Bill processing method and device, equipment, storage medium and program product

Abstract

The invention provides a bill processing method and device, equipment, a storage medium and a program product, and relates to the field of bill pledge business. The method comprises the steps of obtaining a to-be-processed bill data set, generating a corresponding enhanced semantic graph, determining a corresponding collaborative tensor model, performing screening processing on bill data to obtain a compliance candidate bill data set, determining a corresponding take-off solution group, and performing bill risk sorting processing on compliance candidate bills to obtain a corresponding near-optimal syn-position sequence; and pulling the trigger event, updating the enhanced semantic graph and the collaborative tensor model, and according to the updated enhanced semantic graph, the updated collaborative tensor model and the updated near-optimal syn-position sequence, obtaining a target near-optimal syn-position sequence. The bill processing method and device solve the problem that in the prior art, existing bill processing faces risk events and multi-scene collaborative information which are updated in real time, and a bill pool meeting the minimum denomination coverage sequence cannot be efficiently obtained, so that the bill processing efficiency is low.

Description

Technical Field

[0001] The present application relates to the technical field of bill pledge, and in particular to a bill processing method, apparatus, equipment, storage medium and program product. Background Art

[0002] In the bill pledge business, with the continuous development of financial technology, how to efficiently and accurately manage bill pools has become a key issue. Optimizing bill pool management is primarily achieved through integrating multi-source data, building credit relationship models, conducting risk quantification, and conducting compliance screening.

[0003] Currently, existing technologies leverage graph embedding and tensor analysis to achieve risk measurement, with data structures often maintained in a static or weakly incremental manner. Risk quantification employs a two-level embedding and low-rank compression approach, prioritizing overall similarity or default correlation over asynchronous coupling factors. Portfolio optimization approaches narrow the solution set by appending heuristic rules, limiting the ability to reconcile the objective of minimizing denomination with regulatory coverage. This leads to a significant number of ranking combinations requiring reassessment following real-time event triggering.

[0004] However, existing bill processing is faced with real-time updated risk events and multi-scenario collaborative information, and is unable to efficiently obtain a bill pool that meets the minimum denomination coverage priority, resulting in inefficient bill processing. Summary of the Invention

[0005] The present application provides a method, device, equipment, storage medium and program product for processing bills, which are used to solve the problem that the existing bill processing is faced with real-time updated risk events and multi-scenario collaborative information, and is unable to efficiently obtain a bill pool that meets the minimum denomination coverage priority, resulting in low bill processing efficiency.

[0006] In a first aspect, the present application provides a method for processing bills, comprising:

[0007] Obtaining a data set of bills to be processed; wherein the data set of bills to be processed includes bill data of a plurality of bills to be processed;

[0008] Based on the bill dataset to be processed, a corresponding enhanced semantic graph is generated. The nodes of the enhanced semantic graph include bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes. The edges of the enhanced semantic graph include backbone topological edges composed of nodes and supplementary association edges. The supplementary semantic edges include guarantee relationship association edges, macroeconomic association edges, and regulatory event association edges.

[0009] According to the enhanced semantic graph, the corresponding collaborative tensor model is determined;

[0010] The bill data is screened and processed based on the enhanced semantic graph and the collaborative tensor model to obtain a compliant candidate bill dataset; wherein the compliant candidate bill dataset includes bill data of multiple compliant candidate bills;

[0011] Determine the corresponding jump solution group based on the enhanced semantic graph and the compliant candidate bill dataset;

[0012] Based on the jumping solution group, the compliant candidate bills are ranked by bill risk to obtain the corresponding near-priority sequence;

[0013] Pull trigger event;

[0014] According to the triggering event, the enhanced semantic graph and the collaborative tensor model are updated, and according to the updated enhanced semantic graph and the updated collaborative tensor model, the near-priority sequence is updated to obtain the target near-priority sequence, wherein the target near-priority sequence is used to indicate the priority execution order of the bills to be processed.

[0015] In one possible design, based on the invoice dataset to be processed, a corresponding enhanced semantic graph is generated, including:

[0016] Performing data preprocessing on the bill data of the bill to be processed to obtain preprocessed bill data; wherein the data preprocessing includes multiple processes of preset field transformation matrix processing, normalization processing, deduplication processing, composite hash key generation processing, and deduplication processing;

[0017] Extract bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes based on pre-processed bill data;

[0018] Determine the backbone topology edges and supplementary associated edges based on bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes;

[0019] An enhanced semantic graph is generated based on bill nodes, accepting bank nodes, region nodes, industry nodes and event nodes, as well as backbone topological edges and supplementary association edges.

[0020] In one possible design, the corresponding collaborative tensor model is determined based on the enhanced semantic graph, including:

[0021] Perform subgraph extraction on the enhanced semantic graph to obtain the accepting bank-bill subgraph;

[0022] Perform graph convolution on the accepting bank-bill subgraph to obtain the direct credit vector corresponding to the bill to be processed;

[0023] According to the accepting bank-bill subgraph and the direct credit vector, the attention weights associated with the regional nodes and industry nodes are determined, and based on the attention weights, the cross-layer bill vector corresponding to the bill to be processed is obtained;

[0024] According to the cross-layer bill vector, perform dimensionality reduction mapping processing and default geometric semantics preservation processing to obtain the bill low-dimensional feature vector corresponding to the bill to be processed;

[0025] According to the low-dimensional feature vector of the bill, slicing processing is performed, and impact increments are written locally for the slice to obtain a collaborative tensor model.

[0026] In one possible design, the corresponding jump solution group is determined based on the enhanced semantic graph and the compliant candidate bill dataset, including:

[0027] Obtaining bill data of compliant candidate bills based on the compliant candidate bill dataset;

[0028] Calculate the ranking vulnerability index corresponding to the compliant candidate bill based on the enhanced semantic graph and the bill data of the compliant candidate bill;

[0029] Determine the risk zone boundaries based on the sequential vulnerability indicators;

[0030] Determine the take-off solution group based on the risk zone boundary.

[0031] In one possible design, the compliant candidate bills are ranked by bill risk based on the starting solution group to obtain a corresponding near-priority sequence, including:

[0032] According to the jump solution group, instantiation processing is performed to obtain a hot-cold dual-temperature bitter fish network; wherein the hot-cold dual-temperature bitter fish network includes a cold layer for performing local search and a hot layer for performing global search;

[0033] An iterative sorting process is performed on the compliant candidate bills according to the hot-cold dual-temperature bitter fish network to output a near-priority sequence through the hot-cold dual-temperature bitter fish network.

[0034] In one possible design, the bill data is screened and processed based on the enhanced semantic graph and collaborative tensor model to obtain a dataset of compliant candidate bills, including:

[0035] Determine the accepting bank-region exposure matrix based on the enhanced semantic graph;

[0036] Performing a first screening process on the bill data according to the accepting bank-region exposure matrix to obtain a first candidate bill data set;

[0037] Based on the enhanced semantic graph, risk scenario labels, violation semantic distances, and region labels are determined;

[0038] According to the collaborative tensor model, the first candidate bill dataset is subjected to a second screening process in combination with the risk scenario label, violation semantic distance, and region to obtain a second candidate bill dataset;

[0039] performing a second screening process on the first candidate bill dataset according to the collaborative tensor model to obtain a second candidate bill dataset;

[0040] Obtaining the accumulated denomination data and guarantee data of the bills to be processed in the second candidate bill data set;

[0041] According to the collaborative tensor model, the second candidate bill dataset is subjected to a third screening process in combination with the accumulated denomination data and the guarantee data to obtain a compliant candidate bill dataset.

[0042] In a possible design, after updating the enhanced semantic graph and the collaborative tensor model according to the triggering event, and updating the near-priority sequence according to the updated enhanced semantic graph and the updated collaborative tensor model to obtain the target near-priority sequence, the following further steps are included:

[0043] Verifying the target near-priority sequence based on the face value data and guarantee data of the bills to be processed in the target near-priority sequence to obtain a verification result;

[0044] If the verification is successful, the bill pledge pool exit order will be generated based on the target near-priority sequence.

[0045] In one possible design, the bill data includes the face value, issue date, maturity date, issuer, accepting bank, and complete endorsement chain.

