledge graph, the extracted features are relatively simple and the amount of data for anomaly detection is large, and the convolutional neural network needs to be retrained when the enterprise information is updated, which requires a large amount of computing resources, resulting in low accuracy of risk anomaly detection and enterprise risk warning accuracy. If the above factors are solved, the accuracy and processing efficiency of early warning information can be improved, and the waste of communication resources and system computing resources for sending early warning information can be reduced. In order to achieve this effect, the present disclosure first performs adjacent node sampling processing on each risk node, and performs adjacent information aggregation processing on the generated adjacent inductive subgraph set, which can reduce the redundant nodes of the formed adjacent inductive subgraph set, avoid frequent sampling of nodes, reduce the amount of data and the data quality of the adjacent inductive subgraph set, and reduce the overall update when the enterprise node is updated, reducing the waste of computing resources. Then, node embedding and feature fusion are performed on the node aggregation feature vector set, and the semantic information, spatial topology information and enterprise association hierarchical relationship information of the enterprise node are integrated into the node aggregation feature vector set, which can improve the accuracy and richness of the feature vector. Finally, the fused node feature vector set is subjected to multi-layer convolution and classification prediction of the connection edge type, which can reduce the complexity of the anomaly detection model, reduce the waste of computing resources, and improve the detection efficiency and anomaly detection accuracy. Finally, the enterprise abnormal risk probability value sets are classified and averaged, and the generated enterprise warning information is sent to the enterprise terminal. By improving the accuracy of anomaly detection, the waste of communication resources can be reduced.
[0094] Step 108, based on the enterprise risk information set, predict the enterprise risk flow of the enterprise node set included in the above-mentioned enterprise-related knowledge graph to obtain the enterprise information set to be circulated.
[0095] In some embodiments, the execution subject may perform enterprise risk transfer prediction on the enterprise node set included in the enterprise-related knowledge graph according to the enterprise risk information set to obtain an enterprise information set to be transferred. The enterprise information to be transferred in the enterprise information set to be transferred may be information of enterprises infected by risks.
[0096] As an example, the execution entity may utilize a virus propagation model to predict the enterprise risk flow of the enterprise node set included in the enterprise-related knowledge graph according to the enterprise risk information set, and obtain an enterprise information set to be circulated.
[0097] Considering the problems of the above conventional solutions, facing the above technical problem three: because the infectious disease model divides the enterprise risk infection status into infection status, susceptible status and immune status, the division is not fine enough, and the consideration of the risk flow path is too single, resulting in low risk flow prediction accuracy of the circulating enterprise, high enterprise loss risk and low early warning accuracy. Combined with the advantages / technical status, it can be decided to adopt the following solution.
[0098] In some optional implementations of some embodiments, the enterprise risk transfer prediction is performed on the enterprise node set included in the enterprise-related knowledge graph according to the enterprise risk information set to obtain the enterprise information set to be transferred, which may include the following steps:
[0099] The first step is to obtain the historical value transfer information of each enterprise node in the enterprise node set corresponding to the enterprise risk information set to obtain the historical value transfer information set. The historical value transfer information can be the financial income information of the enterprise.
[0100] In the second step, according to the above-mentioned historical value flow information set and the above-mentioned enterprise risk information set, the enterprise value flow status information of each enterprise node in the enterprise node set included in the above-mentioned enterprise association knowledge graph is determined to obtain the enterprise value flow status information set. Among them, the above-mentioned enterprise value flow status information can represent the enterprise's operating ability. The above-mentioned enterprise value flow status information can include five states: complete value flow status information, high-level value flow status information, medium-level value flow status information, low-level value flow status information and risk value flow status information. The above-mentioned complete value flow status information can represent the enterprise's 100% operating ability. The above-mentioned high-level value flow status information can represent that the enterprise's operating ability is greater than or equal to 75% and less than 100%. The above-mentioned medium-level value flow status information can represent that the enterprise's operating ability is greater than or equal to 50% and less than 75%. The above-mentioned low-level value flow status information can represent that the enterprise's operating ability is greater than or equal to 25% and less than 50%. The above-mentioned risk value flow status information can represent that the enterprise's operating ability is less than 25%.
[0101] As an example, the above-mentioned execution entity can use the hierarchical analysis method to quantify the above-mentioned historical value flow information set and the above-mentioned enterprise risk information set to determine the enterprise value flow status information of each enterprise node in the enterprise node set included in the above-mentioned enterprise association knowledge graph, and obtain the enterprise value flow status information set.
