Risk diffusion path prediction method and related equipment

Abstract

The invention provides a risk diffusion path prediction method and related equipment, and relates to the field of big data. The method comprises the following steps: determining multiple types of transaction data confidence degrees of multiple transaction subjects, and determining a target transaction subject from the multiple transaction subjects according to the transaction data confidence degrees; generating a directed topology network according to the target transaction subject, and updating the directed topology network according to the transaction data of the target transaction subject; performing iteration on the transaction data confidence of the nodes according to the edge weights to obtain risk potential energy of each node; and determining a risk diffusion path by taking the node of which the risk potential energy is greater than or equal to a preset risk potential energy threshold value and / or the node of which the risk potential energy amplification is greater than or equal to a preset amplification threshold value as an initial node. The directed topology network can be dynamically updated, the risk diffusion path can be identified based on the updated network, and the accuracy of the risk diffusion path can be improved.

Description

Technical Field

[0001] The present application relates to the field of big data, and in particular to a risk diffusion path prediction method and related equipment. Background Art

[0002] In the field of financial technology, the risk diffusion path can describe the transaction entities that will be affected in sequence when there is a transaction risk. Identifying the risk diffusion path plays an important role.

[0003] In related technologies, risk diffusion paths can be implemented based on a topological network. A topological network consists of multiple nodes, each corresponding to a transaction entity. Nodes are connected by directed edges, representing the transaction relationships between transaction entities. Risk diffusion paths can then be analyzed based on these directed edges.

[0004] However, the above solution has the problem of low recognition accuracy. Summary of the Invention

[0005] The present application provides a risk diffusion path prediction method and related equipment to solve the technical problem of low accuracy of risk diffusion paths.

[0006] In a first aspect, the present application provides a method for predicting risk diffusion paths, comprising:

[0007] determining confidence levels of multiple types of transaction data of multiple transaction entities, and determining a target transaction entity from the multiple transaction entities based on the transaction data confidence levels;

[0008] Generating a directed topological network based on the target transaction subject, and updating the directed topological network based on the transaction data of the target transaction subject, wherein the directed topological network includes nodes and edge weights, wherein the nodes correspond to the target transaction subject, and the edge weights are related to the transaction volumes of the two nodes;

[0009] Iterating the transaction data confidence of the node according to the edge weight to obtain the risk potential of each node;

[0010] The node whose risk potential is greater than or equal to the preset risk potential threshold, and / or the node whose risk potential increase is greater than or equal to the preset increase threshold, is used as the starting node to determine the risk diffusion path. The end node of the risk diffusion path is the node whose similarity with the starting node is greater than or equal to the preset similarity threshold and whose risk potential is less than or equal to the preset risk potential threshold.

[0011] In a second aspect, the present application provides a risk diffusion path prediction device, comprising:

[0012] a confidence determination module, configured to determine the confidence levels of multiple types of transaction data of multiple transaction entities, and determine a target transaction entity from the multiple transaction entities based on the confidence levels of the transaction data;

[0013] a network update module, configured to generate a directed topology network based on the target transaction subject and update the directed topology network based on the transaction data of the target transaction subject, wherein the directed topology network includes nodes and edge weights, wherein the nodes correspond to the target transaction subject and the edge weights are related to the transaction volume of the two nodes;

[0014] a risk potential iteration module, configured to iterate the transaction data confidence of the node according to the edge weight to obtain the risk potential of each node;

[0015] The risk diffusion path identification module is used to use the node whose risk potential is greater than or equal to the preset risk potential threshold, and / or the node whose risk potential increase is greater than or equal to the preset increase threshold, as the starting node to determine the risk diffusion path. The end node of the risk diffusion path is the node whose similarity with the starting node is greater than or equal to the preset similarity threshold and whose risk potential is less than or equal to the preset risk potential threshold.

[0016] In a third aspect, the present application provides an electronic device, characterized in that it includes: a processor, and a memory communicatively connected to the processor;

[0017] The memory stores computer-executable instructions;

[0018] The processor executes the computer-executable instructions stored in the memory to implement the method of the first aspect.

[0019] In a fourth aspect, the present application provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to implement the method of the first aspect when executed by a processor.

[0020] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which implements the method of the first aspect when executed by a processor.

[0021] The present application provides a risk diffusion path prediction method and related equipment. The method includes: determining the confidence of multiple types of transaction data of multiple transaction entities, and determining the target transaction entity from the multiple transaction entities based on the transaction data confidence; generating a directed topological network based on the target transaction entity, and updating the directed topological network based on the transaction data of the target transaction entity, the directed topological network includes: nodes and edge weights, the nodes correspond to the target transaction entities, and the edge weights are related to the transaction volume of the two nodes; iterating the transaction data confidence of the nodes based on the edge weights to obtain the risk potential of each node; using nodes with risk potential greater than or equal to a preset risk potential threshold, and / or nodes with risk potential increase greater than or equal to a preset increase threshold as starting nodes to determine the risk diffusion path, and the end node of the risk diffusion path is a node with a similarity with the starting node greater than or equal to a preset similarity threshold, and a risk potential less than or equal to a preset risk potential threshold. The present application can dynamically update the directed topological network to identify the risk diffusion path based on the updated network, which can improve the accuracy of the risk diffusion path. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0023] Figure 1 A schematic diagram of the structure of a directed topology network provided in this application;

[0024] Figure 2 A flowchart of the steps of a risk diffusion path prediction method provided in this application;

[0025] Figure 3 This is a structural block diagram of a risk diffusion path prediction device provided by this application;

[0026] Figure 4 It is a structural diagram of the electronic device provided in this application.

[0027] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0028] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0029] It should be noted that the risk diffusion path prediction method and related equipment provided in this application can be used in the field of artificial intelligence, and can also be used in any field other than artificial intelligence. The application field of the risk diffusion path prediction method and related equipment in this application is not limited.