[0046] In a second aspect, the present application provides a bill processing device, comprising:

[0047] A first acquisition module is configured to acquire a data set of bills to be processed, wherein the data set of bills to be processed includes bill data of a plurality of bills to be processed;

[0048] A generation module is used to generate a corresponding enhanced semantic graph based on the bill dataset to be processed; wherein the nodes of the enhanced semantic graph include bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes; the edges of the enhanced semantic graph include backbone topological edges composed of nodes and supplementary association edges; the supplementary semantic edges include guarantee relationship association edges, macroeconomic association edges, and regulatory event association edges;

[0049] A first determination module is used to determine a corresponding collaborative tensor model according to the enhanced semantic graph;

[0050] A screening module is used to screen the bill data based on the enhanced semantic graph and the collaborative tensor model to obtain a dataset of compliant candidate bills; wherein the dataset includes bill data of multiple compliant candidate bills;

[0051] The second determination module is used to determine the corresponding jump solution group based on the enhanced semantic graph and the compliant candidate bill dataset;

[0052] The first sorting module is used to sort the risk of the compliant candidate bills according to the starting solution group to obtain the corresponding near-priority sequence;

[0053] Pull module, used to pull trigger events;

[0054] The second sorting module is used to update the enhanced semantic graph and the collaborative tensor model according to the triggering event, and update the near-priority sequence according to the updated enhanced semantic graph and the updated collaborative tensor model to obtain the target near-priority sequence, wherein the target near-priority sequence is used to indicate the priority execution order of the bills to be processed.

[0055] In a third aspect, the present application provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;

[0056] Memory stores computer-executable instructions;

[0057] The processor executes the computer-executable instructions stored in the memory to implement a method for processing bills as described in the first aspect of the invention.

[0058] In a fourth aspect, the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement a method for processing bills as described in the first aspect of the invention.

[0059] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements a method for processing bills according to the invention content of the first aspect.

[0060] The present application provides a method, device, equipment, storage medium and program product for processing bills, including: obtaining a data set of bills to be processed; generating a corresponding enhanced semantic graph based on the data set of bills to be processed; determining a corresponding collaborative tensor model based on the enhanced semantic graph; screening and processing the bill data based on the enhanced semantic graph and the collaborative tensor model to obtain a data set of compliant candidate bills; determining a corresponding jump solution group based on the enhanced semantic graph and the compliant candidate bill data set; performing bill risk sorting processing on the compliant candidate bills based on the jump solution group to obtain a corresponding near-preferred priority sequence; pulling a trigger event; updating the enhanced semantic graph and the collaborative tensor model based on the trigger event, and updating the near-preferred priority sequence based on the updated enhanced semantic graph and the updated collaborative tensor model to obtain a target near-preferred priority sequence. Compared with the prior art, the existing bill processing is faced with real-time updated risk events and multi-scenario collaborative information, and cannot efficiently obtain a bill pool that meets the minimum denomination coverage priority, resulting in low bill processing efficiency. This application proposes to strengthen the semantic graph structure, disassembling the guarantee chain into supplementary edges with guarantee level and counter-guarantee ratio attributes, injecting macroeconomic indicators, and then connecting the event nodes that trigger the event. Through fully structured edge attributes and traceable timelines, this graph model makes liquidity fragility, regulatory gradients, and immediate risk events explicit in the same graph domain and maintains incremental writability, thereby providing a unified data base for subsequent risk propagation and rapid backtracking at any granularity. It can quickly and efficiently obtain a pool of bills that meet the minimum denomination coverage priority, thereby improving the efficiency of bill processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0062] Figure 1 A schematic diagram of the system architecture of a bill processing method provided in an embodiment of the present application;

[0063] Figure 2 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 1 ;

[0064] Figure 3 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 2 ;

[0065] Figure 4 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 3 ;

[0066] Figure 5 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 4 ;

[0067] Figure 6 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 5 ;

[0068] Figure 7 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 6 ;

[0069] Figure 8 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 7 ;

[0070] Figure 9 A schematic diagram of the structure of a bill processing device provided in an embodiment of the present application;

[0071] Figure 10 A schematic structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0072] 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.

[0073] In the embodiments of the present application, words such as "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit differences. It should be noted that in the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described in this application as "exemplary" or "for example" should not be interpreted as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way. In the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more.

[0074] It should be noted that the phrase "at the time of" in the embodiments of the present application can refer to the instant when a certain event occurs or a period of time after the event occurs, and the embodiments of the present application do not specifically limit this. Furthermore, the bill processing method provided in the embodiments of the present application is merely an example, and the bill processing method may include more or less content.

[0075] Loan supervision refers to the control and management process of supervising and managing the implementation of pre-loan conditions in the credit approval document and the payment of credit funds after the credit business is approved, and deciding whether to issue and pay credit funds.

[0076] In the bill pledge business, existing technologies usually follow the general process of "multi-source data integration - credit relationship modeling - risk quantification - compliance screening - portfolio optimization".

[0077] Specifically, the industry practice is to first extract the bill number, face value, issue date, accepting bank, endorsement chain, and guarantee terms from the bill core system, bill exchange interface, and third-party credit information. Using field mapping tables, unified format verification and deduplication are completed. The bill information is then mapped to a relational database or attribute graph for storage. The graph model uses the bill as the central node, connecting to entities such as the accepting bank, region, and industry. Guarantee loops are represented by weighted edges, with weights recording the guarantee level and counter-guarantee ratio. Macroeconomic indicators and liquidity monitoring values ​​are written into the corresponding node attributes using the "date-region" double key, forming a baseline financial knowledge graph with time series characteristics.

[0078] In the risk quantification phase, existing solutions typically employ a two-level embedding mechanism: First, graph convolution or random walks based on adjacency matrices are used to generate initial bill vectors to capture direct credit relationships. This is then iteratively propagated along regional and industry links, with weighted updates to the embedding results using sequences such as liquidity and capital outflow as attention factors. The resulting high-dimensional vectors are then compressed into a low-dimensional space using principal component analysis, variational autoencoding, or weighted singular value decomposition. Simultaneously, a three-dimensional tensor is constructed to store the similarity or default distance of bill pairs under different scenarios. These scenario segments can be discretized and managed by quarterly macroeconomic models, regulatory policies, or historical extreme events, facilitating rapid indexing later.

[0079] In the portfolio generation process, industry tools often employ a "rule-based filtering + heuristic search" approach. The system first eliminates non-compliant bill pairs based on regulatory concentration thresholds, interbank exposure tables, and default correlation coefficients, then gradually tightens the scope based on scenario simulation results. Common optimization algorithms include greedy backtracking, genetic evolution, simulated annealing, or particle swarm optimization. Their objective function considers the portfolio's denomination, worst-case double default coverage ratio, and the remaining maturity of the bills. During the search process, bills with similar risks are allowed to be exchanged within adjacent risk bands to ensure rapid convergence and interpretability.

[0080] After launch, an event-driven mechanism receives real-time information on rating adjustments, regulatory penalties, and bill transfers via a message queue. After locating affected nodes, it incrementally updates graph attributes and tensor slices, subsequently triggering local re-optimization. Adjustment strategies typically employ simulated annealing with increasing temperature or multi-point parallel genetic operations, iterating within a limited time window until constraints are re-satisfied. The final ranking table, verified by coverage enumeration, is then written to the pledge management system, completing the bill pool implementation.

[0081] On the one hand, while existing methods can integrate bills, accepting banks, regions, and macroeconomic variables into a unified knowledge graph or attribute graph, and leverage graph embedding and tensor analysis to implement risk measurement, these data structures are often maintained in a static or weakly incremental manner. High-frequency events such as macroeconomic shocks, regulatory penalties, and rating adjustments are often handled through batch recalculation or simplified weight patching. This results in lags in the timeliness and granularity of risk indicators, making it difficult to accurately reflect the real-time interactions between nodes.

[0082] On the one hand, the commonly used two-level embedding and low-rank compression methods in the risk quantification process focus more on overall similarity or default correlation, lacking specific characterization of asynchronous coupling factors such as guarantee chains, discount flows, and liquidity fragility. As a result, they still deviate from the true default transmission path. While the three-dimensional tensor generated based on this can store historical fragments by scenario, the mapping relationship between regulatory thresholds, compliance rules, or risk events is often maintained by offline scripts, with coarse update granularity, which can easily lead to model gaps.

[0083] On the other hand, at the portfolio optimization level, greedy backtracking, simulated annealing, or genetic algorithms are limited in their convergence speed when the search space expands exponentially. They can only narrow the solution set by adding heuristic rules, which restricts the ability to reconcile the objective of minimizing denomination with regulatory coverage. More importantly, existing tools generally use denomination, maturity, and concentration as primary variables, lacking a clear understanding of the inherent fragility of the ranking structure. When real-time events are triggered, the sheer number of ranking combinations that need to be reassessed can easily lead to computational bottlenecks and make it difficult to quickly identify the critical bill pairs that lead to coverage loss.

[0084] Based on this, the embodiments of the present application provide a bill processing method, device, equipment, storage medium and program product, which can be used in the field of bill pledge business, aiming to solve the above technical problems of the prior art.