[0102] The third step is to generate the enterprise risk response capability value and enterprise risk tolerance value of each enterprise node in the enterprise node set included in the above-mentioned enterprise association knowledge graph according to the above-mentioned enterprise value flow status information set and the above-mentioned enterprise risk information set, and obtain the enterprise risk response capability value set and the enterprise risk tolerance value set. Among them, the enterprise risk response capability value in the above-mentioned enterprise risk response capability value set can represent the probability value of the enterprise recovering to its original state or improving the enterprise's operating capacity after suffering an operational risk. The larger the above-mentioned enterprise risk response capability value, the stronger the enterprise's ability to cope with operational risks. The enterprise risk tolerance value in the above-mentioned enterprise risk tolerance value set can represent the probability value of the enterprise suffering from operational risks resulting in a lower enterprise operating capacity. The smaller the above-mentioned enterprise risk tolerance value, the lower the possibility that the enterprise will suffer from operational risks resulting in a lower enterprise operating capacity.
[0103] As an example, the execution subject may input the enterprise value flow state information set and the enterprise risk information set into a dynamic Bayesian network to obtain an enterprise risk response capability value set and an enterprise risk tolerance value set. Since the dynamic Bayesian network can simulate random processes in a dynamic environment, the dynamic Bayesian network can be used to simulate and predict the dynamic random change process of the enterprise risk response capability and enterprise risk tolerance of each enterprise node in the enterprise node set included in the enterprise association knowledge graph.
[0104] The fourth step is to use a discrete time Markov chain to generate a value flow state transition probability matrix of the above enterprise node set based on the above historical value flow information set, the above enterprise risk response capability value set and the enterprise risk tolerance value set. Among them, the above value flow state transition probability matrix can represent the probability value of the enterprise node switching between the five states. The above enterprise risk response capability value set and the above enterprise risk tolerance value set obey exponential distribution.
[0105] The fifth step is to determine the current value flow state information of each enterprise node in the enterprise node set through the value flow state transfer probability matrix to obtain the current value flow state information set. The current value flow state information can represent the operational capability of the current enterprise node.
[0106] As an example, the execution entity may perform a product operation on the value flow state transition probability matrix and the enterprise value flow state information set to obtain a current value flow state information set.
[0107] In the sixth step, according to the above-mentioned current value flow state information set and the preset division strategy, the enterprise risk state information of each enterprise node in the above-mentioned enterprise node set is determined to obtain the enterprise risk state information set. Among them, the above-mentioned preset division strategy can be a pre-defined strategy information for dividing the risk of the enterprise node set through the current value flow state information set. The above-mentioned enterprise risk state information can characterize the flow capacity of the enterprise risk corresponding to the enterprise node. The above-mentioned enterprise risk state information includes: susceptible state information, latent state information, first infection state information, second infection state information and immune state information. The above-mentioned susceptible state information characterizes that the enterprise node itself has a low risk and a high operational capacity, and is easily infected by individuals with associated states of latent state information, first infection state information and second infection state information. The above-mentioned latent state information can characterize that the enterprise node has been infected with risk, but it occurs within the enterprise, the enterprise risk is low, and after the latent period, it has the ability to transfer risk to the outside, and the risk transfer capacity is low. The above-mentioned first infection state information can characterize that the enterprise node has been infected with risk, the enterprise risk is high, the operational capacity is low, and the risk transfer capacity is high. The above-mentioned second infection status information can indicate that the enterprise node has been infected, the enterprise risk is extremely high, the operation capacity is low, and the risk transfer capacity is extremely high. The above-mentioned immune status information can indicate that the enterprise node has the lowest risk and will not be transferred by risk for a period of time, but will lose immunity and become susceptible after a period of time. For example, the above-mentioned period of time can be one month. The above-mentioned preset division strategy can be that the immune status information corresponds to the complete value flow status information, the susceptible status information corresponds to the high-level value flow status information, the latent status information corresponds to the medium-level value flow status information, the first infection status information corresponds to the low-level value flow status information, and the second infection status information corresponds to the risk value flow status information.
[0108] The seventh step is to determine the node aggregation value of each enterprise node in the enterprise node set in the enterprise-related knowledge graph to obtain a node aggregation value set. The node aggregation value may be the ratio of the number of edges associated with the node enterprise to the number of associated edges included in the entire enterprise-related knowledge graph.
