Artificial intelligence-based enterprise operation big data wisdom supervision platform

Abstract

The application relates to an enterprise operation big data wisdom supervision platform based on artificial intelligence, and belongs to the technical field of enterprise operation management. The platform analyzes the time sequence characteristics and the space characteristics of the neighborhood range of an enterprise node in a graph structure, obtains the neighborhood propagation stability, the propagation expansion capacity and the propagation hub dependency of the enterprise node and each neighborhood node, comprehensively obtains the propagation permission degree which can represent the risk analysis importance degree of the correlation edge between the enterprise node and each neighborhood node, and screens out effective value propagation edges and corresponding risk propagation weights. Then, the interaction information of the neighborhood node is obtained based on the effective value propagation edge set, and the enterprise node is weighted and information is aggregated based on the risk propagation weight to complete risk propagation analysis. The application effectively improves the neighborhood equal-weight aggregation problem of the traditional TGN in the risk propagation analysis of the enterprise correlation network, and significantly improves the reliability of the enterprise operation data risk propagation analysis and supervision output.

Description

Technical Field

[0001] This invention relates to the field of enterprise operation management technology, and in particular to an intelligent supervision platform for enterprise operation big data based on artificial intelligence. Background Technology

[0002] With the continuous advancement of enterprise digital transformation and the construction of data governance systems, the scale of data generated during enterprise operations is showing a sustained growth trend. Against this backdrop, enterprise operation big data supervision platforms, through centralized governance and correlation modeling of enterprise operational data, can promptly detect abnormal changes, potential risks, and compliance issues in enterprise operations, and analyze the scope and spread trends of risk impacts, thereby achieving a shift from monitoring single-point indicators to supervising the overall operational status.

[0003] Enterprise operation big data supervision platforms typically use Temporal Graph Network (TGN) models to learn representations of enterprise nodes and their dynamic interaction relationships. When constructing the historical interaction neighborhood of a target node, the platform mainly determines the neighborhood range based on historical interaction data and uniformly incorporates neighborhood relationships that meet the sampling conditions into the message aggregation and state update process, without distinguishing the actual propagation role of neighborhood relationships in the risk propagation process.

[0004] However, in the process of enterprise operation, there are often a large number of constantly changing relationships between enterprise nodes. The propagation effect of different relationships in the enterprise operation dynamic diagram varies significantly. Without distinguishing the actual propagation effect, weaker neighboring relationships and key propagation neighboring relationships will participate in the target node state representation together. This weakens the influence of key propagation neighboring relationships in the node representation, making it difficult for neighboring relationships that play an important role in risk propagation to be fully reflected in the state update process. Consequently, there is a discrepancy between the propagation structure reflected in the enterprise operation dynamic diagram and the actual risk propagation structure, affecting the accuracy of risk propagation link identification, propagation intensity assessment, and propagation range analysis.

[0005] In other words, the current enterprise operation big data supervision platform lacks sufficient reliability in analyzing the risk propagation of enterprise operation data. Summary of the Invention

[0006] In view of this, the present invention provides an intelligent supervision platform for enterprise operation big data based on artificial intelligence, in order to solve the technical problem of insufficient reliability of current supervision platforms in risk propagation analysis of enterprise operation data.

[0007] This invention discloses an intelligent monitoring platform for enterprise operations based on big data, comprising a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it performs the following steps:

[0008] Acquire enterprise operation data and construct an enterprise operation dynamic graph. Take any enterprise node in the enterprise operation dynamic graph as the target node, and take any of the associated edges formed between the target node and each neighboring node as the edge to be analyzed. Determine the neighborhood propagation stability of the edge to be analyzed based on the time interval and time span of the interaction events corresponding to the edge to be analyzed.

[0009] Based on the difference between the union and intersection of the sets of associated nodes corresponding to the two nodes of the edge to be analyzed, the propagation extension degree of the edge to be analyzed is determined; based on the degree of each node in the set of associated nodes corresponding to the two nodes of the edge to be analyzed, the propagation hub dependency degree of the edge to be analyzed is determined.

