A multi-channel aggregation service routing strategy configuration method and system

Abstract

This invention relates to a method and system for configuring routing strategies for multi-channel aggregation services. The core solution includes: constructing a routing rule library containing general routing rules and channel-specific routing rules; classifying routing strategies into general strategies and channel strategies; allowing general strategies to use only general routing rules from the rule library and not any channel-specific routing rules; and allowing channel strategies to use both general routing rules from the rule library and the channel-specific routing rules corresponding to the channel. Multi-dimensional feature analysis is used to analyze target channel attributes and strategy weights, fusing routing rule conditions, actions, scope of application, and historical feedback to generate semantic fingerprints. A global rule conflict perception graph is constructed using a graph neural network to identify and visualize conflict relationships such as logical coverage, mutual exclusion, and coupling. Candidate rules are automatically reordered, and strategy anchor point boundary locking is supported, enabling adaptive optimization and change tracking.

Description

Technical Field

[0001] This invention relates to the field of multi-channel aggregation service routing policy configuration technology, and in particular to a multi-channel aggregation service routing policy configuration method and system. Background Technology

[0002] Currently, with the increasing complexity and diversity of services offered by multi-channel aggregation service platforms, the intelligent configuration of routing strategies has become a hot technological topic in the industry. Existing technologies have the following shortcomings:

[0003] (1) In the prior art, routing rules are not designed as independent objects. Instead, rules are configured directly on the policy, which results in rules not being shared and a large number of duplicate configurations, increasing configuration and maintenance costs.

[0004] (2) In the prior art, routing rules and policies are not classified and processed, the configuration logic is chaotic and unclear, the rule sharing is poor, which makes it difficult to view, maintain and modify routing policies.

[0005] (3) When two rules recommended at the same time have complex relationships such as coverage, negation, or dependency in terms of conditions or actions, traditional mechanisms often fail to detect and handle them in a timely manner, leading to problems such as high-priority rules being covered by low-priority rules, business process breaks, or multi-point mutual exclusion deadlocks in actual policy operation. This not only affects the accuracy of routing and the consistency of service SLA, but also increases the workload of manual investigation and auditing, and weakens the intelligence and self-driving capabilities of the policy system.

[0006] (4) Although some existing systems have conflict detection functions, most of them are at the post-analysis level and can only discover conflicts that have occurred through running logs, failure alarms and other means. They lack a pre-conflict identification mechanism in the rule recommendation and priority generation stages. Summary of the Invention

[0007] This application provides a method for configuring routing strategies for multi-channel aggregation services, aiming to solve one of the problems or issues of the existing technology mentioned in the background section.

[0008] This application provides a method for configuring routing policies for multi-channel aggregation services, including: constructing a routing rule library containing general routing rules and channel-specific routing rules; dividing routing policies into two categories: general policies and channel policies; general policies can only use general routing rules in the rule library and cannot use any channel-specific routing rules; channel policies can use general routing rules in the rule library and channel-specific routing rules corresponding to the channel.

[0009] The core elements of a routing rule include rule conditions and routing target elements. Furthermore, routing rules are categorized, and a routing rule library containing general routing rules and channel-specific routing rules is constructed. The specific implementation process is as follows:

[0010] Rule type: When configuring a rule, first set the rule type: general routing rule or channel-specific routing rule for a certain channel.

[0011] The specific criteria for classifying general routing rules and channel-specific routing rules are as follows:

[0012] Universal routing rules: These rules are independent of specific channels and can be applied to customer requests across all channels. For example, the rule "Route based on system-defined tags of visiting friends" does not depend on any channel characteristics and can be used by strategies across all channels, enabling cross-channel reuse.

[0013] Channel-specific routing rules: These rules are strongly correlated with specific channels and apply only to customer requests within that channel. For example, "Route based on WeChat Work account affiliation" applies only to WeChat Work channels; "Route based on Douyin service account identifier" applies only to Douyin channels, preventing misuse of the rules.

[0014] Rule conditions: Set according to the actual characteristics of customer requests, such as customer tags, channel type, access node affiliation, customer level, etc., to ensure that the rules can accurately match customer requests.

[0015] Routing target: After a successful rule match, the specific object to which the customer request needs to be routed can be selected based on actual needs, such as an automated response robot, agent skill group, or specific agent.