[0109] Step 8: Based on the node aggregation value set and the enterprise risk status information set, an enterprise risk transfer model is constructed. The enterprise risk transfer model can be a formula for risk probability transfer that represents the probability value of each enterprise node being dynamically infected under the network topology conditions that the volt state information, the first infection state information, the second infection state information have certain risk transfer capabilities and the enterprise associated knowledge graph.
[0110] As an example, the above-mentioned execution entity can adopt an infectious disease dynamics model to construct an enterprise risk flow model based on the above-mentioned node aggregation value set and the above-mentioned enterprise risk status information set.
[0111] The ninth step is to solve the differential equation of the enterprise risk flow model to generate the risk flow probability value of each enterprise node in the enterprise node set to obtain a risk flow probability value set. The risk flow probability value can represent the probability value of an enterprise node being infected by an enterprise node with risk flow through the topological structure of the enterprise-related knowledge graph. The larger the risk flow probability value, the greater the possibility of risk infection. The solution to the differential equation can be the solution of the Euler method.
[0112] Step 10: Determine at least one risk transfer probability value in the risk transfer probability value set that is greater than or equal to a preset risk transfer probability value as a target risk transfer probability value set. The preset risk transfer probability value may be a preset minimum value to be transferred. For example, the preset risk transfer probability value may be 0.7.
[0113] In the eleventh step, the enterprise node set corresponding to the above target risk transfer probability value set is determined as the enterprise information set to be transferred.
[0114] The above technical solution and its related contents, combined with step "step 109", as an inventive point of the embodiment of the present disclosure, solve the technical problem three mentioned in the background technology "Because the infectious disease model divides the enterprise risk infection state into infection state, susceptible state and immune state, the division is not fine enough, and the consideration of the risk flow path is too single, resulting in low risk flow prediction accuracy and low early warning efficiency and reliability, the flow of enterprise risk in various enterprise systems, resulting in low security and stability of the enterprise system, reducing the timeliness of early warning information and the waste of communication resources for sending early warning information". The factors that lead to low security and stability of the enterprise system, reducing the timeliness of early warning information and the waste of communication resources for sending early warning information are often as follows: Because the infectious disease model divides the enterprise risk infection state into infection state, susceptible state and immune state, the division is not fine enough, and the consideration of the risk flow path is too single, resulting in low risk flow prediction accuracy and low early warning efficiency and reliability, the flow of enterprise risk in various enterprise systems. If the above factors are solved, the effect of improving the security and stability of the enterprise system, improving the timeliness of early warning information and reducing the waste of communication resources for sending early warning information can be achieved. In order to achieve this effect, the present disclosure first determines the enterprise value flow state information set through the historical value flow information set, and expands and subdivides it into five states from low to high, and combines the discrete time Markov chain and the dynamic Bayesian network to simulate the enterprise risk response capability value and enterprise risk tolerance value of the enterprise node, accurately describing the change of the enterprise value flow state information of each enterprise node over time under the topological structure formed by the enterprise-related knowledge graph. Secondly, the infectious disease model is linked to the enterprise value flow state information, and the three individuals in the infectious disease model are expanded into susceptible state information, latent state information, first infection state information, second infection state information and immune state information through the five enterprise value flow state information of the enterprise node, as well as the node aggregation value of each node enterprise, to construct an enterprise risk flow model in which the enterprise node is in the latent state, the first infection state and the second infection state, which can improve the prediction accuracy of the enterprise to be circulated and reduce the risk of enterprise loss. Finally, by sending a graded warning to the accurately predicted enterprise information set to be circulated, the accuracy and timeliness of the warning can be improved and the waste of communication resources can be reduced.
[0115] Step 109, use the associated terminal display screen to display each piece of enterprise information to be transferred in the enterprise information set to be transferred in a graded manner, and control the sound and light warning device to issue a graded warning through sounds of different frequencies and flashing lights of different colors that represent the warning level, so as to generate risk warning information for each piece of enterprise information to be transferred in the enterprise information set to be transferred, and send the generated risk warning information set to the enterprise terminal corresponding to the enterprise information set to be transferred.