[0010] The propagation permission of the edge to be analyzed is determined based on the neighborhood propagation stability, the propagation expansion, and the propagation hub dependency. The effective value propagation edge and the corresponding risk propagation weight in the associated edges formed between the target node and each neighboring node are determined based on the propagation permission.

[0011] Risk monitoring of the target node is completed based on the effective value propagation edge and the corresponding risk propagation weight.

[0012] Further, determining the neighborhood propagation stability of the edge to be analyzed includes:

[0013] The time interval between any two interaction events in the interaction events corresponding to the edge to be analyzed is recorded as the adjacent interaction interval. The absolute value of the difference between any adjacent interaction interval and the mean of all adjacent interaction intervals is recorded as the interaction discrete value of any adjacent interaction interval. The normalized value of the mean of the interaction discrete values ​​of all adjacent interaction intervals is recorded as the first neighborhood propagation stability characterization value.

[0014] The normalized value of the ratio of the time span of the interaction event corresponding to the edge to be analyzed to the mean of the time span of the interaction events corresponding to all related edges on the enterprise operation dynamic graph is denoted as the second neighborhood propagation stability characterization value.

[0015] The average of the first neighborhood propagation stability characterization value and the second neighborhood propagation stability characterization value is denoted as the neighborhood propagation stability of the edge to be analyzed.

[0016] Further, determining the propagation extent of the edge to be analyzed includes:

[0017] The two nodes corresponding to the edge to be analyzed are respectively denoted as the current first node and the current second node. Each node that is associated with the current first node forms a first set of associated nodes, and each node that is associated with the current second node forms a second set of associated nodes. The existence of an association relationship means that there is a direct or indirect connection relationship with the current first node or the current second node.

[0018] The union of the first set of associated nodes and the second set of associated nodes is calculated and recorded as the first set calculation result. The intersection of the first set of associated nodes and the second set of associated nodes is calculated and recorded as the second set calculation result. The ratio of the difference between the first set calculation result and the second set calculation result to the total number of nodes in the enterprise operation dynamic graph is recorded as the propagation expansion degree of the edge to be analyzed.

[0019] Further, determining the propagation hub dependency of the edge to be analyzed includes:

[0020] The average degree of each associated node in the first associated node set is denoted as the first degree mean, and the average degree of each associated node in the second associated node set is denoted as the second degree mean. The ratio of the average of the first degree mean and the second degree mean to the total number of nodes in the enterprise operation dynamic graph is denoted as the propagation hub dependency of the edge to be analyzed.

[0021] Further, determining the propagation permission of the edge to be analyzed includes:

[0022] The product of the mean of the propagation spread and the propagation hub dependency and the neighborhood propagation stability is denoted as the propagation permission of the edge to be analyzed.

[0023] Furthermore, determining the effective value propagation edges and corresponding risk propagation weights among the association edges formed between the target node and each neighboring node includes:

[0024] The effective value propagation edge is defined as the association edge whose propagation permission is greater than a preset propagation permission threshold among the association edges formed between the target node and each neighboring node.

[0025] The ratio of the propagation permission degree corresponding to any effective value propagation edge to the sum of the propagation permission degrees corresponding to all effective value propagation edges is denoted as the risk propagation weight of any effective value propagation edge.

[0026] The advantages of this invention compared to the prior art are:

[0027] This invention analyzes the temporal and spatial connectivity characteristics of the neighborhood of enterprise nodes within a graph structure. It obtains the neighborhood propagation stability, propagation expansion capability, and propagation hub dependence of each enterprise node and its neighboring nodes. This comprehensive analysis yields a propagation permission score, representing the importance of risk analysis for the corresponding edges between the enterprise node and its neighbors. Effective value propagation edges and their corresponding risk propagation weights are then selected. Based on the set of effective value propagation edges, the interaction information of neighboring nodes is obtained, and the enterprise node's information is weighted and aggregated based on the risk propagation weights to ultimately complete the risk propagation analysis. This invention successfully achieves the screening and weighting of neighboring nodes for enterprise nodes, ensuring that only edges with high propagation value participate in subsequent risk propagation analysis. It also differentiates the contribution of different edges during the propagation process, effectively improving the problem of equal-weighted neighborhood aggregation in traditional TGN (Traffic Network Network) risk propagation analysis of enterprise networks. This allows the node state update process to more realistically reflect the actual risk propagation structure characteristics between enterprises, significantly improving the accuracy and reliability of enterprise operational data risk propagation analysis and regulatory output. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a flowchart of an intelligent supervision method for enterprise operation big data based on artificial intelligence, provided in Embodiment 1 of the present invention. Detailed Implementation

