A workflow processing method, system, device and medium based on a routing algorithm

Abstract

The application provides a kind of workflow processing method, system, equipment and medium based on routing algorithm, belong to workflow approval technical field, specifically include: after obtaining the approval request transmitted by API gateway, build semantic atlas to extract entity relationship, combine the dynamic weight calculated by real-time state and historical information of approval terminal, generate feature vector;Analysis request basic data, based on basic data again build semantic atlas and generate feature vector by fusing terminal state, historical record;Through rule engine priority layer processing basic data and feature vector, get preliminary routing judgment;Then determine the optimal path after adjusting the routing value through DQN optimization layer;Finally, execute the approval process and push the task, receive feedback to generate visual report, form closed loop optimization. Through multi-layer processing and optimization, improve the accuracy and efficiency of approval routing, reduce congestion, enhance process stability, realize visual report to improve transparency, overall optimization workflow processing.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of workflow approval, and particularly relates to a workflow processing method, system, device and medium based on a routing algorithm. BACKGROUND

[0002] In some financial industries or manufacturing industries, such as banks, insurance companies, and credit approval, claim settlement, risk control processes of securities institutions often involve multiple links such as customer information, customer managers, risk control, law, and loan / claims. It takes a long time to complete an approval link, the experience is poor, and traces are left throughout the process. For manufacturing industries, equipment procurement, spare parts procurement, and quality abnormality processing need to cross factories, supply chains, quality, logistics, and multiple departments of finance, which affects the normal production order.

[0003] Overall, in the enterprise cross-department approval scene, the existing technology is difficult to process approval requests of different systems. For example, the formats of reimbursement approval of the finance system and leave approval of the personnel system are quite different. The system often causes data parsing errors due to format incompatibility after receiving, which reduces the overall work efficiency. Traditional routing decisions are only based on surface data, and the routing is determined according to the approval amount, ignoring the implicit association. Taking project approval as an example, a project involves multiple sub-contracts, and the existing technology may route each sub-contract approval separately without considering the overall association of the project, resulting in an incomplete approval process and an inability to fully grasp the project situation during approval, which is prone to approval judgment deviation. When assigning approval tasks, the approval terminal state is not dynamically monitored. For example, a certain approval terminal has accumulated a large number of high-priority tasks, and the system still assigns new tasks, while other low-load approval terminals are idle, causing resource waste and delaying new task processing, and even causing task blocking, affecting business progress. SUMMARY

[0004] The application provides a workflow processing method based on a routing algorithm, which improves the accuracy and efficiency of approval routing, reduces blocking, enhances process stability, realizes visual reporting to improve transparency, and overall optimizes workflow processing through multi-layer processing and optimization.

[0005] The method comprises the following steps:

[0006] S101: Obtain an approval request transmitted through an API gateway, build a semantic graph of the request content to extract entity relationships, obtain real-time working status of an approval terminal, combine historical information of the request, calculate dynamic weights through collaborative analysis, and obtain a feature vector;

[0007] S102: Analyze the approval request to parse out basic data;

[0008] S103: Based on the basic data, a semantic graph of the request text content is constructed to extract entity relationships, the real-time working status of the approval terminal is obtained according to the associated approval terminal information, the historical approval records are combined, the weight is calculated through collaborative analysis, and the feature vector is obtained;

[0009] S104: The basic data obtained by analysis and the extracted feature vector are processed by the rule engine priority layer to obtain the approval routing judgment;

[0010] S105: In the hybrid decision routing, through the DQN optimization layer, combined with the congestion risk prediction, the content processed by the rule engine priority layer or the basic data obtained by analysis and the feature vector are calculated for the routing value after risk adjustment to determine the optimized routing path;

[0011] S106: According to the path determined by the hybrid decision routing, the approval process is executed and the task is pushed, the operation feedback is received, and the workflow visualization report is generated.

[0012] It should be further explained that step S102 specifically includes:

[0013] The approval request is parsed to extract the approval amount field, the application reason description, the request initiation time, and the source system identifier to form an original data set;

[0014] Based on the pre-defined approval data rule library, the parsed structured data is contextually associated and verified;

[0015] The intent of the original data set content is recognized, the key business elements are extracted, and the urgency label and involved department list are generated in combination with the department affiliation information of the associated approval terminal;

[0016] The parsed basic data is verified, and when a key field is missing, a correction process is performed to fill in the missing information through a pre-set default value or associate historical data to complete the missing information.

[0017] It should be further explained that step S103 specifically includes:

[0018] Based on the parsed basic data, the explicit entity relationships and implicit business associations in the text content are extracted in combination with the approval data rule library and the semantic graph;

[0019] According to the involved department information in the basic data, the real-time working status of the associated approval terminal is collected, and the department collaboration mode data is associated;

[0020] From the historical approval records, the historical cases matching the current request are filtered, the average processing time of the approval nodes in each case, the abnormal interruption ratio, and the user feedback score are counted to generate a historical efficiency feature set;

[0021] The entity relationship of the semantic graph, the real-time working state index set, and the historical efficiency feature set are input into the collaborative analysis module, the weight distribution of the features is adjusted, and a feature vector that fuses content correlation, resource availability, and historical reliability is generated.

[0022] It should be further explained that step S104 specifically includes:

[0023] Based on the basic data parsed in step S102 and the feature vector extracted in step S103, an index of the approval data rule library is constructed, and is stored in categories according to business types and approval levels, and main rules and auxiliary rules are marked;

[0024] The basic data and the feature vector are preprocessed for rule matching, and key parameters related to the rules in the approval data rule library are extracted;

[0025] The features are preliminarily matched through the main rules, and candidate routes that meet the main conditions are screened out;

[0026] The matching results are subjected to contradiction detection, and if there are multiple routes that meet the conditions, the final approval route is determined according to the rule priority.