[0030] The application scenario of this application is to predict the risk diffusion path based on a directed topological network. Figure 1 This is a schematic diagram of a directed topology network provided by this application. Figure 1 As shown, the directed topology network includes eight nodes, N1 to N8. Node N1 is directionally connected to nodes N2 and N4, node N2 is directionally connected to node N3, node N4 is directionally connected to nodes N3 and N6, node N5 is directionally connected to N3, node N7 is directionally connected to N5, and node N8 is directionally connected to nodes N3 and N6. It should be noted that in practice, directed topology networks are more complex, with many more nodes and edges. This is just a simple example.

[0031] based on Figure 1 In the directed topology network shown, a possible risk diffusion path is exemplified as N1->N4->N6.

[0032] In related technologies, the structure of the directed topology network cannot be updated in real time and is fixed. Therefore, predicting the risk diffusion path based on the fixed directed topology network may result in poor prediction accuracy.

[0033] The risk diffusion path prediction method and device provided in this application are designed to address the aforementioned technical issues in the prior art. To more accurately predict risk diffusion paths, this application considers dynamically constructing and updating a directed topological network based on the confidence level of transaction data, enabling it to more accurately reflect current transaction conditions. This allows for more accurate prediction of risk diffusion paths.

[0034] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0035] Figure 2 This is a flowchart of a risk diffusion path prediction method provided by this application, refer to Figure 2 As shown, the above-mentioned risk diffusion path prediction method includes the following steps:

[0036] S201: Determine confidence levels of multiple types of transaction data of multiple transaction entities, and determine a target transaction entity from the multiple transaction entities based on the transaction data confidence levels.

[0037] In some possible implementations, determining the confidence of multiple types of transaction data for multiple transaction entities may include: first, obtaining a transaction data sequence of each type for the transaction entity, and performing a sliding mean calculation on the transaction data sequence based on a preset sliding mean weight; then, scaling the mean residual between the sliding mean and the historical mean using an adaptive scaling parameter to obtain the transaction data confidence; and finally, dynamically adjusting the sliding mean weight based on the transaction data confidence and the mean residual. The present application can calculate the transaction data confidence by using a sliding mean, thereby ensuring the smoothness of the transaction data confidence and thereby improving its accuracy.

[0038] The type of transaction data indicates its source. Multiple types include, but are not limited to, news streams, financial report streams, social media text streams, and high-frequency trading streams. Non-blocking monitoring channels can be established for each of these data stream types to monitor transaction data. Transaction data can also be understood as transaction events, including but not limited to the transaction initiator, counterparty, and transaction volume. When monitoring transaction data, the data source identifier, source timestamp, and transaction data can be encapsulated into a transaction data object for writing to a ring buffer. The data source identifier indicates the type of transaction data, and the source timestamp indicates the time when the transaction data was detected.

[0039] In some possible implementations, the data may be written into the ring buffer through a compare-and-swap spin write operation, thereby improving the write speed compared to a normal write operation.

[0040] After writing to the aforementioned ring buffer, the transaction data object needs to be read from the ring buffer to preprocess it. This allows writing and reading to be performed independently without affecting each other. Specifically, the data source identifier, source timestamp, and transaction data can be extracted from the read transaction data object. The time zone offset and sampling granularity identifier can then be extracted from the transaction data header. The source timestamp can then be converted to a standard timestamp using the time zone offset and sampling granularity identifier, allowing the transaction data to be sorted based on the standard timestamp. For example, the standard timestamp can be a UTC (Coordinated Universal Time) timestamp.

[0041] In some possible implementations, time conversion can be performed according to the following formula:

[0042]

[0043] Among them, τ i Is a standard timestamp in milliseconds. Is the source timestamp, o i is the time zone offset, g i The sampling granularity identifier can be 0, 3, or 6, representing seconds, milliseconds, and microseconds, respectively.

[0044] After performing the aforementioned time conversion, each transaction data object can be sorted according to the standard timestamp to obtain a transaction data object sequence. The transaction data objects in the transaction data object sequence can then be parsed sequentially. Specifically, transaction data can be extracted from the transaction data objects, and the transaction content can be extracted from the transaction data. A lightweight speech parser can then be used to perform lemmatization, dependency segmentation, and part-of-speech filtering on the transaction content. Financial entity lemmas with a word frequency greater than or equal to a preset word frequency threshold are retained to obtain a transaction triple, including a transaction initiator lemma, a counterparty lemma, and a transaction code.

[0045] After obtaining the above-mentioned transaction triples, the transaction initiating subject identifier and the transaction counterparty subject identifier can be determined for the transaction initiating subject term and the transaction counterparty subject term respectively. Taking the transaction initiating subject term as an example, the transaction initiating subject term and each candidate subject information in the real-time alias table can be matched to obtain industry similarity, regional fit, historical co-occurrence times, etc. These three can form a subject similarity vector, and the matching degree between the transaction initiating subject and the candidate subject can be calculated based on the subject similarity vector. When the matching degree is greater than or equal to the preset threshold, the subject identifier corresponding to the candidate subject in the real-time alias table can be used as the transaction initiating subject identifier. The similarity can adopt any vector similarity. For example, the matching degree can refer to the following formula:

[0046]

[0047] Among them, γ jk is the matching degree between the jth transaction initiator and the kth candidate subject, W is the subject mapping matrix, F jk is a vector composed of the industry similarity, regional fit, and historical co-occurrence count between the j-th transaction initiator and the k-th candidate entity. F jk Transpose, F jk′is a vector composed of the industry similarity, regional fit, and historical co-occurrence count between the j-th transaction initiator and the k'th candidate entity. F jk′ The transpose of .

[0048] The above real-time alias table is updated in real time. When a merger, renaming or spin-off is detected, the above update process can be recorded in the above real-time alias table in the form of inheritance chain. j Updated to Then build an inheritance chain in the real-time alias table This ensures continuous traceability.

[0049] To further ensure the accuracy of transaction data, key-value pairs can be generated for transaction data. Specifically, after hashing the transaction initiator identifier of the transaction data, perform an XOR operation with the type corresponding to the transaction data to use as the key. Then, the standard timestamp, the type of transaction data, and the transaction data are used as the value. Finally, the key-value pair is written to the partition queue and a sliding window check is performed to determine whether there is an unknown subject or a subject conflict. If so, the transaction data is pushed to the manual correction pool for manual correction. Of course, after the correction is completed, the real-time alias table and the subject mapping matrix W used in the aforementioned matching calculation can also be updated according to the correction results. In this way, the accuracy of the subject mapping can be further improved.