[0030] The overall concept of this invention is as follows:

[0031] This invention analyzes the temporal and spatial connectivity features of the neighborhood of an enterprise node in a graph structure to obtain the neighborhood propagation stability, propagation expansion capability, and propagation hub dependence of the enterprise node and its neighboring nodes. It then obtains the propagation permission degree, which represents the importance of risk analysis of the corresponding association edges between the enterprise node and its neighboring nodes. Based on this, it selects effective value propagation edges and their corresponding risk propagation weights. Then, it obtains the interaction information of the neighboring nodes based on the set of effective value propagation edges, and performs weighted information aggregation on the enterprise nodes based on the risk propagation weights to finally complete the risk propagation analysis.

[0032] To further illustrate the technical solution of the present invention, specific embodiments are described below.

[0033] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a particular feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. Furthermore, a particular feature, structure, or characteristic in one or more embodiments may be combined in any suitable form, and the terms "comprising," "including," "having," and variations thereof mean "including, but not limited to," unless otherwise specifically emphasized.

[0034] It should be understood that the sequence number of each step in the following embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0035] System Implementation Example:

[0036] Embodiment 1 of the present invention provides an intelligent supervision platform for enterprise operations based on artificial intelligence big data, including a processor and a memory. The processor executes a computer program stored in the memory to implement an intelligent supervision method for enterprise operations based on artificial intelligence big data, such as... Figure 1 As shown, the method includes the following steps:

[0037] S101, acquire enterprise operation data and construct an enterprise operation dynamic graph, take any enterprise node in the enterprise operation dynamic graph as the target node, and take any of the associated edges formed between the target node and each neighboring node as the edge to be analyzed; determine the neighborhood propagation stability of the edge to be analyzed based on the time interval and time span of the interaction events corresponding to the edge to be analyzed.

[0038] To effectively monitor and manage enterprise operational big data, it is first necessary to acquire enterprise operational data. Enterprise operational data includes, but is not limited to: basic enterprise information data, data on interactions between enterprises, and textual information data generated during the enterprise's operations.

[0039] After acquiring enterprise operational data, this embodiment also performs unified preprocessing on various types of data to provide a reliable analytical foundation for subsequent risk analysis. This includes, but is not limited to: 1. Processing missing data using interpolation completion methods and historical record-based missing value imputation methods; 2. Identifying and removing or correcting abnormal data using statistical distribution-based anomaly detection methods; 3. Standardizing the format of data from different sources using unified data encoding and field mapping rules; 4. Converting and aligning time information from various data sources using a unified time base to achieve timestamp standardization; 5. Mapping unified identifiers of enterprise entities from different data sources based on entity resolution technology to complete enterprise entity alignment. Ultimately, this results in a unified, time-consistent, and analytically correlated enterprise operational dataset.

[0040] Then, based on the obtained enterprise operation dataset, for each inter-enterprise interaction record, the initiating enterprise, the receiving enterprise, the interaction time, and the corresponding event attribute information are extracted to form a time-series related event. Subsequently, all related events are organized in chronological order to obtain the enterprise operation time-series event flow, which is further constructed into a graph structure. In this graph structure, the enterprise entities are used as nodes, and the edge relationships and time-series edge sequences are determined according to the event flow. That is, the interaction events that occur between enterprises are used as the related edges between nodes, and the time of event occurrence is used to describe the time-series characteristics of the edges. When the same pair of enterprise entities interacts multiple times at different times, a corresponding time-series edge sequence is formed, ultimately resulting in an enterprise operation dynamic graph that reflects the dynamic evolution of enterprise relationships.