[0027] It should be further explained that step S1044 specifically includes:

[0028] The candidate route set after rule matching is subjected to contradiction feature extraction, the contradiction types are identified, and the contradiction dimensions of each candidate route are marked;

[0029] Based on historical decision information and the current business scenario, the weights of the contradictory rules are reordered;

[0030] For candidate routes with priority contradictions, a weighted scoring method is used to calculate the comprehensive scores, and the route with the highest comprehensive score is selected as the final approval path;

[0031] For no matching request, the abnormal features in the basic data and the key information in the feature vector are extracted, and are marked as an abnormal feature request marker code, with a contradiction reason explanation and associated parameters attached, and an approval route judgment result containing the marked information is output.

[0032] It should be further explained that step S105 specifically includes:

[0033] The approval route judgment result output in step S104 is received, and the basic data parsed in step S102 and the feature vector extracted in step S103 are obtained, and the three types of information are integrated into a decision input data set, and key influence information in the decision input data set is marked;

[0034] Based on the input decision input data set, a blocking risk assessment dimension is constructed, and a risk assessment matrix is formed;

[0035] In the DQN optimization layer, risk matching is performed on the routes processed by the rule engine priority layer to calculate the blocking probability of the routes under the current risk evaluation matrix; for requests not processed by the rule engine, the blocking probability of the routes is calculated based on the basic data and the feature vector;

[0036] The candidate routes are adjusted in combination with the blocking probability, and a route path that meets the business target and has low blocking risk is preferentially selected; if there are multiple equivalent paths, the final optimized route path is determined according to the historical execution efficiency sorting, and the path details are output.

[0037] It should be further explained that step S1053 specifically includes:

[0038] A double-channel risk mapper is established in the DQN optimization layer, a first channel receives the marked route path, and a second channel receives a special request not processed by the rule definition, and a path state snapshot is generated respectively;

[0039] The real-time terminal heartbeat sequence and the cross-department time delay fingerprint are injected into each path state snapshot to form a dynamic risk fingerprint;

[0040] The adaptive risk measurer is called according to the risk fingerprint to output a blocking probability vector of each candidate path;

[0041] The blocking probability vector is coupled with the path business weight to generate a risk-adjusted route value indication and is cached to the decision log.

[0042] The application also provides a workflow processing system based on a routing algorithm, the system comprising:

[0043] A feature extraction module is configured to obtain an approval request transmitted through an API gateway, construct a semantic graph of the request content to extract entity relationships, obtain real-time working states of approval terminals, calculate dynamic weights through collaborative analysis in combination with historical request information, and obtain a feature vector;

[0044] A request analysis module is configured to analyze the approval request to obtain basic data;

[0045] A semantic fusion module is configured to construct a semantic graph of the request text content based on the basic data to extract entity relationships, obtain real-time working states of approval terminals according to associated approval terminal information, calculate weights through collaborative analysis in combination with historical approval records, and obtain a feature vector;

[0046] A priority judgment module is configured to analyze the obtained basic data and the extracted feature vector through the rule engine priority layer to obtain an approval route judgment;

[0047] A routing decision module is configured to, in the hybrid decision routing, combine the blocking risk prediction through the DQN optimization layer, and calculate the routing value of the content processed by the rule engine priority layer or the basic data and feature vector obtained by parsing after risk adjustment to determine the optimized routing path.

[0048] An execution feedback module is configured to execute the approval process and push the task according to the path determined by the hybrid decision routing, receive the operation feedback, and generate a workflow visualization report.

[0049] According to another embodiment of the present application, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the routing algorithm-based workflow processing method when executing the program.

[0050] According to another embodiment of the present application, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the routing algorithm-based workflow processing method when executing the program.

[0051] From the above technical solutions, the present application has the following advantages:

[0052] The method of the present application constructs a semantic graph to extract entity relationships and integrates industry relationships. In project approval, the correlation between sub-contracts and projects can be mined, related approval tasks are reasonably associated with the routing, the approval terminal can obtain comprehensive information, and the approval cycle is shortened. At the same time of obtaining the real-time state of the terminal, the routing is optimized in combination with risk prediction. The approval terminal load is balanced, the task accumulation is avoided, the resource utilization rate is improved, and the overall approval efficiency is improved. The SHAP analysis is used to generate a decision report, the user score is received and fed back to the model. The user can clearly understand the basis of the approval routing, the model is continuously optimized through feedback, unreasonable routing is reduced, and the decision accuracy is improved. The routing is directly calculated through the DQN optimization layer. The adaptive and rapid processing of special requests such as emergency projects is realized, the response is timely, the workflow approval is intelligent, efficient, and adaptive, and the approval efficiency and accuracy are improved. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0054] Figure 1 A routing algorithm-based workflow processing method flowchart;

[0055] Figure 2Flowchart of an embodiment of a routing algorithm based workflow processing method;

[0056] Figure 3 Flowchart of a further embodiment of a routing algorithm based workflow processing method;

[0057] Figure 4 Flowchart of a further embodiment of a routing algorithm based workflow processing method;

[0058] Figure 5 Schematic diagram of a routing algorithm based workflow processing system;

[0059] Figure 6 Schematic diagram of an embodiment of a routing algorithm based workflow processing system;

[0060] Figure 7 Schematic diagram of an electronic device. DETAILED DESCRIPTION

[0061] The method of the present application addresses the problem of rigid approval paths by using a dynamic routing algorithm based on reinforcement learning to optimize approval paths in real time. Intelligent load balancing is used in combination with real-time availability and historical response speed of approvers to make decisions, solving the problem of low approval efficiency and process blockage. An adaptive relationship graph is applied to learn and update the approval path, so that when the organizational structure changes, manual adjustment of approval nodes and rules is no longer needed, providing a basis for routing decisions and visual analysis.