[0085] The inventive concept of this application lies in: To address the aforementioned issues, while researching the efficiency of bill processing, the inventors discovered that, under the premise that constraints such as regulatory double default coverage ratios, accepting bank and regional concentration are not violated, it is impossible to efficiently determine and continuously verify the bill priority sequence for real-time updated risk events and multi-scenario collaborative information, thereby failing to ensure that the pledge pool always maintains the minimum denomination and controllable risk exposure. Based on this, the inventors proposed a strengthened semantic graph structure, combining bill nodes with accepting banks, regions, and industries to form a backbone topology. The guarantee chain is then broken down into supplementary edges with guarantee levels and counter-guarantee ratio attributes, injected with macroeconomic indicators, and then connected to the event nodes that trigger the event. Through fully structured edge attributes and a traceable timeline, this graph model makes liquidity vulnerability, regulatory gradients, and immediate risk events explicit in the same graph domain and maintains incremental writeability, thus providing a unified data base for subsequent risk propagation and rapid backtracking at any granularity. This allows for the rapid output of the minimum denomination coverage priority of the bill pool in a dynamic risk environment. Based on this, the present application proposes a bill processing method to further improve the efficiency of bill processing.

[0086] 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.

[0087] Figure 1 A schematic diagram of the system architecture of a bill processing method provided in an embodiment of the present application. Figure 1 In the above architecture, the above architecture includes at least one of a data acquisition device 11, a processing device 12 and a display device 13.

[0088] It is understood that the structure illustrated in the embodiments of this application does not constitute a specific limitation on the architecture of the bill processing system. In other feasible implementations of this application, the above architecture may include more or fewer components than shown, or may combine or split certain components, or arrange the components differently. The specific configuration can be determined based on the actual application scenario and is not limited here. Figure 1 The components shown can be implemented in hardware, software, or a combination of software and hardware.

[0089] In the specific implementation process, the data acquisition device 11 may include an input / output interface and a communication interface. The data acquisition device 11 may be connected to the processing device through the input / output interface or the communication interface to obtain the bill data of multiple bills to be processed.

[0090] The processing device 12 can process data based on the bill data of the bill to be processed.

[0091] The display device 13 may also be a touch screen display or a screen of a terminal device, which is used to display the above-mentioned content to achieve interaction with the user.

[0092] It should be understood that the above-mentioned processing device can be implemented by a processor reading instructions in a memory and executing the instructions, or it can be implemented by a chip circuit.

[0093] In addition, the network architecture and business scenarios described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Ordinary technicians in this field can know that with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0094] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0095] The technical solution of this application is described in detail below with reference to specific embodiments:

[0096] Figure 2 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 1 ,like Figure 2 As shown, the method includes:

[0097] S201: Obtain a dataset of bills to be processed.

[0098] The unprocessed bill data set includes bill data of multiple unprocessed bills.

[0099] Among them, bill data includes face value, issue date, maturity date, issuer, accepting bank and complete endorsement chain.

[0100] In this embodiment, the face value, issue date, maturity date, issuer, accepting bank and endorsement chain of the bill to be processed are collected.

[0101] S202: Generate a corresponding enhanced semantic graph based on the bill dataset to be processed.

[0102] Among them, the nodes of the enhanced semantic graph include bill nodes, accepting bank nodes, regional nodes, industry nodes and event nodes. The edges of the enhanced semantic graph include backbone topological edges composed of nodes and supplementary association edges. The supplementary semantic edges include guarantee relationship association edges, macroeconomic association edges and regulatory event association edges.

[0103] Specifically, the bill node is a core node, including basic attributes: bill number, issue date, issuer, amount, term, status, etc. It also includes associated attributes: connected to the accepting bank, region, industry, etc. through edges.

[0104] Specifically, the accepting bank node includes business attributes: bank code, name, liquidity vulnerability, regulatory gradient, rating, etc. It also includes geographical attributes: region, which is associated with the region node.

[0105] Specifically, a regional node includes macro attributes: region code, name, GDP growth rate, CPI index, and other economic indicators. It also includes a time dimension: connecting to bills and events in different periods through time index edges.

[0106] Specifically, the industry node includes industry characteristics: industry code, name, prosperity index, policy support, etc. It also includes association attributes: association to the bill node through the issuer.

[0107] Specifically, an event node includes event attributes: event ID, type (regulatory penalty / rating adjustment), time, description, impact scope, etc. It also includes dynamic attributes: connecting to the affected bills or accepting banks through time index edges.

[0108] Specifically, the backbone topology edges include the acceptance relationship: bill → accepting bank (direct business relationship), the region: accepting bank → region (geographic location), and the industry: issuer → industry (economic classification).

[0109] Specifically, the supplementary semantic edges include guarantee relations: bills Bills (within the same guarantee chain) include guarantee level and counter-guarantee ratio attributes. Macroeconomic correlation: Bill → Region (based on issue date and region index), associated with macroeconomic sequence. Regulatory event correlation: Event → Bill / Accepting Bank (based on time index), including impact degree and duration attributes.

[0110] S203. Determine the corresponding collaborative tensor model according to the enhanced semantic graph.

[0111] S204: Screen and process the bill data based on the enhanced semantic graph and collaborative tensor model to obtain a compliant candidate bill dataset.

[0112] The compliant candidate bill data set includes bill data of multiple compliant candidate bills.

[0113] S205: Determine the corresponding jump solution group based on the enhanced semantic graph and the compliant candidate bill dataset.

[0114] S206. Based on the starting solution group, the compliant candidate bills are sorted by bill risk to obtain a corresponding near-priority sequence.

[0115] S207: Pull trigger event.

[0116] S208. Update the enhanced semantic graph and the collaborative tensor model according to the triggering event, and update the near-priority sequence according to the updated enhanced semantic graph and the updated collaborative tensor model to obtain the target near-priority sequence.

[0117] The target near-priority sequence is used to indicate the priority execution order of the bills to be processed.

[0118] Specifically, regulatory penalties, rating adjustments and bill transfer events are pulled in at regular intervals, and the relevant nodes in the graph are located by time-indexed edges and risk increments are written. After synchronizing the increments to the collaborative tensor, the resonance penalty weights of high-risk bills are increased and the temperature of the hot layer is increased. At the same time, a cooling window is set for the affected bills in the cold layer. A round of cold adjustment and hot jump is performed in the updated energy field, so that the sequence continues to iterate in the direction of smaller denominations, more stable coverage and weaker resonance until the energy improvement is below the threshold or the maximum number of iterations is reached.

[0119] The present embodiment provides a method for processing bills, comprising: obtaining a data set of bills to be processed; generating a corresponding enhanced semantic graph based on the data set of bills to be processed; determining a corresponding collaborative tensor model based on the enhanced semantic graph; screening the bill data based on the enhanced semantic graph and the collaborative tensor model to obtain a data set of candidate bills that are compliant; determining a corresponding jump solution group based on the enhanced semantic graph and the data set of candidate bills that are compliant; performing bill risk sorting processing on the candidate bills based on the jump solution group to obtain a corresponding near-preferred priority sequence; pulling a trigger event; updating the enhanced semantic graph and the collaborative tensor model based on the trigger event, and updating the near-preferred priority sequence based on the updated enhanced semantic graph and the updated collaborative tensor model to obtain a target near-preferred priority sequence. Compared with the prior art, the existing bill processing is unable to efficiently obtain a bill pool that meets the minimum denomination coverage priority in the face of real-time updated risk events and multi-scenario collaborative information, resulting in low bill processing efficiency. The present application proposes an enhanced semantic graph structure, which decomposes the guarantee chain into supplementary edges with guarantee level and counter-guarantee ratio attributes, injects macroeconomic indicators, and then connects the event nodes of the trigger event. Through fully structured edge attributes and traceable timelines, this graph model makes liquidity fragility, regulatory gradients, and immediate risk events explicit in the same graph domain and keeps them incrementally writable, thus providing a unified data base for subsequent risk propagation and rapid backtracking at any granularity. It can quickly and efficiently obtain a pool of bills that meet the minimum denomination coverage priority, thereby improving the efficiency of bill processing.

[0120] Figure 3 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 2 ,like Figure 3 As shown, the specific implementation steps of the above S202 include:

[0121] S301 : Preprocess the bill data of the bill to be processed to obtain preprocessed bill data.

[0122] Among them, data preprocessing includes multiple types of preset field transformation matrix processing, normalization processing, deduplication processing, composite hash key generation processing and deduplication processing.

[0123] For example, for N commercial bills to be processed, the face value, issue date, maturity date, issuer, accepting bank, and complete endorsement chain are collected to form a raw field vector.

[0124] Specifically, the system presets the field transformation matrix Weight diagonal matrix With bias And the multi-index normalization is completed based on the sample mean μ, and the standardized vector s is obtained. i .