[0041] The dynamic graph of enterprise operation constructed in this embodiment differs from the structure of a conventional static graph. In a conventional graph structure, edges between nodes are typically used only to indicate a relationship between two nodes, with a unique edge corresponding to each pair of nodes. This edge only reflects the existence of a connection between nodes, not the change of the relationship over time. However, in an enterprise operation scenario, multiple interaction events typically occur between the same pair of enterprise entities at different times, and these interaction events exhibit clear temporal characteristics. Therefore, this embodiment adopts an event-driven dynamic graph construction method, using enterprise entities as graph nodes, relationships between enterprises as graph edges, and recording multiple interaction events between the same pair of enterprise entities as a set of temporal events on the corresponding relationship edges. For any relationship edge... The corresponding set of time-series events can be represented as ,in This indicates that node i and node j are in An interaction event occurred at that moment.

[0042] For ease of analysis, based on the constructed enterprise operation dynamic graph, any enterprise node is taken as the target node, and any of the associated edges formed between the target node and each neighboring node is taken as the edge to be analyzed.

[0043] In business operations, some neighborhood relationships stem from long-term, continuous business collaborations, financial transactions, or resource connections between enterprises, exhibiting strong continuity and stability in their interactions. Conversely, some neighborhood relationships are formed only by sporadic interaction events, resulting in short-lived connections and a lack of regularity in their interactions. For the latter, even if they form connections with target nodes in the graph structure, it is difficult to establish a continuous and stable information transmission path during subsequent risk propagation.

[0044] Therefore, this step first quantifies the ability of the neighborhood relationship between the target node and a certain neighboring node to form a stable propagation channel. As can be seen from the above, when the relationship exists for a long time and the interaction behavior has a strong regularity, it is easier to form a stable information transmission link and thus continuously participate in the risk propagation process. Conversely, if the relationship exists only briefly or the interaction behavior shows obvious random fluctuations, its contribution to the risk propagation process is usually sporadic and uncertain.

[0045] Specifically, by constructing a sequence of time intervals between adjacent interaction events and calculating the average absolute deviation of each time interval from the average interaction interval, the dispersion of interaction behavior in the time dimension is reflected; and by calculating the relative relationship between the life span of associated edges and the average life span of associated edges in the whole graph, the persistence level of the current association relationship relative to the overall network is evaluated.

[0046] Therefore, the neighborhood propagation stability of the edge to be analyzed can be determined, including:

[0047] The time interval between any two interaction events in the interaction events corresponding to the edge to be analyzed is recorded as the adjacent interaction interval. The absolute value of the difference between any adjacent interaction interval and the mean of all adjacent interaction intervals is recorded as the interaction discrete value of any adjacent interaction interval. The normalized value of the mean of the interaction discrete values ​​of all adjacent interaction intervals is recorded as the first neighborhood propagation stability characterization value.

[0048] The normalized value of the ratio of the time span of the interaction event corresponding to the edge to be analyzed to the mean of the time span of the interaction events corresponding to all related edges on the enterprise operation dynamic graph is denoted as the second neighborhood propagation stability characterization value.

[0049] The average of the first neighborhood propagation stability characterization value and the second neighborhood propagation stability characterization value is denoted as the neighborhood propagation stability of the edge to be analyzed.

[0050] Furthermore, as a preferred embodiment, the neighborhood propagation stability of the edge to be analyzed is:

[0051]

[0052] In the formula, Represents the target node Its neighboring nodes The neighborhood propagation stability of the associated edges (i.e., the edges to be analyzed) formed between the nodes is used to characterize the target node. With neighboring nodes The ability of the relationships between them to form stable transmission channels This represents the inverse proportional normalization function. Both represent hyperbolic normalization functions, and their function is to restrict the calculation result output to [0,1], ensuring that the larger the final output result, the higher the value of the target node. Its neighboring nodes The stronger the continuity and persistence of the relationship between them over time, the more likely they are to maintain a stable state of propagation over a long period of time.