[0016] All configured routing rules are stored in the rule base as independent objects that can be selected by subsequent routing strategies and can be edited individually. When a rule needs to be modified, only the independent rule object in the rule base needs to be modified, and all routing strategies that reference the rule will take effect synchronously, without having to modify each routing strategy one by one.

[0017] Configuration Phase - Routing Policy Configuration and Rule Association References:

[0018] After the routing rules are configured, the routing policy management module is used to configure the routing policy. The routing policy acts as a combination object for rules, allowing for flexible combination and priority setting of rules. The specific implementation process is as follows:

[0019] (1) Strategy Classification: Routing strategies are divided into two categories: general strategies and channel strategies. The rules for these two types of strategies are selected in different scenarios, as detailed below:

[0020] ① General Strategy: The general strategy can only use general routing rules in the rule base and cannot use any channel-specific routing rules. It is mainly applicable to general routing scenarios for all channels and can be applied to routing logic for different channels.

[0021] ② Channel strategy: such as Figure 4 As shown, the channel strategy can use general routing rules from the rule base and channel-specific routing rules corresponding to this channel. It cannot use dedicated routing rules for other channels. It is mainly suitable for personalized routing scenarios of channels, taking into account both the reuse of general rules and the adaptation to channel-specific needs.

[0022] After configuring the routing policy, configure the departments to apply the policy to achieve batch application of the policy. The specific implementation process is as follows:

[0023] (1) Association between channel friend access nodes and departments: First, configure the channel friend access nodes of each channel (such as the WeChat Enterprise Account, the Douyin Service Account of Douyin Channel, the access account of 5G Messaging, etc.) to the corresponding departments, that is, clarify which department manages each channel friend access node and establish the association relationship of "department-channel friend access node".

[0024] (2) Binding of strategy and effective department: Select one or more departments for which the strategy should be effective in the routing strategy. After binding, all channel friends access nodes under the management of these departments will automatically take effect with the routing strategy, without the need to configure each access node individually.

[0025] (3) Implementation of global policy: If a routing policy needs to be effective for all channel friends access nodes of the entire enterprise, simply configure the effective department of the policy as the root department (the root department represents the entire enterprise) to realize the global application of the policy and simplify the configuration process of global policy.

[0026] Runtime Phase - Routing Policy Execution Logic:

[0027] When customers from multiple channels request access to the system, the routing execution module will perform routing operations according to the following logic to ensure that customer requests are delivered accurately:

[0028] (1) Identify the channel type of the customer request and the corresponding channel friend access node to clarify the source of the customer request;

[0029] (2) Based on the identified channel friend access nodes, query the department to which they belong;

[0030] (3) Match all routing policies bound to the department and locate the corresponding routing policy according to the channel type requested by the customer;

[0031] (4) Execute the routing rules in sequence according to the priority of the rules configured in the strategy, and determine whether the client request matches the current rule conditions;

[0032] (5) If a rule is matched successfully, the customer request will be routed to the automatic response robot, agent skill group or specific agent set by the rule, and the routing operation will be completed. If all rules fail to match, a general fallback rule can be configured and this fallback rule can be selected for each strategy (configured in the lowest priority position). When all the previous rules fail to match, the preset default routing logic (such as routing to the general agent skill group) will be executed to ensure that the customer request can be responded to.

[0033] Channel policies can use general routing rules from the rule base and channel-specific routing rules corresponding to this channel. The specific configuration of general routing rules and channel-specific routing rules for a channel policy should be based on the actual situation, following the suggested configuration:

[0034] The channel strategy configuration method specifically includes:

[0035] S1: Obtain the target channel attribute characteristics and strategy target weight of the channel strategy to be configured, and extract the semantic elements of each rule in the routing rule base;

[0036] S2: Based on the semantic elements, map each routing rule to a fixed-dimensional rule semantic fingerprint to generate a rule semantic fingerprint set;

[0037] S3: Based on the logical overlap, conditional mutual exclusion, and action conflict relationships between any two rule semantic fingerprints in the rule semantic fingerprint set, construct a global rule conflict perception map containing overlay edges, mutual exclusion edges, and coupling edges;

[0038] S4: Based on the target channel attribute features, retrieve the appropriate rule subset in the global rule conflict perception map to generate an initial candidate rule set and its corresponding candidate semantic fingerprint sub-map;