[0062] The routing algorithm based workflow processing method according to the present application will be described in detail below. For the purpose of explanation and not limitation, specific details are set forth, such as specific system architectures, techniques, in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced in other embodiments that depart from these specific details.

[0063] It should be understood that when used in the specification, the term "comprises" indicates the presence of the described features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. The terms "comprise", "include", "have" and their conjugates mean "including but not limited to", unless otherwise specifically stated.

[0064] The language "one embodiment" or "some embodiments" appearing in the specification are used merely for a convenient description of certain features, structures, or characteristics with respect to one or more embodiments and are not necessarily referring to the same embodiment. Therefore, the language "one embodiment" or "some embodiments" does not necessarily mean the same embodiment in different places throughout the application.

[0065] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of the present application.

[0066] Please refer to Figure 1 As shown in FIG. 1, which is a flowchart of a workflow processing method based on a routing algorithm in an embodiment, the method comprises the following steps:

[0067] Step S101: Obtain an approval request transmitted through an API gateway, build a semantic graph of the request content to extract entity relationships, obtain real-time working status of an approval terminal, combine historical information of the request, calculate dynamic weights through collaborative analysis, and obtain a feature vector.

[0068] In some embodiments, after obtaining the approval request transmitted through the API gateway, the format and integrity of the request are first verified, and invalid or abnormal requests are removed. Subsequently, for the request content, a semantic graph containing multiple levels of entities and the association relationships between the entities is built through semantic analysis technology to accurately extract the entity relationships. At the same time, through the interface with the approval terminal system, real-time working status data of the terminal, such as the current task volume, response speed, and online state, are collected, and historical processing records of such requests are retrieved, including historical approval paths, processing time, approval results, and other information. The entity relationship features in the semantic graph, the real-time state features of the terminal, and the historical information features are collaboratively associated, the weight proportions of the features in the analysis are dynamically adjusted according to the business type of the request, and finally a feature vector that can comprehensively reflect the characteristics of the request and the processing environment is fused and generated.

[0069] As can be seen, the embodiment ensures the reliability of the data source by verifying the validity of the approval request, mines the internal entity association of the request content using the semantic graph technology, combines the real-time processing capacity of the approval terminal and the historical processing rules, and realizes the fusion of features through weight distribution to provide input data for the routing algorithm.

[0070] Step S102: Analyze the approval request to obtain basic data.

[0071] Optionally, the basic data includes but is not limited to: text content, approval amount, urgency label, involved department.

[0072] The embodiment analyzes the obtained approval request, identifies the source of the request, which can be a financial system, a personnel system, etc., performs format adaptation conversion according to the format characteristics of different systems, and unifies the data structure. The explicit basic data such as text content, approval amount, urgency label, and involved department are extracted from the request, and the implicit business attributes such as project type, contract period, and related business processes are mined from the text content through semantic analysis.

[0073] In some specific embodiments, step S102 specifically includes the following steps:

[0074] S1021: Data analysis is performed on the approval request to extract the approval amount field, application reason description, request initiation time, and source system identifier, and an original data set is formed.

[0075] It should be noted that data analysis can be based on BERT-Base model for sequence labeling to extract business elements. Related data of the original data set can also be obtained through API gateway log analysis, and finally fused into an original data set containing structured fields, text elements, and system identifiers.

[0076] S1022: Based on the pre-defined approval data rule library, the structured data extracted is verified in context. Illustratively, it is verified whether the approval amount is within the department authority range and whether the involved department is business-related to the application content, and basic data is generated.

[0077] Optionally, the approval data rule library includes a department authority table and a business relevance rule. When verifying, it is checked through SQL query whether the data meets the rule, and a pass / fail label is generated.

[0078] S1023: Intention recognition is performed on the content of the original data set to extract key business elements, and combined with the department attribution information of the associated approval terminal, an urgency label and an involved department list are generated.

[0079] It should be noted that the urgency label is generated by using a text classification model, the model input is a text vector of the application reason, and the output is a high / medium / low three-class label, and the classification threshold is adjusted according to the text features of historical urgent requests; the involved department list is generated by matching business elements through a department relationship graph.

[0080] S1024: The basic data obtained by parsing is verified, and when a key field is missing, a correction mechanism is triggered to fill in or associate historical data to complete the missing information.

[0081] The integrity check sets a set of key fields, such as amount, involved departments, and urgency level. When a field is missing, the default value is extracted first from the statistical pattern of historical requests of the same type. If it cannot be extracted, it is marked as pending correction and pushed to the manual review interface. After correction, the data is synchronized to the original dataset.

[0082] It can be seen that step S102 achieves compatible processing of approval requests from different source systems, improving the system's versatility. The extracted basic data contains both explicit information and implicit attributes, enriching the data dimensions and improving the efficiency of the entire workflow processing.

[0083] Step S103: Based on the basic data, construct a semantic graph of the request text content to extract entity relationships, obtain the real-time working status of the approval terminal according to the associated approval terminal information, and combine it with historical approval records to calculate weights through collaborative analysis to obtain feature vectors.