[0116] In some embodiments, the execution subject may use the associated terminal display screen to perform hierarchical display for each enterprise information to be transferred in the enterprise information set to be transferred, and control the sound and light warning device to perform level warning through sounds of different frequencies and flashes of different colors that represent the warning level, so as to generate risk warning information for each enterprise information to be transferred in the enterprise information set to be transferred, and send the generated risk warning information set to the enterprise terminal corresponding to the enterprise information set to be transferred. Among them, the hierarchical display may be a hierarchical display of the risk flow probability value. The risk flow probability value may be a probability value of being infected by the risk obtained by the virus transmission model. The associated terminal display screen may be a display screen connected to the execution subject by wireless connection. The sound and light preset device may be a device that performs warning prompts from high to low sounds and flashes by means of sound frequencies from large to small and preset color warning level strategies. The preset color warning level strategy may be a strategy that indicates that the warning level is weakened in sequence according to the seven colors of the rainbow. The risk warning information may be warning information used to prompt the enterprise that there is a risk.
[0117] The above-mentioned embodiments of the present disclosure have the following beneficial effects: the risk warning information generation method of some embodiments of the present disclosure can improve the risk detection accuracy and warning reliability, improve the security and stability of the enterprise system, timely block the risk flow, and reduce the waste of communication resources. Specifically, the reasons for the omission and false detection of relevant enterprise warnings, the reduction of the timeliness and reliability of enterprise warnings, the extension of the warning information processing time and the waste of warning information communication resources are: because the random walk graph embedding model is difficult to capture the multi-association relationship between different types of enterprise nodes, it can only extract local or global features, resulting in low accuracy of enterprise anomaly detection, a large number of erroneous and redundant abnormal enterprise information, resulting in omission and false detection of enterprise warnings, reducing the timeliness and reliability of enterprise warnings, and extending the warning information processing time and the waste of warning information communication resources. Based on this, the risk warning information generation method of some embodiments of the present disclosure can first, in response to the detection of business data query request information, obtain a multi-source heterogeneous business data set of the enterprise from a distributed database. Here, the enterprise multi-source heterogeneous business data set is used for the subsequent construction of the enterprise knowledge graph. Secondly, according to the above-mentioned enterprise multi-source heterogeneous business data set, an initial enterprise association knowledge graph is constructed. Here, since the enterprise multi-source heterogeneous business data set comes from multiple heterogeneous data sources, the data sources are rich, which can improve the coverage of the enterprise-related knowledge graph. Again, the above-mentioned initial enterprise-related knowledge graph is subjected to link prediction processing to obtain the enterprise-related knowledge graph. Here, the accuracy and richness of the enterprise-related knowledge graph can be improved. Subsequently, the above-mentioned enterprise-related knowledge graph is subjected to association relationship screening processing to obtain the screened enterprise-related knowledge graph. Here, it is convenient to perform targeted anomaly detection for different associations in the future, so as to improve the accuracy of subsequent anomaly detection. Next, the risk structure of the above-mentioned screened enterprise-related knowledge graph is identified to obtain a risk enterprise-related network subgraph set. Here, the risk structure of the screened enterprise-related knowledge graph is targeted to improve the accuracy of identification. Afterwards, each enterprise node included in each risk enterprise-related network subgraph in the above-mentioned risk enterprise-related network subgraph set is represented by adjacent node embedding to generate a node embedding feature vector and obtain a node embedding feature vector set. Here, by combining the embedding of spatial features and semantic features of adjacent nodes, deeper and richer feature vectors of nodes can be extracted, improving the quality and accuracy of node embedding feature vectors. Then, based on the above node embedding feature vector set, abnormal risk detection is performed on the above risk enterprise association network subgraph set to obtain an enterprise risk information set. Here, abnormal risk prediction is performed through high-quality node embedding feature vectors to improve the accuracy of detection and obtain more accurate enterprise risk information. Afterwards, based on the above enterprise risk information set, enterprise risk flow prediction is performed on the enterprise node set included in the above enterprise association knowledge graph to obtain an enterprise information set to be circulated.Here, the risk transfer process and risk transfer scope of the enterprise corresponding to the enterprise risk information set can be determined, and more accurate enterprise information to be transferred can be obtained, so as to improve the efficiency of blocking risk transfer. Finally, the associated terminal display screen is used to display each enterprise information to be transferred in the above-mentioned enterprise information set to be transferred, and the sound and light warning device is controlled to perform level warnings through sounds of different frequencies and flashes of different colors that represent the warning level, so as to generate risk warning information for each enterprise information to be transferred in the above-mentioned enterprise information set to be transferred, and the generated risk warning information set is sent to the enterprise terminal corresponding to the above-mentioned enterprise information set to be transferred. Here, the enterprise security and warning accuracy can be improved. It can be obtained that the risk warning information generation method can improve the risk detection accuracy and warning reliability, improve the security and stability of the enterprise system, timely block the risk flow, and reduce the waste of communication resources.