[0039] S5: Perform conflict propagation processing on the candidate semantic fingerprint subgraph, identify covering edges, mutually exclusive edges and coupling edges in the candidate semantic fingerprint subgraph, and generate a set of potential conflict paths marked with conflict type, impact breadth and arbitration basis;

[0040] S6: Based on the conflict type in the potential conflict path set and the strategy target weight, rearrange the rule priority sequence in the initial candidate rule set to generate a preliminary priority sequence;

[0041] S7: Receive the highest non-adjustable rule anchor point and the lowest basic rule anchor point set by the user, combine the coupling edge constraints in the global rule conflict perception graph, lock the rule boundaries of the preliminary priority sequence, and adjust the positions of the unlocked floating rules according to the monotonically increasing overall priority sequence without logical breaks, to generate the final rule priority sequence.

[0042] S8: Manual review, where strategy personnel or reviewers verify the business rationality of the final rule priority sequence, focusing on whether conflict resolution meets expectations and whether anchor point settings are accurate. Once confirmed to be correct, the rule is issued for execution or rolled back for adjustment.

[0043] This application provides a multi-channel aggregation service routing strategy configuration method and system, which has the following beneficial effects:

[0044] (1) Strong rule reusability, greatly improving configuration and maintenance efficiency: By treating routing rules as independent objects, they can be repeatedly selected by multiple routing strategies, effectively avoiding the repeated configuration of rules; when a rule needs to be modified, only the independent rule object in the rule library needs to be modified, and all strategies that reference the rule will take effect synchronously without needing to modify them one by one. This not only reduces configuration and maintenance costs, but also reduces routing anomalies caused by missing modifications, which can well adapt to the core needs of "cost reduction and efficiency improvement" of multi-channel customer service systems.

[0045] (2) Clear classification of rules and policies enhances the convenience of configuration and maintenance: By dividing rules and policies into two categories, general and channel-specific, the applicable scope of each type of rule and policy is clearly defined, the configuration logic is clearer, and the misuse of rules is avoided. At the same time, it is also convenient for staff to view, maintain and modify routing policies, reducing the difficulty of operation. When the routing requirements are relatively simple, it may be possible to reuse all channels of all departments by configuring only one general routing policy. The policy activation configuration is simple and efficient, and adaptable to enterprise-scale management: The reverse binding method of "configuring the effective department on the policy" is adopted. It is not necessary to configure the policy for each department or each channel friend access node one by one (because there may be many departments). It is only necessary to select the effective department on the policy to realize the batch activation of the policy for all access nodes under that department. If the policy needs to be adjusted later, it is only necessary to modify the policy itself or the effective department bound to it. It is not necessary to modify the configuration of multiple departments or access nodes simultaneously, which greatly improves the configuration efficiency, especially suitable for medium and large enterprises with a large number of departments and access nodes. At the same time, the global application of the policy can be realized through the root department configuration, which further simplifies the configuration process of the global policy. It also supports flexible adjustment of the priority of routing rules within the routing policy.

[0046] (3) Highly scalable and adaptable to multi-channel expansion needs: When an enterprise adds a new channel (such as a video account, a Xiaohongshu service account, etc.), it only needs to add dedicated rules and strategies for that channel. There is no need to modify the existing rule and strategy system to quickly realize the routing configuration of the new channel. There are also general rules that can be selected for the routing strategy of the new channel. Clear structure, easy to understand and maintain: The entire routing strategy configuration is divided into rule management, strategy management, and effective object configuration. The responsibilities of each module are clear and the logic is clear. Both configuration personnel and maintenance personnel can quickly understand the system architecture, which is convenient for subsequent system maintenance and iterative upgrades.

[0047] (4) By moving rule conflict identification from post-verification to the recommendation generation stage and introducing a dynamic priority adjustment mechanism based on semantic fingerprints and graph structure reasoning, the accuracy of multi-channel aggregation service routing strategy configuration is significantly improved. Traditional methods usually rely on static priority sorting or manual experience setting, which makes it difficult to detect potential logical conflicts in a timely manner when faced with complex and frequently changing rule sets, leading to distorted routing decisions or even service anomalies. This solution extracts semantic elements such as condition expression structure features, action behavior labels, applicable scope metadata, and historical execution feedback during the rule registration stage to construct a rule semantic fingerprint that does not depend on specific parameter values ​​but can represent logical intent, realizing an abstract expression of the essential function of the rule; on this basis, a global rule conflict perception graph is established, which uses three types of relationships, namely "covering edge", "mutually exclusive edge", and "coupling edge", to accurately characterize the logical dependencies and constraints between rules, so that when adding or modifying rules, their impact on the overall strategy space can be evaluated in real time, effectively avoiding global chaos caused by local adjustments. This mechanism fundamentally overcomes the passive response mode of "configure first and then report errors" in traditional solutions, making conflict identification forward-looking and structural, and greatly reducing the risk of strategy deployment. Attached Figure Description