[0125]

[0126] in, is a piecewise linear function, e p is a basis column vector.

[0127] Furthermore, the composite hash key K is calculated based on the bill number || issue date || issuer i , only keep s with unique keys i .

[0128] S302. Extract bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes based on the pre-processed bill data.

[0129] S303. Determine the backbone topology edges and supplementary associated edges based on the bill nodes, the accepting bank nodes, the region nodes, the industry nodes, and the event nodes.

[0130] S304: Generate an enhanced semantic graph based on the bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes, as well as the backbone topology edges and supplementary association edges.

[0131] For example, the system uses ticket node P i Bind the accepting bank node A as the center in turn j , regional node R k , Industry Node I m , forming a four-level backbone topology of bills-accepting bank-region-industry in the figure.

[0132] Specifically, for any two bills P in the same guarantee chain i ,P l Establish supplementary edge E il , the guarantee level g il and the counter-guarantee ratio η il Write edge attributes.

[0133] Furthermore, access to the macroeconomic sequence M t,r and regulatory indicators U t,r Based on the bill issuance date τ i Complete date-region double alignment with regional index k, Write the bill attribute vector e i .

[0134] Furthermore, according to the discount flow, P i Mapped to the industrial chain node L m , and the macro shock sensitive component γ m Accumulate to L m , so that the nodes in the industrial chain have the ability of periodic pulse feedback.

[0135] Furthermore, the liquidity sequence L of the accepting bank is extracted within the time window [t0-τ,t0] j (t) and net capital outflow C j (t), calculate the liquidity vulnerability V j At the same time, the aggregated regional regulatory intensity index ζ k (q) Forming a regulatory gradient Γ k :

[0136]

[0137] Among them, σ(·) is the standard deviation operator, λ c is the capital outflow weight, ε is the smoothing constant, H j It is a proprietary bill position.

[0138] Specifically, the system will V j Solidified into attribute node L j ,Γ k Solidified as node G k , and respectively with A j ,R k Bidirectional links are formed to make the default correlation factor explicit in the graph.

[0139] Finally, the system monitors events such as regulatory penalties and rating adjustments, and generates event node E u , record the occurrence time θ u With the impact window ω u .

[0140] Specifically, for the bills P affected by the event i or Accepting Bank A j Add an inbound time index edge, the edge attribute is {θ u ,ω u}. So far, the node set {P i ,A j ,R k ,I m ,L j ,G k ,E u} and the enhanced semantic graph G of all directed edges S Write to the graph storage engine for subsequent cross-layer embedding, coordinate tensor construction, and optimize the unique data source for sequence risk.

[0141] It should be noted that the four-level topological structure establishes a three-dimensional risk transmission network of bills-institutions-regions-industries. The guarantee levels (such as general guarantees / joint and several liabilities) and counter-guarantee ratios in the supplementary edges can quantify the intensity of credit risk transmission, realize the directional and attenuation modeling of risk spillovers, and improve the accuracy of correlation risk warnings.

[0142] It should be noted that by embedding the GDP growth rate, PMI and other time series data into the issuance date index, and associating the regional index with spatial variables such as regional credit growth rate and industrial policy, the model has the ability to identify macroeconomic cycles and can quantify the nonlinear impact of economic fluctuations on the bill repayment ability.

[0143] In this implementation, the bill node is centered and sequentially bound to the accepting bank node, regional node, and industry node, forming a four-tiered backbone topology: bill-accepting bank-region-industry. Supplementary edges containing guarantee levels and counter-guarantee ratios are then written between bills within the same guarantee chain. Furthermore, macroeconomic sequences are written based on the issue date and regional index, and liquidity vulnerability and regulatory gradients are written for the accepting bank. Event nodes are generated for regulatory penalties or rating adjustments, and time-indexed edges are used to connect the affected bills or accepting banks, resulting in an enhanced semantic graph. By constructing a financial knowledge graph with dual-dimensional spatiotemporal attributes, this allows for a transition from static association analysis to dynamic risk evolution, thereby improving bill processing efficiency.

[0144] Figure 4 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 3 ,like Figure 4 As shown, the specific implementation steps of the above S203 include:

[0145] S401. Extract subgraphs from the enhanced semantic graph to obtain an accepting bank-bill subgraph.

[0146] S402: Perform graph convolution processing on the accepting bank-bill subgraph to obtain a direct credit vector corresponding to the bill to be processed.

[0147] For example, in G S Extract the accepting bank-bill directed subgraph Construct the sequential adjacency matrix A according to the direction of the creditor's rights dir and the order weight matrix O, and perform a round of graph convolution with the original attribute matrix X of the bill as input to obtain the bill vector that only reflects the direct credit relationship

[0148] H (1) .

[0149] S403. Determine the attention weights associated with the regional nodes and industry nodes based on the accepting bank-bill subgraph and the direct credit vector, and obtain the cross-layer bill vector corresponding to the bill to be processed based on the attention weights.

[0150] For example, in In the shell, along the edge of the region (P i ,R k ) and industry edge (P i ,I m )H (1)Spread outward. Read node liquidity index {V j ,Γ k}Calculate attention weight α uv , the fragile link is due to α uv The larger the size, the higher the aggregation ratio, and the output is a cross-layer bill vector H containing both macro and link risks. (2) .

[0151] S404: Perform dimensionality reduction mapping processing and default geometric semantics retention processing according to the cross-layer bill vector to obtain a bill low-dimensional feature vector corresponding to the bill to be processed.

[0152] Among them, low dimension means the dimension is reduced compared to three-dimensional and cross-layer bill vectors.

[0153] S405. Perform slicing processing based on the low-dimensional feature vector of the bill, and write the impact increment locally for the slice to obtain a collaborative tensor model.

[0154] For example, in order to suppress redundant noise and amplify the default-sensitive dimension, weighted singular value decomposition is performed with the liquidity resonance intensity matrix R as the weight, and the first r feature subspaces are selected to complete the low-rank mapping.

[0155] Furthermore, geometric interpretability is intuitively preserved in this mapping space.

[0156] Among them, the mapping and distance calculation can be specifically expressed as:

[0157]

[0158] d il =‖z i -z l ‖2

[0159] Among them, U r and Σ r are the eigenvectors and singular value submatrices obtained by weighted decomposition, z i is the coordinate of the i-th bill in the low-dimensional space, d il is the default semantic distance between the two bills.

[0160] Furthermore, according to the business scenario label λ, {d il}Layered, filling the 3D tensor body layer by layer

[0161] Among them, each slice retains the default distance and mapping coordinates between bill pairs, forming a prototype of a collaborative tensor covering all historical scenarios.

[0162] Furthermore, the newly written supervisory event node E in the graph is monitored in real time. u Once an event touches a note P via a time-indexed edgei , the system is in the corresponding slice Injection shock increment Δ u , so that the tensor can capture the future risk amplification effect and update the correlation distance in time.

[0163] Furthermore, for the slices that have been injected with shock, the system performs local Laplace smoothing in the distance dimension to weaken the isolated noise peaks and obtain a synergy tensor that maintains historical synergy and integrates the latest shock.

[0164] In this implementation, a subgraph of accepting banks and bills is extracted from a semantic graph, and graph convolution is performed on the original bill attribute matrix to obtain a direct credit vector. When spreading along regional and industry edges, node liquidity indicators are read to calculate attention weights, outputting a cross-layer bill vector containing macro and link risks. A weighted singular value decomposition is performed using the liquidity resonance strength matrix as weight, selecting the top k feature subspaces to obtain low-dimensional coordinates and populating three-dimensional tensor slices based on business scenario labels. When an event node touches a bill via a time-indexed edge, only the impact increment is written locally to the corresponding slice, and local Laplace smoothing is performed to form a collaborative tensor. By constructing a "local-global-dynamic" trinity graph tensor computation framework, a closed loop from micro-bill credit assessment to macro-risk hedging is achieved, thereby improving bill processing efficiency.

[0165] Figure 5 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 4 ,like Figure 5 As shown, the specific implementation steps of the above S204 include:

[0166] S501. Determine the accepting bank-region exposure matrix based on the enhanced semantic graph.

[0167] S502: Perform a first screening process on the bill data according to the accepting bank-region exposure matrix to obtain a first candidate bill data set.

[0168] S503: Determine risk scenario labels, violation semantic distances, and region labels based on the enhanced semantic graph.

[0169] S504. According to the collaborative tensor model, combined with the risk scenario label, violation semantic distance and region, the first candidate bill dataset is subjected to a second screening process to obtain a second candidate bill dataset.

[0170] S505 . Perform a second screening process on the first candidate bill dataset according to the collaborative tensor model to obtain a second candidate bill dataset.