[0050] After the mapping is complete, the transaction data can be divided according to the transaction initiator identifiers, with each transaction data being stored in the corresponding ring buffer. Different ring buffers correspond to different transaction initiators, and their order is sorted according to the standard timestamps, thereby forming multiple types of transaction data sequences for each transaction initiator. A sliding mean calculation can then be performed on these transaction data sequences using pre-set sliding mean weights.

[0051] Specifically, first, for each transaction data in the ring buffer of the transaction initiator, multiple time identifiers can be calculated according to the standard timestamp, including but not limited to: minute identifier, hour identifier and natural day identifier, and the time identifier is written into the header information of the transaction data. Among them, the minute identifier m i Can be That is, for τ i / Round down to 60,000, hour symbol h i Can be Nature Day Logo i Can be

[0052] After obtaining the above-mentioned time stamp, the confidence level of the transaction data can be generated based on the above-mentioned time stamp. Specifically, at the end of the natural day, first, for each type of transaction data sequence of each transaction initiator, a sliding mean with learnable weights is performed on it at the minute level; then, the sliding mean of the minute level is aggregated through a gated aggregator to obtain the sliding mean of the hour level, and further aggregation is performed based on the sliding mean of the hour level to obtain the sliding mean of the natural day level; finally, the difference between the sliding mean of the natural day level and the corresponding historical mean is calculated as the mean residual, and the mean residual is scaled to a type of transaction data confidence level of the transaction initiator through an adaptive scaling parameter. The confidence levels of multiple types of transaction data of the transaction initiator constitute the confidence vector of the transaction initiator.

[0053] The confidence level of the above transaction data can be obtained by mapping it with the following formula:

[0054]

[0055] Among them, c e,m is the confidence level of the transaction data of the mth type of the transaction initiator e, and its value ranges from 0 to 1. σ(·) is a smoothing limit function, α m ,β m is the modal adaptive scaling parameter, r e,m is the residual between the sliding mean and the historical mean of the transaction data of the mth type of transaction initiator e at the natural day level, represents element-wise division, σ m is the historical standard deviation of the mth type of transaction data.

[0056] After obtaining the transaction data confidence level of the transaction initiator, the sliding mean weight can be dynamically adjusted based on the transaction data confidence level and the mean residual. 8. The method according to claim 7, wherein dynamically adjusting the adaptive scaling parameter based on the transaction data confidence level and the mean residual includes:

[0057] When the transaction data sequence is of a social type, the transaction data is a social data stream. If the transaction data confidence is less than a first confidence threshold and the mean residual is greater than or equal to the first residual threshold, the sliding mean weight of the transaction data sequence is reduced. When the transaction data sequence is of a financial report type, the transaction data is a financial report data stream. If the transaction data confidence is greater than a second confidence threshold and the mean residual is greater than or equal to the second residual threshold, the sliding mean weight of the transaction data sequence is increased, but the saturation value of the sliding mean is increased to avoid excessive influence from the financial report data stream. In this way, the sliding mean can be dynamically changed, improving its accuracy.

[0058] After the confidence vector is determined, the sliding mean of multiple types of natural day layers is The confidence vectors are concatenated with the confidence vectors by date index to form a fixed column structure. The mean confidence vectors of all transaction initiators form the mean confidence tensor. Missing positions are filled with zeros to ensure stable tensor sparsity. Confidence labels derived from minute noise ratio and hourly stability are also appended and output to downstream models.

[0059] After completing a prediction, the downstream model returns the deviation vector, updates the aforementioned first confidence threshold and second confidence threshold and modal adaptive scaling parameters, thereby continuously weakening the influence of the noise source within the natural day layer closed loop and forming an adaptive noise suppression mechanism.

[0060] After obtaining the above-mentioned mean confidence tensor, the target transaction subject can be extracted according to the above-mentioned method. Specifically, when the confidence of the transaction data in the mean confidence tensor is greater than or equal to the active threshold, the corresponding transaction initiator can be determined as the target transaction subject.

[0061] S202: Generate a directed topology network based on the target transaction subject, and update the directed topology network based on the transaction data of the target transaction subject. The directed topology network includes: nodes and edge weights. The nodes correspond to the target transaction subject, and the edge weights are related to the transaction volume of the two nodes.

[0062] It is understood that updating a directed topology network includes updating nodes and updating edge weights. Updating nodes includes adding nodes, deleting nodes, splitting nodes, and merging nodes, and updating edge weights includes increasing edge weights and decreasing edge weights.

[0063] In some possible implementations, updating the directed topology network according to the transaction data of the target transaction subject includes:

[0064] If the transaction data includes transaction volume, transaction initiator, and transaction counterparty, the edge weight is updated based on the transaction volume, transaction initiator, and transaction counterparty. For example, an edge can be established from the transaction initiator to the transaction counterparty, and the transaction volume can be added to the edge weight of the edge to record the short-term transaction aggregation effect. When multiple transaction data from the same transaction initiator to the same transaction counterparty appear consecutively, the transaction volume of the multiple transaction data can be locally accumulated first, and the accumulated transaction volume can be written to the edge weight. In this way, the number of times the directed topology network is read and written can be minimized, only one write lock is required, and write diffusion is reduced.

[0065] If the transaction data includes the original transaction subject, the split subject, and the split ratio, the node corresponding to the original transaction subject is split into the nodes of the split subject, and the edge weight of the original transaction subject is split into the edge weights of the split subject according to the split ratio. It is understood that the news data stream may generally include the original transaction subject, the split subject, and the split ratio, so that text matching can be performed on the transaction data to identify whether there is a split scenario.

[0066] If the transaction data includes the original transaction subject and the merged subject, the nodes of the original transaction subject are merged into the merged subject, and the edge weights of the original transaction subject are merged.