[0118] Further references Figure 2 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a risk warning information generating device. These device embodiments are similar to Figure 1 Corresponding to the method embodiments shown, the risk warning information generating device can be specifically applied to various electronic devices.
[0119] like Figure 2As shown, a risk warning information generating device 200 includes: an acquisition unit 201, a construction unit 202, a link prediction processing unit 203, a screening unit 204, a risk structure identification unit 205, an adjacent node embedding representation unit 206, an abnormal risk detection unit 207, a risk flow prediction unit 208 and a level warning unit 209. Among them, the acquisition unit 201 is configured to: in response to detecting the business data query request information, obtain the enterprise multi-source heterogeneous business data set from the distributed database. The construction unit 202 is configured to: construct an initial enterprise association knowledge graph according to the above-mentioned enterprise multi-source heterogeneous business data set. The link prediction processing unit 203 is configured to: perform link prediction processing on the above-mentioned initial enterprise association knowledge graph to obtain the enterprise association knowledge graph. The screening unit 204 is configured to: perform association relationship screening processing on the above-mentioned enterprise association knowledge graph to obtain the filtered enterprise association knowledge graph. The risk structure identification unit 205 is configured to: perform risk structure identification on the above-mentioned filtered enterprise association knowledge graph to obtain the risk enterprise association network subgraph set. The adjacent node embedding representation unit 206 is configured to: perform adjacent node embedding representation on each enterprise node included in each risk enterprise associated network subgraph in the risk enterprise associated network subgraph set to generate a node embedding feature vector and obtain a node embedding feature vector set. The abnormal risk detection unit 207 is configured to: perform abnormal risk detection on the risk enterprise associated network subgraph set according to the node embedding feature vector set to obtain an enterprise risk information set. The risk flow prediction unit 208 is configured to: perform enterprise risk flow prediction on the enterprise node set included in the enterprise associated knowledge graph according to the enterprise risk information set to obtain an enterprise information set to be transferred. The level warning unit 209 is configured to: use the associated terminal display screen to perform graded display for each enterprise information to be transferred in the enterprise information set to be transferred, and control the sound and light warning device to perform graded warning through sounds of different frequencies and flashing lights of different colors that represent the warning level, so as to generate risk warning information for each enterprise information to be transferred in the enterprise information set to be transferred, and send the generated risk warning information set to the enterprise terminal corresponding to the enterprise information set to be transferred.
[0120] It is understandable that the various units recorded in the risk warning information generating device 200 and the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the risk warning information generating device 200 and the units included therein, and will not be described in detail here.
[0121] Reference below Figure 3 , which shows a structural schematic diagram of an electronic device (eg, an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.
[0122] like Figure 3 As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.
[0123] Typically, the following devices may be connected to the I / O interface 305: input devices 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 308 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 309. The communication devices 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as required.
[0124] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.
[0125] It should be noted that the computer-readable medium in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.
[0126] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (Hyper Text Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.
[0127] The above-mentioned computer-readable medium may be included in the above-mentioned electronic device; or it may exist independently without being installed in the electronic device. The above-mentioned computer-readable medium carries one or more programs. When the above-mentioned one or more programs are executed by the electronic device, the electronic device: in response to detecting business data query request information, obtains an enterprise multi-source heterogeneous business data set from a distributed database; constructs an initial enterprise-related knowledge graph based on the above-mentioned enterprise multi-source heterogeneous business data set; performs link prediction processing on the above-mentioned initial enterprise-related knowledge graph to obtain an enterprise-related knowledge graph; performs association relationship screening processing on the above-mentioned enterprise-related knowledge graph to obtain a screened enterprise-related knowledge graph; performs risk structure identification on the above-mentioned screened enterprise-related knowledge graph to obtain a risk enterprise-related network subgraph set; performs adjacent node embedding representation on each enterprise node included in each risk enterprise-related network subgraph in the above-mentioned risk enterprise-related network subgraph set to generate Node embedding feature vectors are obtained to obtain node embedding feature vector sets; based on the above node embedding feature vector sets, abnormal risk detection is performed on the above risk enterprise associated network subgraph set to obtain an enterprise risk information set; based on the above enterprise risk information set, enterprise risk flow prediction is performed on the enterprise node set included in the above enterprise associated knowledge graph to obtain an enterprise information set to be transferred; each enterprise information to be transferred in the above enterprise information set to be transferred is displayed in a graded manner using an associated terminal display screen, and the sound and light warning device is controlled to perform a graded warning through sounds of different frequencies and flashing lights of different colors that represent the warning level, so as to generate risk warning information for each enterprise information to be transferred in the above enterprise information set to be transferred, and the generated risk warning information set is sent to the enterprise terminal corresponding to the above enterprise information set to be transferred.