[0171] S506: Obtain the accumulated denomination data and guarantee data of the bills to be processed in the second candidate bill data set.

[0172] S507 : Based on the collaborative tensor model, the accumulated denomination data and the guarantee data are combined to perform a third screening process on the second candidate bill dataset to obtain a compliant candidate bill dataset.

[0173] For example, reading the latest regulatory bulletin, the upper limit of the concentration of acceptance banks and regional concentration ceiling Mapped to a unified threshold table Y.

[0174] Furthermore, for any bill pair (P i ,P l )'s Accepting Bank-Region Exposure Matrix E il Calculate the compliance score σ il :

[0175]

[0176] in, represents the combined face value exposure of the two bills to the jth accepting bank, represents the exposure of the k-th region.

[0177] Among them, the range σ il ≤1 is compliant, σ il >1 directly enters the elimination queue.

[0178] Furthermore, the collaborative tensor slices are loaded in the order of risk scenario labels λ

[0179] Among them, the slice has embedded the regulatory impact increment Δ in S203 u ,Therefore, layer-by-layer loading can ensure scene isolation and avoid misjudgment caused by cross-scene information folding.

[0180] Furthermore, the default semantic distance d is read for each bill pair in a single scenario λ. il With the region mark R i =R l , if the same region and d il ≥ρ high ” is regarded as a highly positively correlated resonance combination, which is immediately removed and the cause field is recorded to achieve the first round of weakening of spatial dimension resonance risk.

[0181] Furthermore, for the combinations that pass the regional screening, verify their accepting bank identification (A i ,A l ). If the two bills belong to the same accepting bank and correspond to σ il Because the exposure item of the accepting bank exceeds the threshold, the combination is directly divested, thereby reducing the concentration in the accepting bank dimension.

[0182] Furthermore, the denomination F is accumulated for the remaining bill pairs in each scenario.i ,F l and guarantee compensation

[0183] G i ,G l , calculate the worst double default coverage ratio CR min :

[0184]

[0185] in, is the set of remaining bill pairs under scenario λ, S λ Simulate loss demand exposure for the current layer.

[0186] Among them, if CR min Below the regulatory threshold θ cov , then the bill pair is recorded in the undercoverage list to ensure that extreme losses are quantified in advance.

[0187] Furthermore, the system eliminates the corresponding combinations according to the insufficient coverage list and records the cross-scene duplicates based on the key (P i ,P l ) Perform a union to obtain a streamlined and fully compliant set of candidate bills

[0188] Finally, Solidify the index table write cache and record the collection size Together with the scenario coverage matrix, it is used as the only search space for subsequent priority vulnerability banding and bitterfish search, ensuring that the optimization phase runs with the minimum controllable basis.

[0189] It should be noted that the joint screening of default semantic distance (the similarity of default transmission paths extracted based on NLP) and regional economic markers (such as GDP growth rate and credit growth rate dispersion) can identify "risk resonance combinations" that are difficult to capture using traditional methods.

[0190] It should be noted that the linkage verification of the cumulative mechanism of guarantee compensation (including multi-layer compensation of margin, collateral and guarantee) and the worst double default coverage ratio (taking into account the joint probability of major default and minor default) realizes accurate capital measurement under the risk-based capital framework.

[0191] In this embodiment, an accepting bank-region exposure matrix is ​​calculated based on the regulatory acceptance bank concentration cap and regional concentration cap, and bill pairs exceeding the threshold are directly eliminated. The default semantic distance and regional tag are analyzed scenario by scenario, eliminating highly resonant combinations and combinations with high concentration within the same accepting bank. For the remaining bill pairs, the face value and collateral are accumulated. If the worst-case double default coverage ratio falls below the regulatory threshold, the corresponding bill pair is added to the undercoverage list and eliminated, resulting in a set of compliant candidate bills. By establishing a three-tiered protection system of "hard constraints, soft screening, and dynamic verification," a refined risk-return balance is achieved while ensuring compliance, thereby improving bill processing efficiency.

[0192] Figure 6 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 5 ,like Figure 6 As shown, the specific implementation steps of the above S205 include:

[0193] S601. Obtain bill data of a compliant candidate bill based on a compliant candidate bill dataset.

[0194] S602: Calculate the priority vulnerability index corresponding to the compliant candidate bill based on the enhanced semantic graph and the bill data of the compliant candidate bill.

[0195] For example, traversing And for each bill P i Read the remaining days T i , the number of past rolling χ-day discounts κ i , cross-guarantee level δ i , historical margin call ratio μ i and the number of early redemption triggers ρ i .

[0196] In order to highlight the "time-flow-guarantee" coupling weakness of bills in the pledge sequence, the sequence vulnerability is calculated:

[0197]

[0198] Among them, the four weights β r The historical early outflow samples are fitted by maximum likelihood; the exponential powers λ1 and λ2 amplify the nonlinear margin squeeze and early redemption inertia.

[0199] S603. Determine the risk zone boundary based on the sequential vulnerability index.

[0200] After the calculation is completed, Write into the sequence vulnerability index table and associate it with the bill primary key -.

[0201] Further, press Sort by high to low, based on the sequence of adjacent quantile differences Dynamically capture vulnerability jump points.

[0202] Furthermore, in order to prevent excessive segmentation from causing sparse samples within the band, a window foldback coefficient ω is constructed. w And generate the risk zone boundary set as follows

[0203]

[0204] Among them, ∈ band It can be automatically calibrated with the overall pledge balance, ω w Make the bandwidth of high-vulnerability areas relatively convergent, and ensure that bills with the same band are fully homogeneous in terms of residual value and liquidity elasticity.

[0205] S604: Determine a take-off solution group based on the risk zone boundary.

[0206] For example, to enter single risk zone b, first press the face value F i Sort in monotonically decreasing order, then search the risk-free net amount window In the double default resonance record, for any combination (P i ,P l )Write penalty points If the amount of the penalized sequence is the same, the order is kept, and the final output is the internal jump solution S (b,0) .

[0207] Among them, all S (b,0) A cross-band jumping solution group is formed to provide diverse initial energy states for the Kuyu network.

[0208] In order to avoid dimensional explosion and out-of-control sequence jumps, it is stipulated that the individual bitter fish can only perform exchange transitions between the current band b and the adjacent band b±1, and the exchanged notes must meet

[0209] Specifically, the double constraint directly changes the search space from Converges to

[0210] Where r is the average bandwidth, ensuring that the coordinated goals of denomination saving and sequence stability can converge within an acceptable time.

[0211] Furthermore, local subgraphs are generated based on the risk band division

[0212] Specifically, for any two subgraphs If |pq|≤1 then in the high-order graph Connect hyperedges and assign weights:

[0213]

[0214] Among them, γ is the vulnerability gradient weight, ρ reg is the supervisory friction coefficient, ψ pq (t) represents the intensity of regulatory shocks of the cross-band portfolio at time t. Weight w pq It not only measures the monetary cost of cross-band transition, but also dynamically reflects the regulatory impact, making the Kuyu network channel consistent with real business friction.

[0215] Finally, the system sets the local subgraph High-order graph Jump group {S (b,0)} and the complete vulnerability index package written to the optimization cache, generating version number v S4 .

[0216] It should be noted that the Sequential Vulnerability Index integrates four characteristics: remaining days (time value of liquidity), historical discount times (market recognition), cross-collateralization levels (credit enhancement depth), and margin call / early redemption inertia (holder risk appetite). It then uses exponential amplification (e.g., in the form of e^(w*x)) to enhance extreme risk signals. This allows for the precise capture of micro-risk characteristics.

[0217] It's important to note that determining jump points based on adjacent quantile spreads avoids the mean-masking effect of equal-frequency binning, allowing risk band boundaries (e.g., high-risk, medium-risk, and low-risk) to naturally align with data distribution characteristics. The introduction of a window retracement coefficient (e.g., weighted by the inverse of historical volatility) makes risk band boundaries adaptable to time variations.

[0218] In this embodiment, a priority vulnerability index is calculated for each bill. This index is a composite of the remaining days, historical discounts, cross-guarantee levels, and margin call / early redemption inertia, amplified by a preset weighting index. Bills are sorted from high to low by priority vulnerability, and jump points are determined based on the difference between adjacent quantiles. Risk band boundaries are then formed using a window reversion coefficient. Within a single band, bills are sorted in descending order of face value, and penalty points are assigned to double-default resonance records within the risk-free net balance window. The jump solution is then output. Exchanged bills must meet denomination parity constraints and are only allowed to transition between the current band and adjacent bands. By constructing a three-tiered architecture of "micro-feature quantification - dynamic stratification - intelligent optimization," a closed-loop process is achieved for the entire bill portfolio, from risk identification to active management, thereby improving bill processing efficiency.