[0084] In some embodiments, based on the basic data parsed in step S102, the data is preprocessed to filter out business-related elements such as the project's domain and budget category, and business-related points in the data are marked. Based on the text content in the basic data, and combined with industry-specific relationships such as supplier-material-order relationships in procurement approvals, the scope of semantic graph construction is expanded, the entity relationship network is enriched, and entity relationships are extracted. Based on information such as the departments involved in the basic data, the associated approval terminals are determined, and the real-time working status of these terminals is obtained through multi-channel data acquisition. Simultaneously, historical approval records related to the approval request are retrieved, and features such as historical approval cycles, approval pass rates, and cross-departmental collaboration efficiency are extracted. These features are deeply correlated with the basic data and the real-time terminal status, and the dimensions and weights of collaborative analysis are adjusted to calculate feature vectors. This improves analysis efficiency and enhances the comprehensiveness of entity relationship extraction; collecting terminal status more accurately reflects terminal processing capabilities; and combining historical approval records ensures that the feature vectors contain historical experience information.

[0085] Step S104: The parsed basic data and extracted feature vectors are processed using the priority layer of the rule engine to obtain the approval route judgment.

[0086] In some embodiments, a rule base is constructed based on the basic data parsed in step S102 and the feature vector extracted in step S103. The rules are classified and stored according to financial approval, personnel appointment and dismissal approval, or departmental level and company level, and marked as such that large-amount fund approval requires general manager approval and cross-departmental approval requires confirmation from the department coordinator.

[0087] The basic data and feature vectors are preprocessed by rule matching, and the approval amount range and department authority level related to the rules in the rule library are extracted, so as to convert the unstructured data into structured parameters that can be directly matched. A hierarchical matching strategy is adopted, the structured parameters are preliminarily matched by the main rules, and the candidate routes that meet the main conditions are screened out; then the candidate routes are screened again by the auxiliary rules, and the routes that do not meet the rule requirements are excluded. Finally, the matching results are detected for conflicts, and if there are multiple routes that meet the conditions, the final approval route is determined according to the rule priority; if there is no matching rule, it is marked as a special request and processed by the subsequent optimization layer.

[0088] It can be seen that the construction of the classification rule library makes the rule management clearer and improves the efficiency of rule calling and matching; the hierarchical matching and conflict detection mechanism ensures the accuracy of route judgment, avoids decision-making errors caused by rule conflicts, and improves the efficiency of workflow processing.

[0089] Step S105: In the hybrid decision routing, the content processed by the rule engine priority layer or the basic data and feature vectors obtained by parsing are calculated for route value after risk adjustment by the DQN optimization layer, to determine the optimized route path.

[0090] The step S105 of the embodiment receives the approval route judgment result output by the step S104, obtains the basic data of the step S102 and the feature vector of the step S103, integrates the three types of information into a unified decision input data set, and marks the urgency and terminal load. Based on the data set, the blocking risk assessment dimension is constructed, including the historical blocking probability of the approval terminal, the current task queue length, the cross-department cooperation delay record, etc., to form a risk assessment matrix. In the DQN optimization layer, the blocking probability of the route processed by the rule engine is calculated under the current risk assessment matrix; the blocking probability of the potential route is calculated directly based on the basic data and feature vectors for the special request not processed by the rule engine. The value of each candidate route is adjusted in combination with the blocking probability, and the route path that meets the processing speed and pass rate with low blocking risk is preferentially selected; if there are multiple equivalent paths, the final optimized route path is determined according to the historical execution efficiency.

[0091] It can be seen that the multi-source data integration provides comprehensive input for risk assessment and route optimization; the blocking risk assessment makes the route decision consider the potential blocking problem, improving the stability of the workflow; the optimized route path can reduce the blocking risk while ensuring the efficiency, improving the reliability of the approval process.

[0092] Step S106: According to the path determined by the hybrid decision routing, the approval process is executed and the task is pushed, the operation feedback is received, and the workflow visualization report is generated.

[0093] In some embodiments, based on the optimized routing path determined in step S105, the approval process execution engine is started, the approval task is pushed to the corresponding approval terminal, and the execution status of the task is tracked in real time. Feedback from the approval terminal, such as approval, rejection, or return for modification, is received, and information such as approval time and approver is recorded.

[0094] This embodiment generates a workflow visualization report based on the execution status and operational feedback of the approval process, displaying the approval progress and processing status of each node. Simultaneously, SHAP analysis is used to analyze the routing decision-making process, generating a decision explanation visualization report that explains the basis and key influencing factors of the routing decisions. User ratings for the decision explanations are received, and the SHAP analysis results and user ratings are fed back to the user's terminal. In this way, SHAP analysis enhances the interpretability of the decisions, increasing user trust in the system; user rating feedback is used for model optimization, improving the accuracy and adaptability of routing decisions, and achieving continuous optimization.

[0095] In one embodiment of the present invention, based on step S103, the following will provide a possible embodiment and its specific implementation will be described in a non-limiting manner. For example... Figure 2 As shown, step S103 specifically includes:

[0096] S1031: Based on the parsed basic data, combined with the approval data rule base and semantic graph, extract explicit entity relationships and implicit business connections from the text content. Optionally, implicit business connections can be based on the implicit financial audit requirements of the procurement application.

[0097] It should be noted that semantic graph construction can rely on a combination of syntactic analysis and approval data rule base. The approval data rule base can contain approval contracts, invoices, approval roles and relationship templates. Relationships can involve contracts, associations, legal departments, etc. Approval relationships are generated through pattern matching and semantic role annotation, and the database data can be expanded according to new approval types.

[0098] S1032: Based on the departmental information in the basic data, collect the real-time working status of the associated approval terminals, including the current task queue length, the average processing time within the preset time period, the number of suspended tasks, and associate the departmental collaboration mode data.

[0099] In this embodiment, the real-time working status of the associated approval terminal can be collected by calling the load data interface of the corresponding approval terminal based on the information of the relevant departments. The collected content includes the current task queue length, the average processing time within the preset time period, and the number of suspended tasks. It also associates the departmental collaboration mode data to form a real-time status indicator set to assist in subsequent approvals.