[0128] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0129] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0130] The units described in some embodiments of the present disclosure may be implemented by software or by hardware. The described units may also be set in a processor, for example, may be described as: a processor including an acquisition unit, a construction unit, a link prediction processing unit, a screening unit, a risk structure identification unit, an adjacent node embedding representation unit, an abnormal risk detection unit, a risk flow prediction unit and a level warning unit. Among them, the names of these units do not constitute a limitation on the unit itself in some cases. For example, the acquisition unit may also be described as "a unit for acquiring an enterprise multi-source heterogeneous business data set from a distributed database in response to detecting business data query request information."
[0131] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.
[0132] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) and the technical solutions formed.
Claims
1. A method for generating risk warning information, comprising: In response to detecting the business data query request information, acquiring the enterprise multi-source heterogeneous business data set from the distributed database; Constructing an initial enterprise-related knowledge graph based on the enterprise's multi-source heterogeneous business data set; Performing link prediction processing on the initial enterprise-related knowledge graph to obtain an enterprise-related knowledge graph; Performing association relationship screening processing on the enterprise-related knowledge graph to obtain a screened enterprise-related knowledge graph; Performing risk structure identification on the selected enterprise-related knowledge graph to obtain a risk enterprise-related network subgraph set; Performing adjacent node embedding representation on each enterprise node included in each venture enterprise associated network subgraph in the venture enterprise associated network subgraph set to generate a node embedding feature vector, and obtaining a node embedding feature vector set, wherein performing adjacent node embedding representation on each enterprise node included in each venture enterprise associated network subgraph in the venture enterprise associated network subgraph set to generate a node embedding feature vector, and obtaining a node embedding feature vector set includes: For each venture enterprise association network subgraph in the venture enterprise association network subgraph set, the following feature embedding step is performed: Determine a venture enterprise node pair set corresponding to the venture enterprise node group included in the venture enterprise association network subgraph; Determine a node meta-path group of each venture enterprise node pair in the venture enterprise node pair set to obtain a node meta-path group set; Classifying the node meta-path group set to obtain a category node meta-path group; For each class node meta-path in the class node meta-path group, the following feature vector determination step is performed: Performing node space encoding on each category node included in the category node meta-path to generate a node space feature vector, thereby obtaining a node space feature vector group; Determine a starting spatial feature vector of a starting risk enterprise node and a terminating spatial feature vector of a terminating risk enterprise node according to the node spatial feature vector group, wherein the starting risk enterprise node is a risk node at a starting position included in the risk enterprise node pair corresponding to the category node meta-path, and the terminating risk enterprise node is a risk node at a terminating position included in the risk enterprise node pair; Performing node feature mapping processing on the starting spatial feature vector and the ending spatial feature vector to obtain a starting mapping feature vector and an ending mapping feature vector; Determine a node pair weight feature vector of the start mapping feature vector and the end mapping feature vector under the category node meta-path; Normalizing the node pair weight feature vector to obtain a normalized node pair weight feature vector; Determine the product of the normalized node pair weight feature vector and the termination space feature vector as the node path feature vector of the starting venture enterprise node under the category node meta-path; Determine the venture enterprise node group corresponding to the venture enterprise node pair set as the adjacent venture enterprise node group of the starting venture enterprise node; Performing feature concatenation processing on the node path feature vector group corresponding to the adjacent risk enterprise node group to obtain a spatial embedding feature vector; Determine a semantic embedding feature vector of the starting venture enterprise node under the category node meta-path group; Generate a node embedding feature vector set according to the obtained spatial embedding feature vector set and the obtained semantic embedding feature vector set, wherein generating a node embedding feature vector set according to the obtained spatial embedding feature vector set and the obtained semantic