[0067] If the transaction data of the target transaction subject corresponds to an overdue event, the edge weight of the target transaction subject is reduced, and the target transaction subject is set as the starting node of the risk diffusion path. In this way, the prediction speed of the risk diffusion path can be further improved.

[0068] This application can combine multiple events to update the network, so that the network update covers as many change scenarios as possible, which helps to improve the accuracy of the network and thus improve the accuracy of the risk diffusion path.

[0069] In summary, this application can update nodes, update edge weights, split nodes, split edge weights, merge nodes, and merge edge weights, integrate multiple scenarios to realize directed topology networks, and ensure the accuracy of node and edge weights.

[0070] S203: Iterate the transaction data confidence of the node according to the edge weight to obtain the risk potential of each node.

[0071] In some implementations, the edge curvature of the two nodes is first determined based on their target edge weights and the similarity between the two nodes. Then, the confidence level of the transaction data for the node is used as the initial residual vector for the node, and the residual vector for the node is iteratively updated based on the initial residual vector according to the edge curvature. Finally, the initial risk potential of the node is determined based on the residual vector, and the risk potential of each node is iteratively updated based on the initial risk potential. The risk potential can be determined by updating the residual vector based on the edge curvature, further improving the accuracy of the risk potential.

[0072] The target edge weight includes the total outgoing edge weight of the transaction initiating node and the total incoming edge weight of the counterparty node. The total outgoing edge weight is the sum of the edge weights of the edges where the node is the starting node. For example, Figure 1As shown, the total outbound edge weight of the transaction initiation node corresponding to node N4 is the sum of the edge weights from node N4 to N3 and the edge weights from node N4 to node N6. The total inbound edge weight is the edge weight of the edge with the node as the end node, for example, refer to Figure 1 As shown, the total incoming edge weight of node N3 is the sum of the edge weight from N4 to N3, the edge weight from nodes N2 to N3, the edge weight from nodes N5 to N3, and the edge weight from nodes N8 to N3.

[0073] The similarity of a node can be the similarity of the node's attributes, which is used to indicate the similarity between the attributes of the transaction entities corresponding to the nodes. The similarity can be calculated by representing the attributes of each dimension of the transaction entity as a vector.

[0074] After obtaining the node similarity and edge weights, for each edge, the edge curvature can be calculated based on the similarity of the two nodes connected by the edge and the edge weight of the edge. The edge curvature is used to indicate the structural similarity and transaction volume concentration of the two nodes. The edge curvature is positively correlated with the target edge weight and negatively correlated with the similarity. Specifically, the edge curvature can be calculated using the following formula:

[0075]

[0076] Among them, κ u,v (d) is the edge curvature of nodes u and v on day d, is the total outgoing edge weight of node u, is the total incoming edge weight of node v, cosθ u,v (d) is the similarity between two nodes u and v.

[0077] After obtaining the above-mentioned edge curvature, the above-mentioned edge curvature can be updated to the residual vector of each node to determine the initial potential energy based on the residual vector. Generally, the larger the residual vector, the greater the initial potential energy, which means that the probability of the node being the starting point of the risk is greater.

[0078] In some embodiments, the attenuation coefficient can be further adjusted according to the edge curvature to improve the accuracy of the attenuation coefficient. Compared with a fixed attenuation coefficient, the present application can adjust the attenuation coefficient more dynamically to improve the prediction accuracy. Specifically, when the edge curvature is greater than a preset edge curvature threshold, the attenuation coefficients of the two nodes are updated according to the difference between the edge curvature of the two nodes and the preset edge curvature threshold, so as to iteratively update the residual vector of the node based on the initial residual vector according to the edge curvature and the attenuation coefficient. The iterative update of the residual vector in combination with the dynamic attenuation coefficient can further improve the accuracy of the residual vector.

[0079] The attenuation coefficient is used to indicate the degree of adjustment of the residual when diffusing along the corresponding edge. The attenuation coefficient is positively correlated with the difference between the edge curvature of the two nodes and the preset edge curvature threshold. Specifically, the attenuation coefficient can be updated using the following formula:

[0080]

[0081] Among them, φ u,v (d) is the decay coefficient of the edge between nodes u and v on the dth day, φ u,v (d-1) is the attenuation coefficient of the edge between nodes u and v on day d-1, λ is the sensitivity parameter, κ thr is the preset edge curvature threshold, κ max is the maximum value of the edge curvature. 1{.} is an indicator function used to determine whether the edge curvature is greater than the preset edge curvature threshold. When the edge curvature is greater than the preset edge curvature threshold, the output is 1, and when the edge curvature is less than or equal to the preset edge curvature threshold, the output is 0.

[0082] Furthermore, the similarity between the node feature and the target label of the node can also be calculated as the feature label resonance of the node; then, for each edge, when the feature label resonance of the two nodes of the edge is greater than or equal to a preset resonance threshold, the edge attribute of the edge is generated based on the feature label resonance of the two nodes of the edge; finally, the residual vector of the node is iteratively updated based on the initial residual vector according to the edge attribute, the edge curvature, and the attenuation coefficient. Based on this, the initial risk potential energy of the node determined according to the residual vector includes: determining the initial risk potential energy of the node according to the residual vector and the edge attribute. Iterating the residual vector in combination with the edge attribute can further improve the accuracy of the residual vector and the risk potential energy.

[0083] The target tag indicates the occurrence of a preset event, for example, a transaction anomaly. The edge attribute can be z u,v =1{ρ u >ρ c}(h u ⊙y u )+1{ρ v >ρ c}(h v ⊙y v ), where 1{ρ u >ρ c} is used to determine whether the characteristic label resonance degree of node u is greater than the preset resonance degree threshold. If it is greater, it outputs 1; if it is less than or equal to, it outputs 0. u ⊙y u is the element-wise multiplication result of the vectors corresponding to the node features of node u and the target label of node u, 1{ρ v >ρc} is used to determine whether the resonance degree of the characteristic label of node v is greater than the preset resonance degree threshold. If it is greater, it outputs 1; if it is less than or equal to, it outputs 0. u ⊙y u It is the element-wise multiplication result between the node feature of node v and the target label of node v.

[0084] Among them, the residual vector and edge attributes can be spliced ​​into a joint matrix to cover all risk characteristics.