[0100] S1033: Screen historical cases matching the current request in approval type, involved department, and amount interval from historical approval records, and statistically calculate the average processing time efficiency, abnormal interruption ratio, and user feedback score of each case, to generate a historical efficiency feature set including time efficiency, stability, and satisfaction.

[0101] The historical efficiency feature generation of the embodiment adopts a multi-condition matching rule to screen historical cases matching the current request in approval type, involved department, and amount interval, and statistically calculate the average processing time efficiency, abnormal interruption ratio, and user feedback score of each case, to generate a historical efficiency feature set.

[0102] S1034: Input the entity relationship of the semantic atlas, the real-time working state index set, and the historical efficiency feature set into a collaborative analysis module, adopt a business rule driven cross-dimension association strategy, adjust the weight distribution of the features, and generate a feature vector integrating content correlation, resource availability, and historical reliability.

[0103] The embodiment converts the basic data into quantifiable feature vectors through data fusion and adjustment, to provide input for hybrid decision routing. The weight distribution of the features is adjusted by pre-setting an emergency request priority to distribute low-load terminals, the features are standardized, the global weight is adjusted through association rules, and finally the feature vector is generated to meet the use requirements of users.

[0104] In an embodiment of the present application, based on step S104, a possible embodiment will be given below to non-restrictively describe the specific implementation scheme. As shown in the following formula (1), step S104 specifically includes: Figure 3

[0105] Step S1041: Based on the basic data analyzed in step S102 and the feature vector extracted in step S103, an index of the approval data rule library is constructed, and is stored in categories according to business type and approval level, and is marked with main rules and auxiliary rules.

[0106] The embodiment can establish an index by calculating the association degree of rules and business type and approval level; the association degree = rule application times / total rule times. The rule classification results are stored by using a hash table, to realize fast query and call of rules.

[0107] Step S1042: Perform rule matching preprocessing on the basic data and the feature vector, and extract key parameters related to the rules in the approval data rule library.

[0108] Step S1043: Perform preliminary matching on the features by the main rules, to screen out candidate routes meeting the main conditions; the candidate routes are sorted according to the matching degree = number of rule clauses met / total number of rule clauses; the auxiliary rules reserve the candidate routes meeting the conditions. ​

[0109] Step S1044: Contradiction detection is performed on the matching result. If there are multiple routes meeting the conditions, the final approval route is determined according to the rule priority.

[0110] If there is no matching rule, the request is marked as a request that needs to be processed by the optimization layer, and the approval route judgment result is output. Optionally, the rule priority can be that the amount rule is prior to the department rule.

[0111] The rule priority of the embodiment is determined by the analytic hierarchy process. When there is no matching rule, the request type is marked by an exception marking algorithm to ensure process connection. Optionally, a marking value = 1 represents a special request, and 0 represents a normal request.

[0112] The embodiment takes basic data and a feature vector as input, indexes, realizes fast positioning of rules, adopts a hierarchical matching strategy to accurately screen candidate routes, and solves the contradiction between multiple rules through a contradiction detection mechanism to finally output a clear approval route judgment result while ensuring the accuracy of route judgment under clear rules.

[0113] Step S1044 of the embodiment further comprises the following steps:

[0114] Step S10441: Contradiction feature extraction is performed on the candidate route set after rule matching, contradiction types are identified, and the contradiction dimension of each candidate route is marked.

[0115] Optionally, the contradiction type can involve rule type contradiction, business scenario contradiction, and priority contradiction. The contradiction dimension can involve the simultaneous triggering of the amount rule and the department rule, the overlapping of the time range of the emergency rule and the regular rule, and the like.

[0116] Step S10442: Based on historical decision information and the current business scenario, the weight of the contradictory rule is reordered.

[0117] It should be noted that the historical decision information can consider the approval pass rate and the average time consumption after the rule hits within a period of time. The current business scenario can be combined with the scene based on the actual approval state, such as holidays, sudden business peaks, and the like. The weight of the contradictory rule is reordered by improving the priority of the rule with a high pass rate and prioritizing the rule with a low time consumption in an emergency scenario.

[0118] The priority adjustment method of the embodiment can adjust the priority in combination with historical performance and scene characteristics.

[0119] Optionally, the priority score = 0.6 x historical priority score + 0.3 x hit pass rate in a preset time period + 0.1 x current scene adaptation coefficient. Here, the emergency scene coefficient = 1.2, and the daily scene coefficient = 1.0.

[0120] Wherein, the historical priority score is assigned according to a preset priority rule; the hit pass rate in a preset time period is equal to the number of times of passing after hitting / the total number of times of hitting; and the scene adaptation coefficient is adjusted according to a current business scene.

[0121] Step S10443: For the candidate routes with priority conflicts, a weighted scoring method is used to calculate a comprehensive score, and the route with the highest comprehensive score is selected as the final approval path.

[0122] Optionally, the weighted scoring method can use a linear weighted model to calculate the comprehensive score for the candidate routes with priority conflicts.

[0123] The specific configuration mode is comprehensive score = 0.5 x dynamic priority score + 0.3 x (historical pass rate / 100) + 0.2 x (1- average time consumption / base time consumption).

[0124] Wherein, the historical pass rate and the average time consumption are obtained based on data in a preset time period, for example, the historical pass rate = 90% is scored 0.9, the average time consumption = 2 days is scored 0.8, and the base time consumption = 3 days. The embodiment can also be adjusted according to business targets, for example, the time consumption coefficient is increased to 0.3 when efficiency is emphasized.