[0048] Figure 1 This is the main flowchart of a multi-channel aggregation service routing strategy configuration method;

[0049] Figure 2 This is a sub-flowchart of a multi-channel aggregation service routing strategy configuration method;

[0050] Figure 3 This is another sub-flowchart of a multi-channel aggregation service routing strategy configuration method;

[0051] Figure 4 This is a schematic diagram illustrating the selection of channel strategy rules in one embodiment;

[0052] Figure 5 This is an application environment diagram of a multi-channel aggregation service routing strategy configuration method in one embodiment;

[0053] Figure 6 This is a schematic diagram of the structure of a computer device provided in one embodiment of this application. Detailed Implementation

[0054] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0055] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0056] This application provides a method for configuring routing policies for multi-channel aggregation services. The method includes: constructing a routing rule library containing general routing rules and channel-specific routing rules; dividing routing policies into two categories: general policies and channel policies; general policies can only use general routing rules in the rule library and cannot use any channel-specific routing rules; channel policies can use general routing rules in the rule library and channel-specific routing rules corresponding to the channel.

[0057] Channel policies can use general routing rules from the rule base and channel-specific routing rules corresponding to this channel. The specific configuration of general routing rules and channel-specific routing rules for a channel policy should be based on the actual situation, following the suggested configuration:

[0058] like Figure 1 As shown, the channel strategy configuration method specifically includes:

[0059] S1: Obtain the target channel attribute features and strategy target weights of the channel strategy to be configured, and extract the semantic elements of each rule in the routing rule base. The semantic elements include: conditional expression structure features, action behavior tags, applicable scope metadata, and historical execution feedback summaries.

[0060] S2: Based on the semantic elements, each routing rule is mapped to a fixed-dimensional rule semantic fingerprint to generate a set of rule semantic fingerprints that represent the comparability of rule logical intent.

[0061] S3: Based on the logical overlap, conditional mutual exclusion, and action conflict relationships between any two rule semantic fingerprints in the rule semantic fingerprint set, construct a global rule conflict perception graph containing overlay edges, mutual exclusion edges, and coupling edges.

[0062] S4: Based on the target channel attribute features, retrieve a suitable subset of rules from the global rule conflict perception graph to generate an initial candidate rule set and its corresponding candidate semantic fingerprint subgraph. The initial candidate rule set includes general rules and channel-specific rules.

[0063] S5: Perform conflict propagation processing on the candidate semantic fingerprint subgraph, identify covering edges, mutually exclusive edges and coupling edges in the candidate semantic fingerprint subgraph, and generate a set of potential conflict paths marked with conflict type, impact breadth and arbitration basis.

[0064] S6: Based on the conflict type in the potential conflict path set and the strategy target weight, rearrange the rule priority sequence in the initial candidate rule set to generate a preliminary priority sequence with context-sensitive calibration.

[0065] S7: Receive the highest non-adjustable rule anchor point and the lowest basic rule anchor point set by the user, combine the coupling edge constraints in the global rule conflict perception graph, lock the rule boundaries of the preliminary priority sequence, and adjust the positions of the unlocked floating rules according to the monotonically increasing overall priority sequence without logical breaks, to generate the final rule priority sequence.

[0066] S8: Manual review, where strategy personnel or reviewers verify the business rationality of the final rule priority sequence, focusing on whether conflict resolution meets expectations and whether anchor point settings are accurate. Once confirmed to be correct, the rule is issued for execution or rolled back for adjustment.

[0067] Step S1: Obtain the target channel attribute features and strategy target weights of the channel strategy to be configured, and extract the semantic elements of each rule in the routing rule base. The semantic elements include: conditional expression structure features, action behavior tags, applicable scope metadata, and historical execution feedback summaries. Specifically, this includes:

[0068] S1.1: Perform multi-dimensional feature scanning on the target channel access node to be configured with channel strategy to obtain a set of target channel attribute features including channel type identifier, service level agreement threshold, historical fault frequency statistics and the capability profile of the connected system.