[0085] The update of the residual vector can be expressed as follows:

[0086]

[0087] in, is the updated residual vector, is the residual vector before updating, η is the updating step size of the residual vector, Represents the set of adjacent nodes of node n. n,v is the attenuation coefficient of the edge between nodes n and v, κ n,v is the edge curvature of nodes n and v, z n,v is the edge attribute of nodes n and v. Iterate until When , it is considered to be converged and ε is a very small value.

[0088] Before iterating the residual vector, you can also retrieve the transaction entity e corresponding to the node j The historical variance vector of and historical average Calculate the gain term. Identify significant dimensions through the gating function, and immediately dilute the low-gain dimensions. The specific gain-gating threshold is jointly calculated as:

[0089]

[0090] in, For the transaction subject j The gain-gating threshold is calculated jointly, σ(·) is the Sigmoid gating function, α is the temperature coefficient, which is used to control the steepness of Sigmoid, and diag(σ) represents the vector Put it into the diagonal matrix, τ1 is the significant threshold, ⊙ is the Hadamard multiplication, and Element-level preservation or attenuation of node vectors is performed to prevent low-quality noise from penetrating into the structural layer.

[0091] In some implementations, a damping accumulator can be deployed on the node side to address typical pulses, such as redemptions and convertible bond price increases. The accumulator first performs a vector summation of the instantaneous potential energies corresponding to multiple messages arriving in the same time slot. If the modulus of the instantaneous potential energies is greater than the node damping coefficient, the instantaneous potential energy is scaled proportionally to prevent extreme quotes from causing local gradient collapse.

[0092] In some embodiments, the iterative update based on the initial risk potential to obtain the risk potential of each node includes: first, calculating the inertia coefficients of the two nodes based on the historical risk potentials corresponding to the two nodes when conducting the most recent transaction, and the average historical risk potential of the two nodes, and determining the propagation multiplier of the two nodes based on the baseline propagation multiplier, the inertia coefficient, and the transaction confidence, wherein the inertia coefficient is positively correlated with the two historical risk potentials and negatively correlated with the average historical risk potential, and the propagation multiplier is positively correlated with the baseline propagation multiplier, the inertia coefficient, and the transaction confidence. Then, for any node, the risk potential of the node is updated based on the propagation multiplier between the node's adjacent entry node and the node, the risk potential of the adjacent entry node at the previous time, and the time of the most recent transaction from the adjacent entry node to the node, wherein the adjacent entry node is the node that initiates a transaction to the node. Updating the risk potential in combination with the propagation multiplier can improve the accuracy of the risk potential.

[0093] The inertia coefficient can be calculated using the following formula:

[0094]

[0095] Among them, η u,v is the inertia coefficient of the edge between nodes u and v, ρ is a constant, is the average historical risk potential energy of node u, is the average historical risk potential energy of node v, is the historical risk potential energy corresponding to the most recent transaction of node u, is the historical risk potential energy corresponding to the most recent transaction of node u, is the historical risk potential energy corresponding to the node v when it made the most recent transaction.

[0096] After obtaining the above inertia coefficient, it can be written into the edge header information. In this way, the jump caused by the hedge fund's large-scale position adjustment will amplify η u,v , and when it is stable η u,v ≈1.

[0097] The propagation rate can be calculated using the following formula:

[0098]

[0099] Among them, λ u,v Represents the propagation multiplier of the edge between node u and node v, which is used to indicate the transmission speed or strength factor. is the baseline propagation rate, η u,v is the inertia coefficient of the edge between nodes u and v, α c is the coefficient, c u,v is the transaction confidence of nodes u and v.

[0100] The risk potential of a node can be updated by the following formula:

[0101]

[0102] Among them, p e (t) is the risk potential energy of node e after updating at time t, δ e represents the node self-damping coefficient; ι g is the cross-partition gain coefficient; is the set of adjacent inbound nodes of node e, is the node that initiates transactions to node e, that is, the starting node of the edge connected with node e as the end node; seg(u) and seg(e) are the market partitions to which nodes u and e belong respectively; Δt is the propagation window length, λ u,e is the propagation rate of the edge between node u and node e, p u (t - ) is the node potential energy of node u before the update at time t, τ u,e is the last transaction time between nodes u and e, ζ is the time decay length, V d is the node set in the directed topology network, p k (t - ) is the node potential energy of node k before the update at time t, ∈ is the anti-zero drift constant; P tot Conservation constants defined for the system.

[0103] The above risk potential formula simultaneously completes cross-market alignment, time decay and total normalization in a single convolution-integration operation, ensuring steady-state convergence and avoiding the dissipation of financial potential by high-frequency oscillations in snapshot updates.

[0104] In some implementations, node potential, node self-damping coefficient, and propagation multiplier can also be written to the version cache. If the monitoring module finds that the risk potential variance of all nodes in the directed topology network exceeds the trading circuit breaker threshold, the system automatically shortens the time decay length ζ and restarts the propagation cycle to re-update the risk potential, achieving a multi-period self-stabilizing closed loop across the market from morning to afternoon to end of the day.

[0105] S204: A node whose risk potential is greater than or equal to a preset risk potential threshold, and / or a node whose risk potential increase is greater than or equal to a preset increase threshold, is used as a starting node to determine a risk diffusion path. The end node of the risk diffusion path is a node whose similarity with the starting node is greater than or equal to a preset similarity threshold and whose risk potential is less than or equal to the preset risk potential threshold.

[0106] The risk potential increase refers to the difference between the updated risk potential and the pre-update risk potential. After determining the end node of the risk diffusion path based on the risk potential, the starting node of the risk diffusion path can be determined. The risk diffusion path can include multiple nodes arranged in sequence.

[0107] The similarity between the end node and the start node may be the similarity between the industries of the two.

[0108] When identifying risk diffusion paths, a meta-learning catastrophic function can be invoked to load a small number of labeled risk chain samples, rapidly adapt the risk map to the kernel, and output a catastrophic response surface. This response surface can self-correct marginal risk elasticity across cycles and market fluctuations, providing a quantitative basis for edge-level latency and cost re-assignment. The risk map can include the risk potential of each node in a directed topological network.