[0125] Step S10444: For the request without matching, the abnormal features in the basic data and the key information in the feature vector are extracted, and are marked as an abnormal feature request marker code, with an additional contradictory reason explanation and associated parameters, and an approval route judgment result containing the marked information is output.

[0126] Optionally, the abnormal feature request marker code is generated for the request without matching rules, and contains an abnormal type, a timestamp, associated parameters and a contradictory rule ID. The marker code format can be AAAA-Abnormal Type-Timestamp-Associated Parameter Hash Value, which facilitates quick positioning of problem scenes during subsequent model training.

[0127] The embodiment solves the problem of rigid route decision caused by rule conflicts or incomplete coverage of the rule engine through contradictory feature extraction, priority adjustment and scoring; and accurately marks the request without matching rules, realizes the process of rule matching, conflict resolution and problem marking, and improves the flexibility and accuracy of route decision.

[0128] In an embodiment of the present application, based on step S105, a possible embodiment will be given below to non-restrictively describe the specific implementation scheme. As shown in FIG. 10, step S105 specifically includes: Figure 4

[0129] ​Step S1051: Receive the approval route judgment result output by step S104, and obtain the basic data parsed by step S102 and the feature vector extracted by step S103. Integrate the three types of information into a unified decision input data set, and mark the key influence information in the data.

[0130] The key influence information can relate to the urgency, terminal load, etc.

[0131] The embodiment realizes information integration by calculating the field correlation degree of different source data; uses a key information extraction algorithm to mark key influence information based on the influence weight of the field on the route decision. The configuration method of the field correlation degree is correlation degree = number of common fields / total number of fields.

[0132] Step S1052: Based on the input data set, construct a congestion risk assessment dimension, including the historical congestion probability of the approval terminal, the current task queue length, and the cross-department coordination delay record, to form a risk assessment matrix.

[0133] The embodiment can construct a risk assessment method based on the analytic hierarchy process, and the risk weight of each dimension is determined by combining historical data verification; the terminal historical congestion probability, task queue length and other indicators are converted into risk values in the 0-1 interval. For example, if the queue length exceeds the threshold proportion x the basic risk coefficient.

[0134] Step S1053: In the DQN optimization layer, first perform risk matching on the route processed by the rule engine priority layer, calculate the congestion probability of the route under the current risk assessment matrix; for special requests that have not been processed by the rule engine, calculate the congestion probability of the potential route based on the basic data and the feature vector.

[0135] Optionally, the congestion probability calculation method is used to obtain a comprehensive probability by weighted summation of each risk dimension; for example, congestion probability = Σ (dimension risk value x dimension weight). Perform feature comparison between the route features and the risk assessment matrix to realize rapid risk matching.

[0136] Step S1054: Adjust each candidate route in combination with the congestion probability, and preferentially select a route path with low congestion risk and in line with the business goal; if there are multiple equivalent paths, determine the final optimized route path according to the historical execution efficiency sorting, and output the path details.

[0137] The route adjustment method used in the embodiment is based on the risk influence coefficient set according to the sensitivity of the business to congestion; the path sorting algorithm is used to sort the paths based on the weighted sum of the adjusted route and the historical execution efficiency, and determine the optimal route.

[0138] It can be seen that the embodiment receives and associates rule results, basic data and feature vectors; extracts risk information from the terminal state library and the historical database to construct an evaluation matrix; configures a DQN optimization layer processing unit, which has built-in risk matching and value adjustment logic, to perform risk accounting and sorting on candidate routes; sets a path decision module to determine the final route according to the risk value and historical efficiency and output the final route. Moreover, in the DQN optimization layer, the route is adjusted in combination with the blocking risk, the path with low risk and high efficiency is preferentially selected, adaptive routing decision is realized for special requests, and finally the optimized route path is output, thereby improving the stability and efficiency of the workflow approval.

[0139] For step S1053 of the embodiment, the following steps are further included:

[0140] Step S10531: A double-channel risk mapper is established inside the DQN optimization layer, a first channel receives the marked route path, and a second channel receives the special request not subjected to rule definition, and path state snapshots are respectively generated;

[0141] Step S10532: Real-time terminal heartbeat sequences and cross-department time delay fingerprints are injected into each path state snapshot to form dynamic risk fingerprints;

[0142] Step S10533: An adaptive risk measurer is called according to the risk fingerprints to output the blocking probability vector of each candidate path;

[0143] In the embodiment, the adaptive risk measurer can receive the risk fingerprints generated by step S10532, and the risk fingerprints are a multi-dimensional vector containing real-time terminal heartbeat features and cross-department time delay features. The measurer compares the vector with each template in the pre-constructed historical risk template library. The comparison is performed by vector similarity calculation, specifically, the Euclidean distance between the current fingerprint and each template is calculated. The template with higher similarity is considered to be closer to the historical blocking scenario. The measurer accordingly gives each candidate path a blocking probability value: the probability corresponding to the template with the highest similarity is given the maximum weight, and the remaining templates contribute smaller weights in descending order of similarity. Finally, the weighted probability value constitutes the blocking probability vector of the path, which is used for subsequent risk adjustment.

[0144] Step S10534: The blocking probability vector is coupled with the path business weight to generate a risk-adjusted route value indication and cached to the decision log.

[0145] As can be seen, step S1053 first sends the paths from the rule definition and the special request into the dual-channel risk mapper respectively, and each path is immediately snapped. Then, the system pulls the heartbeat sequence of the approval terminal and the cross-department collaboration time delay fingerprint in real time, superimposes them on the corresponding snapshot, and forms a unique and non-replayable dynamic risk fingerprint. After the adaptive risk measurer reads the fingerprint, it quickly gives a blocking probability vector, couples the vector with the business weight, and generates a routing value indication that can be directly used for sorting. The entire process is completed in milliseconds, and all intermediate results are written into the decision log for subsequent auditing and visualization.