[0069] There are two input methods for obtaining the target channel attribute features.

[0070] The first method is a system-automated multi-dimensional feature scanning: taking the unique identifier of the channel node as input, extracting the channel type identifier (API, message queue, gateway, etc.) from the service aggregation platform, obtaining thresholds such as response latency, availability, and recovery time from the SLA management module, statistically analyzing the frequency of historical faults and repair time from the operation and maintenance logs, and analyzing profiles such as concurrency capability, protocol type, and stability score from the interaction logs, and outputting a structured feature set after standardization processing.

[0071] The second type is natural language input: suitable for new channels or manual correction scenarios. Natural language provides information, which is then semantically parsed and mapped to a large language model, and also aligned to four fields: channel type identifier, SLA threshold, historical fault statistics, and capability profile.

[0072] The system performs multi-dimensional feature scanning on the target channel access nodes for configuring channel policies. The input data consists of the unique identifier of the channel access node and its associated basic metadata. The data acquisition system first retrieves the access node's registration information in the service aggregation platform, obtaining the channel type identifier through an interface. This includes, but is not limited to, API, message queue, and gateway channels, and records them as structured fields. Subsequently, the system retrieves the corresponding SLA threshold parameters for the node from the Service Level Agreement (SLA) management module. These parameters include key performance indicators such as maximum allowable response latency, minimum availability requirements, and maximum fault recovery time, mapping these parameters to standardized numerical features. For historical fault frequency statistics, the system automatically connects to the operation and maintenance monitoring log database, filtering the target node's occurrence frequency, average repair time, and main fault type distribution within a set time window (e.g., the last 30 days), and outputs the data as a statistical vector. Furthermore, in the process of generating the capability profile of the docking system, capability parameters such as maximum concurrent processing capacity, supported protocol types, interface stability scores, and historical peak loads are extracted by analyzing the data interaction logs and interface call performance reports between the target node and upstream and downstream systems. These parameters are then transformed into comparable profile vectors using feature normalization. Through the above multi-dimensional feature scanning and data standardization processing, the raw information of a single channel access node is transformed into a set of target channel attribute features, including channel type identifier, service level protocol threshold, historical fault frequency statistics, and the capability profile of the docking system. This achieves structured, multi-level, and scalable data representation, providing high-quality input for subsequent strategy target weight calculation and rule matching models.

[0073] S1.2: Based on the business scenario constraints in the target channel attribute feature set, perform weighted calculation processing on the preset strategy optimization dimensions to generate a strategy target weight vector that represents the relative importance of accuracy, availability and timeliness.

[0074] Obtain the target channel attribute feature set output in step S1.1. Inputs include channel type identifier, service level agreement threshold, historical fault frequency statistics, and the capability profile of the connected system. Perform business scenario constraint analysis on the target channel attribute feature set. Establish basic weight allocation rules for accuracy, availability, and timeliness as three optimization dimensions for different channel types and service level agreement thresholds. Using a multi-dimensional weighted calculation method, first, a preset business scenario weight template is selected based on the channel type identifier, and initial weight coefficients are assigned to the three dimensions of accuracy, availability, and timeliness. The initial weight coefficients are dynamically adjusted according to the service level agreement threshold, setting a weight increase for accuracy corresponding to high service level agreements and a weight increase for availability or timeliness corresponding to medium and low service level agreements. The comprehensive target weight for each optimization dimension is calculated using the following weighting formula:

[0075]

[0076] Among them, w i Let a be the target weight for the i-th optimization dimension (accuracy, usability, or timeliness). i Let x be the initial scene weight coefficient for the i-th dimension. i This is the attribute feature correction factor for the i-th dimension. The availability target weights are corrected using historical failure frequency statistics, appropriately increasing the availability coefficient for channels with high failure frequency and maintaining the original distribution for channels with low failure frequency. Combined with the capability profile of the interface system, the timeliness target weights are adjusted a second time based on system response capability and concurrent processing capability, increasing the weight of timeliness targets for high-performance systems and decreasing the weight for low-performance systems. All corrected optimization dimension target weight vectors are normalized to ensure the sum of the three indicators equals 1, serving as the basic parameter input for subsequent dynamic adjustment of strategy priorities. Through the above multi-factor weighting and normalization process, the channel attribute features extracted in the previous step are transformed into strategy target weight vectors representing accuracy, availability, and timeliness, achieving a refined quantitative expression of diverse business demands during strategy configuration.