[0053] in, Target node of reaction Its neighboring nodes The smaller the dispersion of the interaction events corresponding to the associated edges formed between them in the time dimension, the closer the time interval between each interaction event is, and the stronger the continuity and regularity of the association relationship. Indicates associated edges The total number of interactive events in the corresponding time-series interactive event set; Represents two consecutive interaction events (the first one) The second interaction event and the first The time interval between each interaction event, also known as the adjacent interaction interval, Indicates the average interaction time interval; This represents the degree of deviation of the k-th interaction time interval from the average interaction time interval, also known as the interaction discrete value, which is the degree of difference between the current interaction rhythm and the overall interaction rhythm. The larger the value, the more obvious the deviation of the current interaction behavior from the normal rhythm, and therefore the lower the neighborhood propagation stability.

[0054] in, Indicates associated edges The ratio of the time span of an interaction event to the time span of the average interaction events across the entire graph is used to characterize the relative persistence of the interaction relationship. The larger the result, the more it indicates that the relationship persists for a significantly longer period than the network average, indicating a stronger ability to maintain the relationship and a greater likelihood of forming a long-term and stable risk propagation channel. Indicates associated edges The time of occurrence of the first interaction event in the corresponding set of sequential interaction events. Indicates associated edges The difference between the occurrence times of the last interaction event in the corresponding set of time-series interaction events. This represents the duration from the first interaction to the last interaction in the relationship; M represents the total number of all related edges in the enterprise operation dynamic graph. This represents the time span of the a-th related edge in the enterprise operation dynamic diagram, that is, the time span corresponding to the a-th related edge. value.

[0055] S102, based on the difference between the union and intersection of the sets of associated nodes of the two nodes corresponding to the edge to be analyzed, determine the propagation extension degree of the edge to be analyzed, and based on the degree of each node in the set of associated nodes of the two nodes corresponding to the edge to be analyzed, determine the propagation hub dependency degree of the edge to be analyzed.

[0056] The target node was obtained through the previous step. Its neighboring nodes The stability of neighborhood propagation of the associated edges is considered. However, this parameter only reflects whether the relationship can exist stably in the long term, and does not reflect the strength of the propagation effect of the relationship in the dynamic diagram of enterprise operations. Furthermore, even if different associated edges have high propagation stability, their positions and connectivity within the graph structure can still differ significantly. For example, some associated edges only connect nodes within a local neighborhood, and their risk propagation range is relatively limited; while other associated edges can connect groups of nodes between different local structures, forming cross-regional and cross-level information transmission paths during risk propagation, thus having a greater impact on the scope and efficiency of risk diffusion. Therefore, stable neighborhood relationships do not necessarily have the same propagation value.

[0057] Therefore, this step further analyzes the information propagation capability of the associated edges (edges to be analyzed) formed between adjacent nodes. When the neighborhood relationship has both high stability and strong propagation capability, it is more likely to become a key connection unit in the risk propagation link. Conversely, even if the association relationship exists for a long time, if it only plays a local connection role, its contribution to the overall risk propagation process is still relatively limited.

[0058] Specifically, in enterprise operation networks, some risk propagation processes do not rely on uniform diffusion but rather on a small number of highly connected nodes to complete cross-regional propagation and risk amplification. The quantification of propagation coverage capability characterizes the potential node range that the area connected by the associated edge can reach. The larger the node range covered by both ends of the associated edge, the more enterprise nodes the risk information can affect after propagation through that associated edge, and its potential propagation range is wider. Simultaneously, when the neighboring nodes as a whole have high connectivity, it indicates that the area where the associated edge is located has strong hub propagation characteristics, and risks are more easily spread to a wider area through these highly connected nodes.

[0059] Therefore, based on the above ideas, the propagation expansion capability and propagation hub dependency of the edges associated with the target node and its neighboring nodes can be obtained respectively. The propagation coverage capability reflects how many potential objects the risk can propagate to, and the propagation hub dependency reflects the structure through which the risk spreads.

[0060] Therefore, the propagation extent of the edge to be analyzed can be determined first, including:

[0061] The two nodes corresponding to the edge to be analyzed are respectively denoted as the current first node and the current second node. Each node that is associated with the current first node forms a first set of associated nodes, and each node that is associated with the current second node forms a second set of associated nodes. The existence of an association relationship means that there is a direct or indirect connection relationship with the current first node or the current second node.