embedding feature vector set includes: For each spatial embedding feature vector in the spatial embedding feature vector set, the following embedding feature vector determination step is performed: Performing nonlinear transformation processing on the spatial embedding feature vector to obtain a transformed spatial embedding feature vector; Performing node semantic encoding on the starting risk enterprise node corresponding to the converted spatial embedding feature vector to obtain a risk node semantic feature vector; Determine the product of the semantic embedding feature vector corresponding to the spatial embedding feature vector and the converted spatial embedding feature vector as a path semantic feature vector; Normalizing the path semantic feature vector to obtain a normalized path semantic feature vector; Determine the product of the normalized path semantic feature vector and the space embedding feature vector as a node embedding feature vector; According to the node embedding feature vector set, abnormal risk detection is performed on the risk enterprise association network subgraph set to obtain an enterprise risk information set; According to the enterprise risk information set, enterprise risk flow prediction is performed on the enterprise node set included in the enterprise association knowledge graph to obtain an enterprise information set to be circulated; Use the associated terminal display screen to display the levels of each piece of enterprise information to be transferred in the enterprise information set to be transferred, and control the sound and light warning device to issue level warnings through sounds of different frequencies and flashing lights of different colors that represent the warning levels, so as to generate risk warning information for each piece of enterprise information to be transferred in the enterprise information set to be transferred, and send the generated risk warning information set to the enterprise terminal corresponding to the enterprise information set to be transferred.
2. The method according to claim 1, wherein: The step of constructing an initial enterprise-related knowledge graph based on the enterprise's multi-source heterogeneous business data set includes: Performing data fusion processing on the enterprise multi-source heterogeneous business data set to obtain a fused enterprise multi-source heterogeneous business data set; Constructing a business ontology data set according to the merged enterprise multi-source heterogeneous business data set, wherein the business ontology data in the business ontology data set is one of the following: business entity class, business object attribute and business object association relationship; For each business field in the business field set included in the fused enterprise multi-source heterogeneous business data set, the following mapping relationship determination step is performed: Determine a field similarity value between the business field and each business ontology data in the business ontology data set to obtain a field similarity value group; Filtering at least one field similarity value greater than or equal to a preset similarity threshold from the field similarity value group as a target field similarity value, and obtaining a target field similarity value group; The business ontology data group corresponding to the target field similarity value group is determined as the business ontology data group corresponding to the business field, and a field ontology mapping relationship group is obtained; the enterprise business data corresponding to each business ontology data in the business ontology data set is extracted from the enterprise multi-source heterogeneous business data set through the obtained field ontology mapping relationship group set, and a business entity relationship triple set is obtained; Performing instance deduplication processing on the business entity relationship triple set to obtain a deduplicated business entity relationship triple set; The deduplicated business entity relationship triple set is input into the graph database to obtain an initial enterprise-related knowledge graph.
3. The method according to claim 1, wherein: The risk structure identification is performed on the selected enterprise-related knowledge graph to obtain a risk enterprise-related network subgraph set, including: Performing self-correlation edge removal processing on the filtered enterprise-related knowledge graph to obtain a post-removal enterprise-related knowledge graph; Performing sub-graph segmentation processing on the enterprise-related knowledge graph after removal to obtain an enterprise-related network sub-graph set; Performing association relationship classification processing on the enterprise association network subgraph set to obtain an association relationship network subgraph group set; For each association relationship network subgraph group in the association relationship network subgraph group set, risk structure identification is performed on each association relationship network subgraph in the association relationship network subgraph group to obtain a risk enterprise association network subgraph group; The obtained venture enterprise association network subgraph group set is determined as the venture enterprise association network subgraph set.