[0219] Figure 7 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 6 ,like Figure 7 As shown, the specific implementation steps of the above S206 include:

[0220] S701: According to the jump solution group, instantiation processing is performed to obtain a hot and cold dual-temperature bitterfish network.

[0221] The hot-cold dual-temperature bitterfish network includes a cold layer for performing local search and a hot layer for performing global search.

[0222] S702. Perform iterative sorting on the compliant candidate bills according to the hot-cold dual-temperature bitter fish network to output a near-priority sequence through the hot-cold dual-temperature bitter fish network.

[0223] For example, import the jump solution group {S (b,0)} and the risk zone adjacency graph Instantiate the bitter fish individual node, transition channel and dual temperature indicators.

[0224] Specifically, for each initial sequence S (b,0) Calculate the baseline energy:

[0225]

[0226] Among them, F q is the face value of the qth note in the sequence; π ql is a tensor The scenario weight of the corresponding bill pair in (·) + represents the positive part operator.

[0227] Among them, the reference energy provides a scale for subsequent temperature distribution, and the system is based on θ c <θ h Initialize cold and hot temperatures.

[0228] Alternatively, in the cold layer, the bitter fish individuals only follow the inner edge ε of the same zone. b Perform fine-tuning so that the transition probability satisfies the deformed normal kernel:

[0229]

[0230] Where ΔF ij =F i -F j ,σ b is the variance of the amount within the band, is the vulnerability penalty factor, For note i in subgraph neighborhood.

[0231] Among them, the cold layer quickly approaches the local minimum of the denomination using gradient descent, and records the convergence trajectory in real time.

[0232] Furthermore, when continuous v c Denomination reduction rate of generation iteration Below the threshold ε flat When the current optimal sequence is locked as the cold layer snapshot S (c)And push it to the hot layer to prevent low temperature oscillation from consuming extra rounds.

[0233] Furthermore, the hot layer reads the collaborative tensor To S (c) Perform risk scans to identify high-resonance clusters

[0234] Among them, if Then the temperature is raised dynamically:

[0235]

[0236] Furthermore, enabling the cross-band transition edge ε pq (|pq|=1) Take a big leap and quickly escape from the risk trap.

[0237] Optionally, after each transition, the system calculates the composite energy based on the denomination, the worst coverage gap, and the resonance penalty:

[0238]

[0239] Among them, CR min is the worst double default coverage of the current sequence, Ψ is the tensor resonance density, ω gap ,ω res is the weight.

[0240] Among them, only Or the Metropolis criterion will only be adopted when the probability of temperature acceptance is met, ensuring that the amount reduction is synchronized with the risk dilution.

[0241] Finally, the double temperature layer is alternately executed to continuous v h wheel And CR min ≥θ cov When , the training is terminated. At this time, the network weight {W c ,W h} and output the global near-priority sequence S ★ ,The network instance is registered as a reusable optimizer, supporting subsequent event-driven online retraining.

[0242] It should be noted that the hot layer achieves a large cross-band jump by raising its temperature, breaking through the local optimal trap of the cold layer. Collaborative tensor scanning can identify hidden high-resonance clusters, reducing the portfolio's value at risk compared to traditional simulated annealing algorithms.

[0243] It's important to note that the alternating dominance of the dual-thermal layer energy function enables the portfolio to automatically adapt to market conditions. During periods of low volatility, the cold layer dominates, increasing returns; during periods of high volatility, the hot layer activates, reducing the portfolio's maximum drawdown. Large jumps in the hot layer, combined with the Metropolis criterion, proactively accept short-term inferior solutions under stressful circumstances, preventing the portfolio from experiencing liquidity depletion. The dual-thermal layer energy trajectory (cold layer convergence curve / hot layer jump path) forms a traceable log of the optimization process.

[0244] In this example, the jump solution group is instantiated as a "cold-hot" dual-temperature bitterfish network. A baseline energy is calculated for each sequence. The cold layer fine-tunes the ranking using a deformed normal kernel only along the intra-band edges until the rate of decrease in the amount falls below a threshold. The currently optimal sequence is then pushed to the hot layer. The hot layer scans the collaborative tensor to identify highly resonant clusters. After the temperature is raised, large jumps are made along the cross-band edges, and new sequences are accepted or rejected using the Metropolis criterion. The dual-temperature layers iterate alternately until the global energy curve stabilizes, outputting a near-optimal ranking sequence. By deeply integrating simulated annealing and collaborative filtering, an optimization engine with biological intelligence characteristics is constructed, thereby improving bill processing efficiency.

[0245] Figure 8 A schematic diagram of a bill processing method provided in an embodiment of the present application Figure 7 ,like Figure 8 As shown, the above S208 further includes:

[0246] It should be noted that the specific implementation methods of steps S207 and S208 are:

[0247] Step 1: The scheduling module calls the knowledge graph subscription interface before each iteration to batch pull new regulatory penalties, credit rating adjustments, and bill transfer events that have occurred since the previous round.

[0248] The pulling process is strictly sorted using timestamps and then written into the event buffer. The scheduling module also records the capture window [t last ,t now ] and version number v S5 , ensuring that external asynchronous information is aligned with the clock domain within the pledge business, providing a unique timing anchor point for subsequent mapping.

[0249] Step 2: The event parser reads the cache and stores the event in the enhanced semantic graph G. S The time index edge in the target bill node P is located one by one i or accepting bank node A j .

[0250] Furthermore, each event is converted into an incremental risk label according to the event-intensity dictionary. A mechanism of incremental overlay and historical trace retention is employed: new risk control fields are injected into node attributes. Meanwhile, old values ​​are shifted back along the timeline to ensure complete traceability of the causal chain.

[0251] Step 3: The tensor mapping service receives the labeled node update request and translates the risk increment of the node layer into a collaborative tensor scene increments.

[0252] Among them, the slice positioning + local writing strategy is adopted during mapping, and overwriting is performed only on the row-column plane where the affected bills are located.

[0253] Furthermore, the scene index table is refreshed synchronously to avoid the loss of real-time performance caused by recalculating the entire tensor. After the update is completed, the tensor version number is increased to v S6 , output incremented tensor

[0254] Step 4: Optimize the controller Pushed to the dual-temperature bitter fish network, the network first calls the energy evaluation module to re-evaluate the composite energy potential of all the sequences under study

[0255] Among them, for sequences containing high-risk incremental bills, the system automatically increases the resonance penalty weight and correspondingly increases the thermal layer temperature θ h , forming a new energy terrain

[0256] Furthermore, the cooling window of the affected nodes is locked in the cold layer to prevent them from being selected into the cold layer again in the short term, and to guide the search path away from the latest risk center.

[0257] Step 5: The network performs a complete cold adjustment-hot jump alternation in the latest energy field.

[0258] Specifically, the cold layer fine-tunes the sequence with the updated deformed normal transition kernel to quickly absorb the potential for denomination reduction; the hot layer enables cross-band large-step jumps based on the new high temperature threshold to replace the high covariance sequence.

[0259] During the entire process, the system monitors the three trajectories of total denomination, double default coverage ratio and collaborative tensor resonance density in real time to ensure that the sequence converges towards the triple goals of smaller amount, more stable coverage and weaker resonance until the global energy decline curve stabilizes.

[0260] Step 6: If the absolute value of the energy improvement for multiple consecutive iterations is lower than the threshold ε stop , and all regulatory constraints If the condition still satisfies the requirement, the system outputs the current sequence and marks it as the latest global near-optimal solution. Otherwise, the control flow returns to step 1, continues to pull events and triggers the next round of adaptive optimization until the termination condition is met or the maximum iteration limit set by the system is reached.

[0261] S801. Verify the target near-priority sequence based on the face value data and guarantee data of the bills to be processed in the target near-priority sequence to obtain a verification result.

[0262] S802. If the verification is successful, the bill pledge pool exit order is generated according to the target near-priority sequence.

[0263] For example, the global near-priority sequence generated in S208 is retrieved. Write the remaining face value of each bill, the corresponding exposure value of the accepting bank and the guarantee compensation amount into the priority verification cache at one time

[0264] Furthermore, the cache synchronization records the version number v S6 and timestamp, locking all numeric fields used in the evaluation.

[0265] Furthermore, any two bills are selected in order from the front to the back, their remaining face value and guarantee offset amount are accumulated in real time, and the current coverage value is calculated and compared with the regulatory threshold θ cov Make a dynamic comparison.

[0266] Furthermore, the minimum coverage generated during the enumeration process is continuously written into the same memory table to form a high-performance online minimum register.