[0062] The union of the first set of associated nodes and the second set of associated nodes is calculated and recorded as the first set calculation result. The intersection of the first set of associated nodes and the second set of associated nodes is calculated and recorded as the second set calculation result. The ratio of the difference between the first set calculation result and the second set calculation result to the total number of nodes in the enterprise operation dynamic graph is recorded as the propagation expansion degree of the edge to be analyzed.

[0063] The formula for the propagation extension of the edge to be analyzed is as follows:

[0064]

[0065] In the formula, Represents the target node Its neighboring nodes The propagation extent of the corresponding associated edge (the edge to be analyzed) is used to characterize the potential propagation range that the associated edge can reach. The larger the result, the wider the structural coverage of the node connected by the associated edge in the network, so that it can reach more potential propagation objects in the process of risk propagation and has a stronger propagation extent.

[0066] in, This indicates the relationship between the target node and the enterprise operation dynamic diagram. The set consisting of all nodes that have a relationship, that is, the first / second set of related nodes. This indicates the relationship between the target node and the enterprise operation dynamic diagram. Neighboring nodes The set of all nodes that have an association relationship, that is, the second / first associated node set, where an association relationship means that there is a direct or indirect connection relationship with the current node; It represents the union of two sets of related nodes, reflecting the total number of neighboring nodes that can be covered by both sides of the two nodes; This represents the intersection of two sets of associated nodes, that is, the number of neighboring nodes that the two nodes are connected to; their difference is... This indicates the number of nodes that can be additionally covered rather than repeatedly covered on both sides of the currently analyzed associated edge (the edge to be analyzed). The larger the difference, the lower the degree of overlap in the neighborhood structure, and the more obvious the difference in the set of nodes connected by each side. This indicates that the associated edge connects two relatively independent local regions with stronger cross-structure information transmission capabilities and potential risk propagation and expansion capabilities. Conversely, it indicates that its neighborhood structure is highly overlapping and mainly plays a role in information interaction and propagation within the local region. This represents the total number of nodes in the enterprise operation dynamic diagram. Its role in the denominator is to normalize the overall propagation spread.

[0067] Furthermore, it is possible to further configure and determine the propagation hub dependency of the edge to be analyzed, including:

[0068] The average degree of each associated node in the first associated node set is denoted as the first degree mean, and the average degree of each associated node in the second associated node set is denoted as the second degree mean. The ratio of the average of the first degree mean and the second degree mean to the total number of nodes in the enterprise operation dynamic graph is denoted as the propagation hub dependency of the edge to be analyzed.

[0069] The formulaic representation of the propagation hub dependency of the edge to be analyzed is as follows:

[0070]

[0071] In the formula, Represents the target node Its neighboring nodes The propagation hub dependency of the corresponding associated edges (edges to be analyzed) is used to characterize the target node. Its neighboring nodes The degree to which the associated edges rely on "high-connectivity nodes (hub nodes)" during the propagation of the graph structure. The larger the result, the higher the overall connectivity of the neighboring nodes connected by the associated edge, and the more the risk propagation process relies on a few high-connectivity nodes for transit and diffusion, exhibiting strong centralized propagation characteristics; conversely, it indicates that the connection structure of the neighboring nodes is more dispersed, the risk propagation path does not rely on obvious hub nodes, and the propagation process tends to spread more evenly.

[0072] in, Represents the target node The average degree of the neighborhood of the target node. The average degree of each node in the set of associated nodes is used to reflect the target node. The average number of connections each neighboring node has in the neighborhood, i.e., the connection density level of the local network structure of that node; Represents the target node The set of associated nodes in the neighborhood The number of nodes in Represents the set of associated nodes Middle node The degree, where degree is an existing metric representing the degree of a node. The number of directly connected nodes and the connectivity in a graph structure. Represents the target node The set of associated nodes in the neighborhood The sum of the connectivity of all neighboring nodes in the middle; Similarly, among them Representing neighboring nodes The set of associated nodes in the neighborhood The number of nodes in Represents the set of associated nodes Middle node The higher the average degree of a node, the more active the neighboring nodes are as a whole, indicating that the area has a high-density network structure and is more likely to form multi-path information propagation and risk diffusion effects. Conversely, a lower average degree indicates that the neighboring nodes are sparsely connected, the local structure is relatively loose, and the risk propagation ability is weak.