4. A risk warning information generating device, comprising: An acquisition unit, configured to acquire enterprise multi-source heterogeneous business data sets from a distributed database in response to detecting business data query request information; A construction unit, configured to construct an initial enterprise-related knowledge graph based on the enterprise's multi-source heterogeneous business data set; A link prediction processing unit is configured to perform link prediction processing on the initial enterprise-related knowledge graph to obtain an enterprise-related knowledge graph; A screening unit is configured to perform a screening process on the enterprise-related knowledge graph to obtain a screened enterprise-related knowledge graph; A risk structure identification unit is configured to perform risk structure identification on the screened enterprise-related knowledge graph to obtain a risk enterprise-related network subgraph set; The adjacent node embedding representation unit is configured to perform adjacent node embedding representation on each enterprise node included in each venture enterprise associated network subgraph in the venture enterprise associated network subgraph set to generate a node embedding feature vector and obtain a node embedding feature vector set, wherein the adjacent node embedding representation on each enterprise node included in each venture enterprise associated network subgraph in the venture enterprise associated network subgraph set to generate a node embedding feature vector and obtain a node embedding feature vector set comprises: for each venture enterprise associated network subgraph in the venture enterprise associated network subgraph set, performing the following feature embedding steps: determining the A venture enterprise node pair set corresponding to a venture enterprise node group included in a venture enterprise association network subgraph; determining a node meta-path group of each venture enterprise node pair in the venture enterprise node pair set to obtain a node meta-path group set; classifying the node meta-path group set to obtain a category node meta-path group; for each category node meta-path in the category node meta-path group, executing the following feature vector determination step: performing node space encoding on each category node included in the category node meta-path to generate a node space feature vector to obtain a node space feature vector group; determining a starting space feature vector of a starting venture enterprise node and a terminating space feature vector of a terminating venture enterprise node based on the node space feature vector group, wherein the starting venture enterprise node is a risk node located at a starting position included in the venture enterprise node pair corresponding to the category node meta-path, and the terminating venture enterprise node is a risk node located at a terminating position included in the venture enterprise node pair; performing node feature mapping processing on the starting space feature vector and the terminating space feature vector to obtain a starting mapping feature vector and a terminating mapping feature vector; determining a node pair weight feature vector of the starting mapping feature vector and the terminating mapping feature vector under the category node meta-path; The weight feature vector is normalized to obtain a normalized node pair weight feature vector; the product of the normalized node pair weight feature vector and the termination space feature vector is determined as the node path feature vector of the starting venture enterprise node under the category node meta-path; the venture enterprise node group corresponding to the venture enterprise node pair set is determined as the adjacent venture enterprise node group of the starting venture enterprise node; the node path feature vector group corresponding to the adjacent venture enterprise node group is feature spliced to obtain a spatial embedding feature vector; the semantic embedding feature vector of the starting venture enterprise node under the category node meta-path group is determined;Generate a node embedding feature vector set according to the obtained spatial embedding feature vector set and the obtained semantic embedding feature vector set, wherein the generating the node embedding feature vector set according to the obtained spatial embedding feature vector set and the obtained semantic embedding feature vector set comprises: for each spatial embedding feature vector in the spatial embedding feature vector set, perform the following embedding feature vector determination steps: perform nonlinear transformation processing on the spatial embedding feature vector to obtain a transformed spatial embedding feature vector; perform node semantic encoding on the starting risk enterprise node corresponding to the transformed spatial embedding feature vector to obtain a risk node semantic feature vector; determine the product of the semantic embedding feature vector corresponding to the spatial embedding feature vector and the transformed spatial embedding feature vector as a path semantic feature vector; perform normalization processing on the path semantic feature vector to obtain a normalized path semantic feature vector; determine the product of the normalized path semantic feature vector and the spatial embedding feature vector as a node embedding feature vector; an abnormal risk detection unit configured to perform abnormal risk detection on the risk enterprise association network subgraph set according to the node embedding feature vector set to obtain an enterprise risk information set; A risk transfer prediction unit is configured to perform enterprise risk transfer prediction on the enterprise node set included in the enterprise association knowledge graph according to the enterprise risk information set, and obtain a set of enterprise information to be transferred; The level warning unit is configured to use the associated terminal display screen to display levels for each piece of enterprise information to be transferred in the enterprise information set to be transferred, and to control the sound and light warning device to issue level warnings through sounds of different frequencies and flashing lights of different colors that represent the warning levels, so as to generate risk warning information for each piece of enterprise information to be transferred in the enterprise information set to be transferred, and send the generated risk warning information set to the enterprise terminal corresponding to the enterprise information set to be transferred.
5. An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 3.
6. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.
Citation Information
Patent Citations
Data processing method and device, model training method and device, risk identification method and device and storage medium
CN114942997A
Enterprise risk conduction prediction method and system based on graph attention mechanism
CN116187767A
Knowledge graph-based listed enterprise illegal risk prediction method
CN116384551A
Enterprise credit risk assessment method based on cascaded hypergraph neural network
CN117350845A
Method, system and equipment for evaluating credit risk of listed company based on mapping knowledge domain
CN117522124A