[0146] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0147] The following is an embodiment of a routing algorithm-based workflow processing system provided by the embodiments of the present disclosure. The system and the routing algorithm-based workflow processing method described above belong to the same inventive concept. Details not described in the routing algorithm-based workflow processing system embodiment can be referred to the routing algorithm-based workflow processing method embodiment described above.

[0148] As shown in Figure 5 , the system comprises:

[0149] The feature extraction module 201 is configured to obtain an approval request transmitted through an API gateway, construct a semantic graph of the request content to extract entity relationships, obtain real-time working status of an approval terminal, calculate dynamic weights through collaborative analysis in combination with historical request information, and obtain a feature vector.

[0150] The request analysis module 202 is configured to analyze the approval request to obtain basic data.

[0151] The semantic fusion module 203 is configured to construct a semantic graph of the request text content based on the basic data to extract entity relationships, obtain real-time working status of an approval terminal according to associated approval terminal information, calculate weights through collaborative analysis in combination with historical approval records, and obtain a feature vector.

[0152] The priority judgment module 204 is configured to use a rule engine priority layer to process the basic data obtained through analysis and the feature vector extracted, and obtain an approval routing judgment.

[0153] The routing decision module 205 is configured to use a DQN optimization layer to calculate a routing value of the content processed by the rule engine priority layer or the basic data and the feature vector obtained through analysis after risk adjustment in a hybrid decision routing, and determine an optimized routing path.

[0154] The feedback execution module 206 is configured to execute the approval process and push the task according to the path determined by the hybrid decision routing, receive operation feedback, and generate a workflow visualization report. Figure 6 As shown in the figure, the SHAP analysis is used to analyze the routing decision process, and a decision explanation visualization report is generated to explain the basis and key influencing factors of the routing decision. The SHAP analysis result and user score feedback are fed back to the user terminal. In this way, the SHAP analysis enhances the explainability of the decision and improves the user's trust in the system; the user score feedback is used for model optimization to improve the accuracy and adaptability of the routing decision, and continuous optimization is achieved.

[0155] As shown in the figure, the SHAP analysis is used to analyze the routing decision process, and a decision explanation visualization report is generated to explain the basis and key influencing factors of the routing decision. The SHAP analysis result and user score feedback are fed back to the user terminal. In this way, the SHAP analysis enhances the explainability of the decision and improves the user's trust in the system; the user score feedback is used for model optimization to improve the accuracy and adaptability of the routing decision, and continuous optimization is achieved. Figure 7 As shown in the figure, the SHAP analysis is used to analyze the routing decision process, and a decision explanation visualization report is generated to explain the basis and key influencing factors of the routing decision. The SHAP analysis result and user score feedback are fed back to the user terminal. In this way, the SHAP analysis enhances the explainability of the decision and improves the user's trust in the system; the user score feedback is used for model optimization to improve the accuracy and adaptability of the routing decision, and continuous optimization is achieved.

[0156] In embodiments of the application, the electronic device includes, but is not limited to, a laptop computer, a desktop computer, a workstation, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices such as personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections, and relationships, and their functions, are meant to be examples only, and are not intended to limit implementations of the applications described and / or claimed in this document.

[0157] In embodiments of the application, the processor 101 can be implemented by using at least one of an application-specific integrated circuit, a programmable logic device, a field programmable gate array, a processor, a controller, a microcontroller, a microprocessor, an electronic unit designed to perform the functions described herein, and in some cases, such implementation can be implemented in a controller. For software implementation, the implementation of such as processes or functions can be implemented with separate software modules allowing at least one function or operation to be performed, and the software code can be implemented by a software application (or program) written in any appropriate programming language and stored in a memory and executed by a controller.

[0158] The display module 103 is configured to display information input by a user or information provided to a user. The display module 103 can include a display panel, which can be configured in the form of a liquid crystal display, an organic light-emitting diode, etc.

[0159] Memory 102 can be used to store software programs as well as various data. Memory 102 can include high-speed random access memory, and can also include nonvolatile memory such as at least one magnetic disk storage device, flash memory device, or other nonvolatile solid-state storage device.

[0160] The application also provides a storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the workflow processing method based on the routing algorithm.