[0267] Optionally, if the coverage falls below θ for the first time during the enumeration process cov The engine immediately interrupts subsequent traversal and writes the corresponding bill pair, sequence position, and number of gaps in the run log when triggered. This immediate breakpoint-based processing eliminates the remaining meaningless computing load and provides coordinates for subsequent weak link location.

[0268] Optional, if exhaustive enumeration is completed and the minimum coverage is always not less than θ cov The system determines that the current ranking meets the worst double default requirements and immediately The token is marked as ready to be withdrawn from the pool. The pledge management system interface is then called to solidify the priority table, write back the pledge status, and archive the risk control files, allowing the bill pool to enter the actual implementation phase.

[0269] Optionally, if insufficient coverage exists, the verification engine aggregates all triggering bill pairs recorded in the log, using the coverage gap as the primary key, calculates the gap contribution, and outputs a minimum default inferior solution subset. This subset only includes bills that actually caused the threshold to be breached, avoiding amplifying the scope of reflux.

[0270] Furthermore, the minimum default inferior solution subset and corresponding gap markers are sent back to the compliance pruning service in S204, requesting an increase in the risk weights for the relevant collaborative tensor slices during the next threshold update, and dynamically tightening the accepting bank-region concentration threshold. Through tensor-pruning coupled adjustments, bill portfolios with overlapping risks are weakened.

[0271] Finally, after the pruning strategy is executed, the next round of S204-S208 is automatically triggered. If the coverage requirement is still not met after reaching the set iteration limit, a risk warning is issued to management, along with a detailed list of the minimum default inferior solution subset, for manual intervention or business parameter adjustment. After receiving manual resolution feedback or parameter adjustment instructions, the process is restarted to ensure that the pledge pool always meets regulatory requirements and maintains a dynamic balance between funds and risks.

[0272] In this embodiment, the face value and collateral offset of any two bills are enumerated in real time, from the earliest to the latest priority, and the minimum coverage ratio is recorded. If the coverage ratio falls below the regulatory threshold for the first time, the triggering bill pair is located and the minimum default inferior solution subset is formed, which is then written back to the pruning service. If the worst-case double default coverage ratio requirement is still met after the full enumeration, the current priority sequence is solidified as the pledge pool exit priority and written back to the pledge management system. By establishing a three-layer protection system of "real-time enumeration, dynamic pruning, and intelligent solidification," bill pledge pool management is shifted from static compliance to dynamic risk management, thereby improving bill processing efficiency.

[0273] Figure 9 A schematic diagram of a bill processing device provided in an embodiment of the present application is shown in FIG. Figure 9 As shown, the device includes: a first acquisition module 91, a generation module 92, a first determination module 93, a screening module 94, a second determination module 95, a first sorting module 96, a pulling module 97 and a second sorting module 98.

[0274] The first acquisition module 91 is used to acquire a data set of bills to be processed; wherein the data set of bills to be processed includes bill data of a plurality of bills to be processed;

[0275] A generation module 92 is configured to generate a corresponding enhanced semantic graph based on the bill dataset to be processed; wherein the nodes of the enhanced semantic graph include bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes; the edges of the enhanced semantic graph include backbone topological edges composed of nodes and supplementary association edges; the supplementary semantic edges include guarantee relationship association edges, macroeconomic association edges, and regulatory event association edges;

[0276] A first determination module 93 is configured to determine a corresponding collaborative tensor model according to the enhanced semantic graph;

[0277] A screening module 94 is configured to screen the bill data based on the enhanced semantic graph and the collaborative tensor model to obtain a dataset of compliant candidate bills, wherein the dataset includes bill data of a plurality of compliant candidate bills;

[0278] A second determination module 95 is configured to determine a corresponding jump solution group based on the enhanced semantic graph and the compliant candidate bill dataset;

[0279] The first sorting module 96 is used to sort the candidate bills by bill risk according to the starting solution group to obtain a corresponding near-priority sequence;

[0280] Pull module 97, used to pull trigger events;

[0281] The second sorting module 98 is used to update the enhanced semantic graph and the collaborative tensor model according to the triggering event, and update the near-priority sequence according to the updated enhanced semantic graph and the updated collaborative tensor model to obtain the target near-priority sequence, wherein the target near-priority sequence is used to indicate the priority execution order of the bills to be processed.

[0282] In one possible design, based on the invoice dataset to be processed, a corresponding enhanced semantic graph is generated, including:

[0283] The generating module 92 is further configured to perform data preprocessing on the bill data of the bill to be processed to obtain preprocessed bill data; wherein the data preprocessing includes multiple processes of presetting field transformation matrix processing, normalization processing, deduplication processing, composite hash key generation processing, and deduplication processing;

[0284] Extract bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes based on pre-processed bill data;

[0285] Determine the backbone topology edges and supplementary associated edges based on bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes;

[0286] An enhanced semantic graph is generated based on bill nodes, accepting bank nodes, region nodes, industry nodes and event nodes, as well as backbone topological edges and supplementary association edges.

[0287] In one possible design, the corresponding collaborative tensor model is determined based on the enhanced semantic graph, including:

[0288] The first determining module 93 is further configured to extract subgraphs from the enhanced semantic graph to obtain an accepting bank-bill subgraph;

[0289] Perform graph convolution on the accepting bank-bill subgraph to obtain the direct credit vector corresponding to the bill to be processed;

[0290] According to the accepting bank-bill subgraph and the direct credit vector, the attention weights associated with the regional nodes and industry nodes are determined, and based on the attention weights, the cross-layer bill vector corresponding to the bill to be processed is obtained;

[0291] According to the cross-layer bill vector, perform dimensionality reduction mapping processing and default geometric semantics preservation processing to obtain the bill low-dimensional feature vector corresponding to the bill to be processed;

[0292] According to the low-dimensional feature vector of the bill, slicing processing is performed, and impact increments are written locally for the slice to obtain a collaborative tensor model.

[0293] In one possible design, the corresponding jump solution group is determined based on the enhanced semantic graph and the compliant candidate bill dataset, including:

[0294] The second determining module 95 is further configured to obtain bill data of a compliant candidate bill based on the compliant candidate bill dataset;

[0295] Calculate the ranking vulnerability index corresponding to the compliant candidate bill based on the enhanced semantic graph and the bill data of the compliant candidate bill;

[0296] Determine the risk zone boundaries based on the sequential vulnerability indicators;

[0297] Determine the take-off solution group based on the risk zone boundary.

[0298] In one possible design, the compliant candidate bills are ranked by bill risk based on the starting solution group to obtain a corresponding near-priority sequence, including:

[0299] The first sorting module 96 is further configured to instantiate the jump solution group to obtain a hot-cold dual-temperature bitterfish network; wherein the hot-cold dual-temperature bitterfish network includes a cold layer for performing local search and a hot layer for performing global search;

[0300] An iterative sorting process is performed on the compliant candidate bills according to the hot-cold dual-temperature bitter fish network to output a near-priority sequence through the hot-cold dual-temperature bitter fish network.

[0301] In one possible design, the bill data is screened and processed based on the enhanced semantic graph and collaborative tensor model to obtain a dataset of compliant candidate bills, including:

[0302] The screening module 94 is further configured to determine an accepting bank-region exposure matrix based on the enhanced semantic graph;

[0303] Performing a first screening process on the bill data according to the accepting bank-region exposure matrix to obtain a first candidate bill data set;

[0304] Based on the enhanced semantic graph, risk scenario labels, violation semantic distances, and region labels are determined;

[0305] According to the collaborative tensor model, the first candidate bill dataset is subjected to a second screening process in combination with the risk scenario label, violation semantic distance, and region to obtain a second candidate bill dataset;

[0306] performing a second screening process on the first candidate bill dataset according to the collaborative tensor model to obtain a second candidate bill dataset;

[0307] Obtaining the accumulated denomination data and guarantee data of the bills to be processed in the second candidate bill data set;

[0308] According to the collaborative tensor model, the second candidate bill dataset is subjected to a third screening process in combination with the accumulated denomination data and the guarantee data to obtain a compliant candidate bill dataset.

[0309] In a possible design, after updating the enhanced semantic graph and the collaborative tensor model according to the triggering event, and updating the near-priority sequence according to the updated enhanced semantic graph and the updated collaborative tensor model to obtain the target near-priority sequence, the following further steps are included:

[0310] Verifying the target near-priority sequence based on the face value data and guarantee data of the bills to be processed in the target near-priority sequence to obtain a verification result;

[0311] If the verification is successful, the bill pledge pool exit order will be generated based on the target near-priority sequence.

[0312] In one possible design, the bill data includes the face value, issue date, maturity date, issuer, accepting bank, and complete endorsement chain.

[0313] The present embodiment provides a bill processing device that can execute a bill processing method of the above embodiment. Its implementation principle and technical effects are similar and will not be described in detail in this embodiment.