[0109] Based on the disaster response surface, the latency cost and risk weight are recalculated for each edge. High-response areas compress latency and reduce cost to improve connectivity, while low-response areas extend latency and increase cost to mitigate error propagation, creating a dynamic metric.

[0110] A heuristic shortest path search is performed on dynamic metrics, and a step-by-step potential energy conversion heuristic function is used to accumulate node potential energy and travel time in real time to generate the first set of candidate paths with the highest reachability from the start node to the end node.

[0111] For each candidate path, the meta-learning catastrophic function is called to estimate the trigger probability, and the arrival time distribution is obtained by combining the edge delay. The system outputs the probability-time dual-index vector Z(π) at one time, which is calculated as follows:

[0112]

[0113] Among them, P(π) is the probability, is the time, L is the length of the candidate path; For node v l The disaster response coefficient, For node v l Updated risk potential; For node v l and v l+1 Dynamic risk weights for the edges; is the dynamic delay; v l and v l+1 The time decay scale of θ l,l+1 is the industry embedding angle between the lth node and the l+1th node; α, β, γ are the adaptive hyperparameters after meta-learning; σ(·) is the Sigmoid function.

[0114] according to Double sorting sorts candidate paths in descending order, creating a list of risk diffusion paths. This list is pushed to the monitoring panel in real time for on-duty personnel to review and is simultaneously written to the retrospective warehouse AR as a posteriori correction benchmark, completing a closed-loop prediction-verification process.

[0115] After obtaining the above risk diffusion path, it is parsed into a three-column structure of "path identifier, trigger probability, estimated arrival time" and pushed to the monitoring panel. The monitoring panel sorts the paths in descending order of probability and ascending order of remaining arrival time, so that on-duty personnel can quickly locate the most urgent risk chain during high-concurrency situations.

[0116] When the trigger probability is higher than the first-level threshold, the client pops up a window and triggers an audio-visual signal; when the probability is between the first and second-level thresholds, only a highlight label is added to the list row; the risk diffusion path below the second-level threshold is folded into the background monitoring sub-panel.

[0117] In addition, it is also possible to rely on unified entity coding rules to continuously capture exchange announcements, subject rating adjustments and multi-channel sudden public opinions, generate real risk trajectory streams through hash aggregation and write them into the posterior buffer according to the timeline, providing original facts for deviation analysis.

[0118] The trajectory aligner is called to match the actual risk trajectory with the same-day forecast list section by section, and the arrival time difference, trigger status and non-trigger reason of each path are output; the result is vectorized into deviation gradient and flows into the model self-correction module to become a quantitative feedback signal.

[0119] The self-correction module reads the deviation gradient and performs hierarchical processing: when the overall gradient is lower than the steady-state threshold ∈, the data is only archived; if the error is concentrated on the local edge, the relevant edge weights and inertia coefficients are dynamically fine-tuned; if the error covers multiple chains and spans multiple market segments, the lightweight retraining process of the link is directly triggered.

[0120] After lightweight retraining is complete, new weights are generated and connected to the real-time data stream in parallel as shadow instances. The scheduler continuously compares the output distribution of the shadow instances with that of the production instances. When the difference converges within the observation window, it automatically performs lossless stream switching, ensuring zero service interruption during the upgrade process.

[0121] Figure 3This is a structural block diagram of a risk diffusion path prediction device provided by this application, refer to Figure 3 As shown, the risk diffusion path prediction device 400 of the present application includes:

[0122] The confidence determination module 401 is configured to determine the confidence of multiple types of transaction data of multiple transaction entities, and determine a target transaction entity from the multiple transaction entities based on the confidence of the transaction data.

[0123] The network update module 402 is used to generate a directed topology network based on the target transaction entity and update the directed topology network based on the transaction data of the target transaction entity. The directed topology network includes: nodes and edge weights, the nodes correspond to the target transaction entity, and the edge weights are related to the transaction volume of the two nodes.

[0124] The risk potential iteration module 403 is configured to iterate the transaction data confidence of the nodes according to the edge weights to obtain the risk potential of each node.

[0125] The risk diffusion path identification module 404 is used to use the node whose risk potential is greater than or equal to the preset risk potential threshold, and / or the node whose risk potential increase is greater than or equal to the preset increase threshold, as the starting node to determine the risk diffusion path. The end node of the risk diffusion path is the node whose similarity with the starting node is greater than or equal to the preset similarity threshold and whose risk potential is less than or equal to the preset risk potential threshold.

[0126] In a possible implementation, the risk potential iteration module 403 is further configured to:

[0127] According to the target edge weights of the two nodes and the similarity of the two nodes, the edge curvature of the two nodes is determined, where the edge curvature is positively correlated with the target edge weight and negatively correlated with the similarity, and the target edge weight includes the total outgoing edge weight of the transaction initiating node and the total incoming edge weight of the transaction counterparty node; the transaction data confidence of the node is used as the initial residual vector of the node, and the residual vector of the node is iteratively updated based on the initial residual vector according to the edge curvature; the initial risk potential of the node is determined according to the residual vector, and the risk potential of each node is obtained by iteratively updating based on the initial risk potential.

[0128] In a possible implementation, the risk potential iteration module 403 is further configured to:

[0129] When the edge curvature is greater than a preset edge curvature threshold, the attenuation coefficients of the two nodes are updated according to the difference between the edge curvature of the two nodes and the preset edge curvature threshold. The attenuation coefficient is used to indicate the degree of adjustment of the residual when diffusing along the corresponding edge, and the attenuation coefficient is positively correlated with the difference. The residual vector of the node is iteratively updated based on the initial residual vector according to the edge curvature and the attenuation coefficient.

[0130] In a possible implementation, the risk potential iteration module 403 is further configured to:

[0131] Calculate the similarity between the node feature and the target label of the node as the feature label resonance of the node; for each edge, when the feature label resonance of the two nodes of the edge is greater than or equal to a preset resonance threshold, generate the edge attribute of the edge according to the feature label resonance of the two nodes of the edge; iteratively update the residual vector of the node based on the initial residual vector according to the edge attribute, the edge curvature, and the attenuation coefficient; determine the initial risk potential energy of the node according to the residual vector and the edge attribute.