[0073] in, Characterize the target node Its neighboring nodes The overall connectivity activity of the local structure reflects the structural density level of the network regions on both sides of the associated edge (the edge to be analyzed). This indicates that the average of them is calculated. This represents the total number of nodes in the enterprise operation dynamic diagram. Its role in the denominator is to normalize the overall dependence on the propagation hub.

[0074] S103, determine the propagation permission of the edge to be analyzed based on the neighborhood propagation stability, the propagation expansion and the propagation hub dependence, and determine the effective value propagation edge and the corresponding risk propagation weight in the associated edges formed between the target node and each neighboring node based on the propagation permission.

[0075] After obtaining the neighborhood propagation stability, propagation expansion, and propagation hub dependency of the edge to be analyzed, the propagation permission of the edge to be analyzed can be obtained by fusing the three factors:

[0076]

[0077] In the formula, Represents the target node Its neighboring nodes The propagation permission of the corresponding associated edges (edges to be analyzed) reflects the overall effectiveness of the associated edges in the risk propagation process. The larger the output value, the stronger the target node. Its neighboring nodes The corresponding edges between them can not only maintain a stable propagation state in the long term, but the nodes connected by the edges also have a wider structural coverage and stronger structural support for risk diffusion in the network, indicating that the edges have more value in participating in risk propagation analysis.

[0078] This allows us to obtain the propagation permission degree of the corresponding edges between the target node and each of its neighboring nodes, and based on the propagation permission degree, remove low-value propagation edge relationships of the target node. Specifically, this can be achieved by setting a propagation permission degree threshold. ,when If the risk propagation value corresponding to a certain associated edge is deemed insufficient to form a risk propagation path with actual influence, it is removed from the propagation analysis neighborhood of the target node. The remaining associated edges then become the effective value propagation edges of the target node. The settings can be customized according to actual needs; the preferred settings in this embodiment are as follows. .

[0079] For the remaining edge relationships with high risk propagation value, they can be denoted as the set M of effective value propagation edges for the target node, and the risk propagation weight of each effective value propagation edge in the set can be further calculated:

[0080]

[0081] In the formula, Represents the target node Its neighboring nodes The risk propagation weight of the corresponding effective value propagation edge, Represents the target node Its neighboring nodes The propagation permission of the effective value propagation edge corresponding to the two. Represents the target node Its neighboring nodes The propagation permission of the effective value propagation edge corresponding to the two.

[0082] At this point, the effective value propagation edges and their corresponding risk propagation weights within the neighborhood of the target node can be determined.

[0083] S104, complete the risk supervision of the target node based on the effective value propagation edge and the corresponding risk propagation weight.

[0084] After obtaining the effective value propagation edges of the target node and acquiring the risk propagation weights of the effective value propagation edges, the interaction information of the neighboring nodes can be obtained based on the set of effective value propagation edges, and the target node's information can be aggregated based on the risk propagation weights to obtain aggregated information.

[0085] The acquired aggregated information is input into the TGN memory update module to update the historical state of the target node, resulting in the updated node memory state. The low-dimensional representation vector of the target node is obtained by embedding the mapping function, and the risk propagation intensity between nodes is calculated by the propagation probability function. This is used to characterize the possibility and degree of risk propagation between nodes, thereby obtaining the node-level risk assessment result and the system-level risk diffusion degree.

[0086] At this point, risk monitoring of the target node can be completed.

[0087] The embodiments of the present invention effectively improve the problem of neighborhood equal-weight aggregation in the traditional TGN in the risk propagation analysis of enterprise-related networks, and can significantly improve the accuracy and credibility of enterprise operational data risk propagation analysis and regulatory output.