[0314] In the specific implementation of the aforementioned method for processing bills, each module can be implemented as a processor, and the processor can execute computer-executable instructions stored in the memory, so that the processor executes the aforementioned method for processing bills.

[0315] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 10 As shown, the electronic device 100 includes: at least one processor 101 and a memory 102. The electronic device 100 also includes a communication component 103. 101, the memory 102 and the communication component 103 are connected via a second bus 104.

[0316] In a specific implementation process, at least one processor 101 executes the computer execution instructions stored in the memory 102, so that at least one processor 101 executes a bill processing method as executed by the electronic device side above.

[0317] The specific implementation process of the processor 101 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.

[0318] 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.

[0319] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk storage.

[0320] The second 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 as address buses, data buses, control buses, etc. 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.

[0321] The above-mentioned functions implemented by the electronic device and the main control device have introduced the solutions provided by the embodiments of the present invention. It can be understood that in order to implement the above-mentioned functions, the electronic device or the main control device includes hardware structures and / or software modules corresponding to the execution of each function. In combination with the units and algorithm steps of the various examples described in the embodiments disclosed in the embodiments of the present invention, the embodiments of the present invention can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the technical solution of the embodiments of the present invention.

[0322] 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, it is used to implement the above-mentioned bill processing method.

[0323] 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.

[0324] 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 as discrete components in an electronic device or a main control device.

[0325] The present application also provides a computer program product, which includes: a computer program, which is stored in a readable storage medium. At least one processor of an electronic device can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the electronic device executes the solution provided by any of the above embodiments.

[0326] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented using hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0327] So far, the technical solution of the present application has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the scope of protection of the present application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solution of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, ordinary technicians in this field should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solution to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for processing bills, characterized in that: include: Obtaining a data set of bills to be processed; wherein the data set of bills to be processed includes bill data of a plurality of bills to be processed; Generate a corresponding enhanced semantic graph based on the bill dataset to be processed; wherein the nodes of the enhanced semantic graph include bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes; the edges of the enhanced semantic graph include backbone topological edges and supplementary association edges composed of the nodes; the supplementary semantic edges include guarantee relationship association edges, macroeconomic association edges, and regulatory event association edges; Determining a corresponding collaborative tensor model according to the enhanced semantic graph; The bill data is screened and processed according to the enhanced semantic graph and the collaborative tensor model to obtain a compliant candidate bill dataset; wherein the compliant candidate bill dataset includes bill data of a plurality of compliant candidate bills; Determining a corresponding jump solution group according to the enhanced semantic graph and the compliant candidate bill dataset; According to the jump solution group, the compliant candidate bills are sorted by bill risk to obtain a corresponding near-priority sequence; Pull trigger event; According to the triggering event, the enhanced semantic graph and the collaborative tensor model are updated, and according to the updated enhanced semantic graph and the updated collaborative tensor model, the near-priority sequence is updated to obtain a target near-priority sequence, wherein the target near-priority sequence is used to indicate the priority execution order of the bills to be processed.

2. The method according to claim 1, characterized in that Generating a corresponding enhanced semantic graph according to the bill dataset to be processed includes: Performing data preprocessing on the bill data of the bill to be processed to obtain preprocessed bill data; wherein the data preprocessing includes multiple processes of preset field transformation matrix processing, normalization processing, deduplication processing, composite hash key generation processing, and deduplication processing; Extracting bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes based on the pre-processed bill data; Determine the backbone topology edges and supplementary associated edges based on the bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes; The enhanced semantic graph is generated based on the bill nodes, accepting bank nodes, region nodes, industry nodes and event nodes, as well as the backbone topology edges and the supplementary association edges.

3. The method according to claim 1, characterized in that Determining a corresponding collaborative tensor model according to the enhanced semantic graph includes: Performing subgraph extraction on the enhanced semantic graph to obtain an accepting bank-bill subgraph; Performing graph convolution processing on the accepting bank-bill subgraph to obtain a direct credit vector corresponding to the bill to be processed; Determining attention weights associated with the region node and the industry node based on the accepting bank-bill subgraph and the direct credit vector, and obtaining a cross-layer bill vector corresponding to the bill to be processed based on the attention weights; Performing dimensionality reduction mapping processing and default geometric semantics preservation processing according to the cross-layer bill vector to obtain a bill low-dimensional feature vector corresponding to the bill to be processed; According to the low-dimensional feature vector of the bill, slicing processing is performed, and impact increments are written locally for the slice to obtain the collaborative tensor model.

4. The method according to claim 1, wherein The determining of a corresponding jump solution group according to the enhanced semantic graph and the compliant candidate bill dataset includes: Obtaining bill data of compliant candidate bills based on the compliant candidate bill dataset; Calculating a ranking vulnerability index corresponding to the compliant candidate bill based on the enhanced semantic graph and the bill data of the compliant candidate bill; Determining risk zone boundaries based on the ranked vulnerability indicators; The take-off solution group is determined according to the risk zone boundary.

5. The method according to claim 4, characterized in that The step of performing bill risk ranking processing on the compliant candidate bills according to the jump solution group to obtain a corresponding near-priority sequence includes: According to the jump solution group, instantiate processing is performed to obtain a hot-cold dual-temperature bitter fish network; wherein the hot-cold dual-temperature bitter fish network includes a cold layer for performing local search and a hot layer for performing global search; An iterative sorting process is performed on the compliant candidate bills according to the hot and cold dual-temperature bitter fish network to output the near-priority sequence through the hot and cold dual-temperature bitter fish network.

6. The method according to any one of claims 1 to 5, characterized in that The step of screening the bill data according to the enhanced semantic graph and the collaborative tensor model to obtain a compliant candidate bill dataset includes: Determining an accepting bank-region exposure matrix based on the enhanced semantic graph; performing a first screening process on the bill data according to the accepting bank-region exposure matrix to obtain a first candidate bill data set; Determining risk scenario labels, violation semantic distances, and region labels based on the enhanced semantic graph; performing a second screening process on the first candidate bill dataset according to the collaborative tensor model, in combination with the risk scenario label, the violation semantic distance, and the region, to obtain a second candidate bill dataset; performing a second screening process on the first candidate bill dataset according to the collaborative tensor model to obtain a second candidate bill dataset; Obtaining the accumulated denomination data and guarantee data of the bills to be processed in the second candidate bill data set; According to the collaborative tensor model, combined with the accumulated denomination data and the guarantee data, a third screening process is performed on the second candidate bill dataset to obtain the compliant candidate bill dataset.

7. The method according to any one of claims 1 to 5, characterized in that After updating the enhanced semantic graph and the collaborative tensor model according to the trigger event, and updating the near-priority sequence according to the updated enhanced semantic graph and the updated collaborative tensor model to obtain a target near-priority sequence, the method further includes: Verifying the target near-priority sequence based on the denomination data and guarantee data of the bills to be processed in the target near-priority sequence to obtain a verification result; If the verification is successful, the bill pledge pool exit order is generated based on the target near-priority sequence.

8. The method according to any one of claims 1 to 5, characterized in that The bill data includes the face value, issue date, maturity date, issuer, accepting bank and complete endorsement chain.

9. A bill processing device, characterized in that: include: A first acquisition module is configured to acquire a data set of bills to be processed; wherein the data set of bills to be processed includes bill data of a plurality of bills to be processed; A generation module is configured to generate a corresponding enhanced semantic graph based on the bill dataset to be processed; wherein the nodes of the enhanced semantic graph include bill nodes, accepting bank nodes, region nodes, industry nodes, and event nodes; the edges of the enhanced semantic graph include backbone topological edges and supplementary association edges formed by the nodes; and the supplementary semantic edges include guarantee relationship association edges, macroeconomic association edges, and regulatory event association edges; A first determining module is used to determine a corresponding collaborative tensor model according to the enhanced semantic graph; a screening module, configured to screen the bill data according to the enhanced semantic graph and the collaborative tensor model to obtain a dataset of compliant candidate bills; wherein the dataset of compliant candidate bills includes bill data of a plurality of compliant candidate bills; A second determination module is configured to determine a corresponding jump solution group based on the enhanced semantic graph and the compliant candidate bill dataset; A first sorting module is configured to perform bill risk sorting on the compliant candidate bills according to the jump solution group to obtain a corresponding near-priority sequence; Pull module, used to pull trigger events; The second sorting module is used to update the enhanced semantic graph and the collaborative tensor model according to the triggering event, and update the near-priority sequence according to the updated enhanced semantic graph and the updated collaborative tensor model to obtain a target near-priority sequence, wherein the target near-priority sequence is used to indicate the priority execution order of the bills to be processed.

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 the computer program is executed by a processor.

Images