[0132] In a possible implementation, the risk potential iteration module 403 is further configured to:

[0133] According to the historical risk potential energies corresponding to the two nodes when conducting the most recent transaction, and the average historical risk potential energies of the two nodes, the inertia coefficients of the two nodes are calculated, and the propagation multipliers of the two nodes are determined based on the baseline propagation multiplier, the inertia coefficient, and the transaction confidence. The inertia coefficient is positively correlated with the two historical risk potential energies and negatively correlated with the average historical risk potential energies. The propagation multiplier is positively correlated with the baseline propagation multiplier, the inertia coefficient, and the transaction confidence. For any node, the risk potential of the node is updated according to the propagation multiplier between the adjacent entry node of the node and the node, the risk potential energy of the adjacent entry node at the previous time, and the most recent transaction time from the adjacent entry node to the node. The adjacent entry node is the node that initiates a transaction to the node.

[0134] In a possible implementation, the network update module 402 is further configured to:

[0135] If the transaction data includes transaction volume, transaction initiator and counterparty, the edge weight is updated according to the transaction volume, transaction initiator and counterparty; if the transaction data includes the original transaction subject, the split subject and the split ratio, the node corresponding to the original transaction subject is split into the node of the split subject, and the edge weight of the original transaction subject is split into the edge weight of the split subject according to the split ratio; if the transaction data of the target transaction subject corresponds to an overdue event, the edge weight of the target transaction subject is reduced, and the target transaction subject is set as the starting node.

[0136] In one possible implementation, the confidence determination module 401 is further used to: obtain each type of transaction data sequence of the transaction subject, and perform sliding mean calculation on the transaction data sequence according to a preset sliding mean weight; scale the mean residual between the sliding mean and the historical mean through an adaptive scaling parameter to obtain the transaction data confidence; and dynamically adjust the sliding mean weight according to the transaction data confidence and the mean residual.

[0137] In one possible implementation, the confidence determination module 401 is further used to: when the transaction data sequence is of a social type, if the transaction data confidence is less than a first confidence threshold and the mean residual is greater than or equal to a first preset residual threshold, then reduce the transaction data sequence for sliding mean weighting; when the transaction data sequence is of a financial report type, if the transaction data confidence is greater than a second confidence threshold and the residual is greater than or equal to a second preset residual threshold, then increase the transaction data sequence for sliding mean weighting.

[0138] This embodiment provides a risk diffusion path prediction device that can execute a risk diffusion path prediction method of the above embodiment. Its implementation principle and technical effects are similar and will not be described in detail in this embodiment.

[0139] Figure 4 This is a schematic diagram of the structure of the electronic device provided by this application. Figure 4 As shown, the electronic device 600 provided in this embodiment includes: at least one processor 601 and a memory 602. Optionally, the electronic device 600 further includes a communication component 603. The processor 601, the memory 602 and the communication component 603 are connected via a bus 604.

[0140] During the specific implementation process, at least one processor 601 executes the computer-executable instructions stored in the memory 602, so that the at least one processor 601 performs the above method.

[0141] The specific implementation process of the processor 601 can be found in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here in this embodiment.

[0142] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly implemented by a hardware processor or implemented by a combination of hardware and software modules in the processor.

[0143] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (NVM), such as at least one disk memory.

[0144] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be classified into address buses, data buses, and control buses. For ease of illustration, the buses in the drawings of this application are not limited to just one bus or just one type of bus.

[0145] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0146] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0147] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0148] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0149] It should be noted that for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all optional embodiments, and the actions and modules involved are not necessarily required by this application.

[0150] It should be further noted that, although the various steps in the flowchart are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps may be performed in other orders. Moreover, at least a portion of the steps in the flowchart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily performed at the same time, but may be performed at different times. The execution order of these sub-steps or stages is not necessarily to be performed in sequence, but may be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0151] It should be understood that the above-described device embodiments are merely illustrative, and the device of the present application may also be implemented in other ways. For example, the division of units / modules in the above-described embodiments is merely a logical functional division, and actual implementations may employ other division methods. For example, multiple units, modules, or components may be combined or integrated into another system, or some features may be omitted or not implemented.

[0152] In addition, unless otherwise specified, the functional units / modules in the various embodiments of the present application may be integrated into a single unit / module, each unit / module may exist physically separately, or two or more units / modules may be integrated together. The aforementioned integrated units / modules may be implemented in the form of hardware or software program modules.

[0153] If the integrated unit / module is implemented in hardware, the hardware may be digital circuits, analog circuits, etc. The physical implementation of the hardware structure includes, but is not limited to, transistors, memristors, etc. Unless otherwise specified, the processor may be any appropriate hardware processor, such as a CPU, GPU, FPGA, DSP, and ASIC. Unless otherwise specified, the storage unit may be any appropriate magnetic storage medium or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), etc.

[0154] If the integrated unit / module is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a memory and includes a number of instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned memory includes various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0155] In the above embodiments, the description of each embodiment has its own emphasis. For parts not described in detail in a particular embodiment, please refer to the relevant description of other embodiments. The technical features of the above embodiments can be combined in any way. To keep the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0156] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. This application is intended to encompass any modifications, uses, or adaptations of the present application that follow the general principles of the present application and include common knowledge or customary techniques in the art that are not disclosed in this application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims. It should be understood that the present application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is limited only by the appended claims.

Claims

1. A risk diffusion path prediction method, characterized in that: include: determining confidence levels of multiple types of transaction data of multiple transaction entities, and determining a target transaction entity from the multiple transaction entities based on the transaction data confidence levels; Generating a directed topological network based on the target transaction subject, and updating the directed topological network based on the transaction data of the target transaction subject, wherein the directed topological network includes nodes and edge weights, wherein the nodes correspond to the target transaction subject, and the edge weights are related to the transaction volumes of the two nodes; Iterating the transaction data confidence of the node according to the edge weight to obtain the risk potential of each node; The node whose risk potential is greater than or equal to the preset risk potential threshold, and / or the node whose risk potential increase is greater than or equal to the preset increase threshold, is used as the starting node to determine the risk diffusion path. The end node of the risk diffusion path is the node whose similarity with the starting node is greater than or equal to the preset similarity threshold and whose risk potential is less than or equal to the preset risk potential threshold.