[0088] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An AI-based enterprise operation big data intelligent supervision platform, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it performs the following steps: Acquire enterprise operation data and construct an enterprise operation dynamic graph. Take any enterprise node in the enterprise operation dynamic graph as the target node, and take any of the associated edges formed between the target node and each neighboring node as the edge to be analyzed. Determine the neighborhood propagation stability of the edge to be analyzed based on the time interval and time span of the interaction events corresponding to the edge to be analyzed. Based on the difference between the union and intersection of the sets of associated nodes corresponding to the two nodes of the edge to be analyzed, the propagation extension degree of the edge to be analyzed is determined; based on the degree of each node in the set of associated nodes corresponding to the two nodes of the edge to be analyzed, the propagation hub dependency degree of the edge to be analyzed is determined. The propagation permission of the edge to be analyzed is determined based on the neighborhood propagation stability, the propagation expansion, and the propagation hub dependency. The effective value propagation edge and the corresponding risk propagation weight in the associated edges formed between the target node and each neighboring node are determined based on the propagation permission. Risk monitoring of the target node is completed based on the effective value propagation edge and the corresponding risk propagation weight.

2. The intelligent supervision platform for enterprise operation big data based on artificial intelligence according to claim 1, characterized in that, Determining the neighborhood propagation stability of the edge to be analyzed includes: The time interval between any two interaction events in the interaction events corresponding to the edge to be analyzed is recorded as the adjacent interaction interval. The absolute value of the difference between any adjacent interaction interval and the mean of all adjacent interaction intervals is recorded as the interaction discrete value of any adjacent interaction interval. The normalized value of the mean of the interaction discrete values ​​of all adjacent interaction intervals is recorded as the first neighborhood propagation stability characterization value. The normalized value of the ratio of the time span of the interaction event corresponding to the edge to be analyzed to the mean of the time span of the interaction events corresponding to all related edges on the enterprise operation dynamic graph is denoted as the second neighborhood propagation stability characterization value. The average of the first neighborhood propagation stability characterization value and the second neighborhood propagation stability characterization value is denoted as the neighborhood propagation stability of the edge to be analyzed.

3. The intelligent supervision platform for enterprise operation big data based on artificial intelligence according to claim 1, characterized in that, Determining the propagation extent of the edge to be analyzed includes: The two nodes corresponding to the edge to be analyzed are respectively denoted as the current first node and the current second node. Each node that is associated with the current first node forms a first set of associated nodes, and each node that is associated with the current second node forms a second set of associated nodes. The existence of an association relationship means that there is a direct or indirect connection relationship with the current first node or the current second node. The union of the first set of associated nodes and the second set of associated nodes is calculated and recorded as the first set calculation result. The intersection of the first set of associated nodes and the second set of associated nodes is calculated and recorded as the second set calculation result. The ratio of the difference between the first set calculation result and the second set calculation result to the total number of nodes in the enterprise operation dynamic graph is recorded as the propagation expansion degree of the edge to be analyzed.

4. The intelligent supervision platform for enterprise operation big data based on artificial intelligence according to claim 3, characterized in that, Determining the propagation hub dependency of the edge to be analyzed includes: The average degree of each associated node in the first associated node set is denoted as the first degree mean, and the average degree of each associated node in the second associated node set is denoted as the second degree mean. The ratio of the average of the first degree mean and the second degree mean to the total number of nodes in the enterprise operation dynamic graph is denoted as the propagation hub dependency of the edge to be analyzed.

5. The intelligent supervision platform for enterprise operation big data based on artificial intelligence according to any one of claims 1 to 4, characterized in that, Determining the propagation permission of the edge to be analyzed includes: The product of the mean of the propagation spread and the propagation hub dependency and the neighborhood propagation stability is denoted as the propagation permission of the edge to be analyzed.

6. The intelligent supervision platform for enterprise operation big data based on artificial intelligence according to claim 1, characterized in that, Determine the effective value propagation edges and corresponding risk propagation weights in the association edges formed between the target node and each neighboring node, including: The effective value propagation edge is defined as the association edge whose propagation permission is greater than a preset propagation permission threshold among the association edges formed between the target node and each neighboring node. The ratio of the propagation permission degree corresponding to any effective value propagation edge to the sum of the propagation permission degrees corresponding to all effective value propagation edges is denoted as the risk propagation weight of any effective value propagation edge.

Images