[0161] The storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium may, for example, be but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination thereof. More specific examples (non-exhaustive list) of the readable storage medium include an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0162] The above description of disclosed embodiments enables one of ordinary skill in the art to make or use the application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for workflow processing based on a routing algorithm, characterized in that, The method comprises: S101: Obtain an approval request transmitted through an API gateway, build a semantic graph of the request content to extract entity relationships, obtain real-time working status of an approval terminal, combine historical request information, calculate dynamic weights through collaborative analysis, and obtain a feature vector; S102: Analyze the approval request to obtain basic data; S103: Based on the basic data, build a semantic graph of the request text content to extract entity relationships, obtain real-time working status of an approval terminal according to associated approval terminal information, combine historical approval records, calculate weights through collaborative analysis, and obtain a feature vector; S104: Process the basic data and the extracted feature vector obtained through analysis by using a rule engine priority layer to obtain an approval routing judgment; Step S1041: Based on the basic data analyzed in step S102 and the feature vector extracted in step S103, build an index of an approval data rule library, store the index according to business types and approval levels, and mark main rules and auxiliary rules; Step S1042: Perform rule matching preprocessing on the basic data and the feature vector, and extract key parameters related to rules in the approval data rule library; Step S1043: Preliminarily match the features through the main rules, and select candidate routes that meet the main conditions; the candidate routes are sorted according to a matching degree = a number of rule clauses met / a total number of rule clauses; the auxiliary rules retain candidate routes that meet the conditions; Step S1044: Perform contradiction detection on the matching results, and determine a final approval routing according to rule priorities if there are multiple routes that meet the conditions; Step S1044 further comprises the following steps: Step S10441: Extract contradictory features from the candidate route set after rule matching, identify contradiction types, and mark contradiction dimensions of each candidate route; Step S10442: Based on historical decision information and a current business scenario, reorder weights of the contradictory rules; The historical priority score is assigned according to a rule preset priority; a hit pass rate in a preset time period = a number of times of passing after hitting / a total number of hits; A scenario adaptation coefficient is adjusted according to a current business scenario; Step S10443: For candidate routes with priority contradictions, calculate a comprehensive score by using a weighted scoring method, and select a route with the highest comprehensive score as a final approval path; Step S10444: For unmatched requests, extract abnormal features in the basic data and key information in the feature vector, mark the abnormal features as an abnormal feature request marker code, attach contradictory reason explanations and associated parameters, and output an approval routing judgment result containing the marker information; S105: In the hybrid decision routing, perform routing value calculation after risk adjustment on the content processed by the rule engine priority layer or directly on the basic data and the feature vector obtained through analysis by using a DQN optimization layer in combination with a congestion risk prediction, and determine an optimized routing path; Step S1051: Receive the approval routing judgment result output in step S104, obtain the basic data analyzed in step S102 and the feature vector extracted in step S103, integrate the three types of information into a unified decision input data set, and mark key influence information in the data; Step S1052: Based on the input decision input data set, the blocking risk assessment dimension is constructed to form a risk assessment matrix; Step S1053: In the DQN optimization layer, the risk matching of the route processed by the rule engine priority layer is calculated, and the blocking probability of the route under the current risk assessment matrix is calculated; for the requests not processed by the rule engine, the blocking probability of the route is calculated based on the basic data and the feature vector; Step S1054: Adjust each candidate route according to the blocking probability, and preferentially select the route path with low blocking risk and meeting the business target; if there are multiple equivalent paths, the final optimized route path is determined according to the historical execution efficiency sorting, and the path details are output; S106: According to the path determined by the hybrid decision routing, the approval process is executed and the task is pushed, the operation feedback is received, and the workflow visualization report is generated.

2. The workflow processing method based on routing algorithm according to claim 1, characterized in that, Step S102 specifically comprises: Data analysis is performed on the approval request to extract the approval amount field, application reason description, request initiation time and source system identifier to form an original data set; Based on the pre-defined approval data rule library, the parsed structured data is contextually associated and verified; Intention recognition is performed on the original data set content to extract key business elements, and an urgency label and a list of involved departments are generated in combination with the department affiliation information of the associated approval terminal; When a missing key field is detected, a correction process is performed to fill in or associate historical data to complete the missing information by using a pre-set default value.

3. The workflow processing method based on routing algorithm according to claim 1, characterized in that, Step S103 specifically comprises: Based on the parsed basic data, the explicit entity relationship and implicit business association in the text content are extracted in combination with the approval data rule library and the semantic graph; According to the involved department information in the basic data, the real-time working state of the associated approval terminal is collected, and the department cooperation mode data is associated; From the historical approval records, historical cases matching the current request are selected, the average processing time efficiency, the abnormal interruption ratio and the user feedback score of each case are counted, and a historical efficiency feature set is generated; The entity relationship of the semantic graph, the real-time working state index set and the historical efficiency feature set are input into the collaborative analysis module to adjust the weight distribution of the features, and a feature vector integrating content relevance, resource availability and historical reliability is generated.

4. The workflow processing method based on routing algorithm according to claim 1, characterized in that, Step S1053 specifically comprises: A double-channel risk mapper is established in the DQN optimization layer, the first channel receives the marked route path, and the second channel receives the special request not processed by the rule definition, and path state snapshots are generated respectively; Real-time terminal heartbeat sequence and cross-department time delay fingerprint are injected into each path state snapshot to form a dynamic risk fingerprint; According to the risk fingerprint, an adaptive risk measurer is called to output the blocking probability vector of each candidate path; The blocking probability vector is coupled with the path service weight to generate a risk-adjusted route value indication and cached to a decision log.

5. A routing algorithm based workflow processing system, characterized in that, The system is used to implement the workflow processing method based on the routing algorithm as claimed in any one of claims 1 to 4. The system comprises: The feature extraction module is configured to acquire the approval request transmitted through the API gateway, construct a semantic graph of the request content to extract entity relationships, acquire real-time working status of the approval terminal, combine historical request information, calculate dynamic weights through collaborative analysis, and obtain a feature vector. The request analysis module is configured to analyze the approval request to obtain basic data. The semantic fusion module is configured to construct a semantic graph of the request text content based on the basic data to extract entity relationships, acquire real-time working status of the approval terminal according to associated approval terminal information, combine historical approval records, calculate weights through collaborative analysis, and obtain a feature vector. The priority judgment module is configured to analyze the obtained basic data and feature vector, perform priority layer processing by using a rule engine, and determine an approval routing result. The routing decision module is configured to perform risk-adjusted route value calculation on the content processed by the rule engine or directly on the obtained basic data and feature vector, determine an optimized routing path, and perform the optimized routing path. The execution feedback module is configured to execute an approval process according to the determined routing path, push a task, receive operation feedback, and generate a workflow visual report.

6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps of the workflow processing method based on the routing algorithm as claimed in any one of claims 1 to 4.

7. A storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the steps of the workflow processing method based on the routing algorithm as claimed in any one of claims 1 to 4.

Images