2. The method according to claim 1, characterized in that The iterative step of determining the confidence level of the transaction data of the node according to the edge weight to obtain the risk potential of each node includes: Determine the edge curvature of the two nodes based on the target edge weights of the two nodes and the similarity of the two nodes, wherein the edge curvature is positively correlated with the target edge weight and negatively correlated with the similarity, and the target edge weight includes the total outgoing edge weight of the transaction initiating node and the total incoming edge weight of the transaction counterparty node; Using the transaction data confidence of the node as the initial residual vector of the node, and iteratively updating the residual vector of the node based on the initial residual vector according to the edge curvature; The initial risk potential energy of the node is determined according to the residual vector, and the risk potential energy of each node is obtained by iteratively updating based on the initial risk potential energy.

3. The method according to claim 2, characterized in that The iteratively updating the residual vector of the node based on the initial residual vector according to the edge curvature includes: When the edge curvature is greater than a preset edge curvature threshold, updating the attenuation coefficients of the two nodes according to the difference between the edge curvature of the two nodes and the preset edge curvature threshold, wherein the attenuation coefficient is used to indicate the degree of adjustment of the residual when diffusing along the corresponding edge, and the attenuation coefficient is positively correlated with the difference; The residual vector of the node is iteratively updated based on the initial residual vector according to the edge curvature and the attenuation coefficient.

4. The method according to claim 3, characterized in that The iteratively updating the residual vector of the node based on the initial residual vector according to the edge curvature and the attenuation coefficient includes: Calculating the similarity between the node feature and the target label of the node as the feature label resonance degree of the node; For each edge, when the characteristic label resonance degree of the two nodes of the edge is greater than or equal to a preset resonance degree threshold, generating an edge attribute of the edge according to the characteristic label resonance degree of the two nodes of the edge; Iteratively updating the residual vector of the node based on the initial residual vector according to the edge attribute, the edge curvature, and the attenuation coefficient; The determining the initial risk potential energy of the node according to the residual vector includes: An initial risk potential energy of the node is determined according to the residual vector and the edge attribute.

5. The method according to claim 4, characterized in that The iterative updating based on the initial risk potential energy to obtain the risk potential energy of each node includes: Calculate the inertia coefficients of the two nodes based on the historical risk potential energies corresponding to the two nodes when conducting the most recent transaction, and the average historical risk potential energies of the two nodes, and determine the propagation magnifications of the two nodes based on the baseline propagation magnification, the inertia coefficient, and the transaction confidence, wherein the inertia coefficient is positively correlated with the two historical risk potential energies and negatively correlated with the average historical risk potential, and the propagation magnification is positively correlated with the baseline propagation magnification, the inertia coefficient, and the transaction confidence; For any node, the risk potential of the node is updated based on the propagation multiplier between the node and its adjacent entry node, the risk potential of the adjacent entry node at the previous time, and the time of the most recent transaction from the adjacent entry node to the node. The adjacent entry node is the node that initiates a transaction to the node.

6. The method according to claim 1, characterized in that The updating of the directed topology network according to the transaction data of the target transaction subject includes: If the transaction data includes transaction volume, transaction initiator, and transaction counterparty, then updating the edge weight according to the transaction volume, transaction initiator, and transaction counterparty; If the transaction data includes an original transaction subject, a split subject, and a split ratio, then split the node corresponding to the original transaction subject into nodes of the split subject, and split the edge weight of the original transaction subject into edge weights of the split subject according to the split ratio; If the transaction data of the target transaction subject corresponds to an overdue event, the edge weight of the target transaction subject is reduced, and the target transaction subject is set as the starting node.

7. The method according to claim 1, characterized in that Determining the confidence levels of multiple types of transaction data of multiple transaction entities includes: Obtaining a transaction data sequence of each type of the transaction subject, and performing a sliding mean calculation on the transaction data sequence according to a preset sliding mean weight; Scaling the mean residual between the sliding mean and the historical mean using an adaptive scaling parameter to obtain the transaction data confidence; The sliding mean weight is dynamically adjusted according to the transaction data confidence and the mean residual.

8. The method according to claim 7, characterized in that The dynamically adjusting the adaptive scaling parameter according to the transaction data confidence and the mean residual includes: When the transaction data sequence is of a social type, if the transaction data confidence is less than a first confidence threshold and the mean residual is greater than or equal to a first preset residual threshold, reducing the transaction data sequence for sliding mean weighting; When the transaction data sequence is of a financial report type, if the transaction data confidence is greater than a second confidence threshold and the residual is greater than or equal to a second preset residual threshold, the transaction data sequence is increased for sliding mean weighting.

9. A risk diffusion path prediction device, characterized in that: include: a confidence determination module, configured to determine the confidence levels of multiple types of transaction data of multiple transaction entities, and determine a target transaction entity from the multiple transaction entities based on the confidence levels of the transaction data; a network update module, configured to generate a directed topology network based on the target transaction subject and update the directed topology network based on the transaction data of the target transaction subject, wherein the directed topology network includes nodes and edge weights, wherein the nodes correspond to the target transaction subject and the edge weights are related to the transaction volume of the two nodes; a risk potential iteration module, configured to iterate the transaction data confidence of the node according to the edge weight to obtain the risk potential of each node; The risk diffusion path identification module is used to use the node whose risk potential is greater than or equal to the preset risk potential threshold, and / or the node whose risk potential increase is greater than or equal to the preset increase threshold, as the starting node to determine the risk diffusion path. The end node of the risk diffusion path is the node whose similarity with the starting node is greater than or equal to the preset similarity threshold and whose risk potential is less than or equal to the preset risk potential threshold.

10. An electronic device, characterized in that: include: a processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 8 when executed by a processor.

12. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 8 when executed by a processor.

Images