[0077] For example, in a typical application scenario, the channel to be configured is a high SLA dedicated line channel for financial payments, with a service level agreement threshold of 99.99%, a historical failure frequency of 2 times per year, and a system capability profile showing a response latency of less than 50 milliseconds and support for tens of thousands of concurrent requests per second. In this scenario, based on the financial payment business template, the initial weight coefficients for accuracy are set to 0.6, availability to 0.3, and timeliness to 0.1. Combining the high SLA threshold, the accuracy correction factor is set to 1.2, the availability correction factor to 0.9, and the timeliness correction factor to 1.0. Substituting these values ​​into the formula, the weighted numerators of the three optimization dimensions are calculated and normalized to obtain the final strategy objective weight vector: accuracy accounts for approximately 0.67, availability approximately 0.22, and timeliness approximately 0.11. The output results show that this step can significantly improve the discernment and decision-making accuracy of various business demand indicators during strategy configuration based on actual business scenarios and technical conditions, providing a scientific and reasonable data foundation for subsequent dynamic adjustment of rule priorities. By changing the initial template and correction factors, flexible mapping can be achieved for different business needs across different types of channels (such as social message push channels and IoT data collection channels), effectively supporting intelligent recommendation and priority control of routing strategies under multiple scenarios and objectives.

[0078] S1.3: Perform syntax tree traversal and structural analysis on the conditional expression of each routing rule stored in the routing rule base to extract the structural feature sequence of the conditional expression, which includes field reference paths, operator type identifiers, and threshold interval forms.

[0079] The routing rule base is a centralized database or knowledge base that stores all defined routing policies. The routing rules are primarily pre-configured and maintained by system administrators or business experts based on historical business needs, traffic scheduling experience, and system operation and maintenance specifications. Each rule clearly specifies which target channel or processing node traffic should be routed to under specific conditions, such as request source, message content, and system load.

[0080] S1.4: Based on the rule logic body corresponding to the structural feature sequence of the condition expression, the rule execution actions and metadata constraints are extracted and tagged to generate a complete set of rule semantic elements containing action behavior tags, applicable scope metadata and historical execution feedback summaries.

[0081] Using the conditional expression structure feature sequence of each routing rule extracted in step S1.3 as input, standardized extraction of action behavior tags is performed for each rule's logical body. Key action elements in the rule definition, such as action type, target channel, and processing method, are hierarchically parsed and mapped into a structured set of action behavior tags. Metadata constraint fields associated with the same rule body are normalized, including but not limited to the applicable channel list, department whitelist, time window limit, and initial priority value, transforming various constraint parameters into a unified format of applicable scope metadata vector. Based on historical policy execution logs, statistical indicators such as the number of hits, average response latency, and failure attribution for each rule in the recent period are collected. A historical execution feedback summary reflecting the actual operating status of the rule is calculated through sliding window aggregation. A multi-source element fusion mechanism is employed to align the three semantic elements—action behavior tags, applicable scope metadata, and historical execution feedback summary—with the conditional expression structure feature sequence, forming a complete and non-redundant set of rule semantic elements. This processing method expands the structured conditional expression features output from the previous step into a multi-dimensional semantic description covering rule actions, application scope, and dynamic feedback, achieving a comprehensive representation of the logical intent of routing rules at both static and dynamic levels.

[0082] Step S2: Based on the semantic elements, map each routing rule to a fixed-dimensional rule semantic fingerprint to generate a set of rule semantic fingerprints representing the comparability of rule logical intent. Specifically, this includes:

[0083] S2.1: Perform field alignment and format normalization on four types of original semantic elements: conditional expression structural features, action behavior labels, applicable scope metadata, and historical execution feedback summary, in order to generate a set of standard rule elements with a unified data schema and eliminate dimensional differences caused by multi-source heterogeneous data.

[0084] For the four categories of original semantic elements extracted from the routing rule base—conditional expression structural features, action behavior tags, applicable scope metadata, and historical execution feedback summaries—field-level data structure standardization processing is performed sequentially. For conditional expression structural features, based on preset field reference path specifications and operator type dictionaries, a unified naming mapping is applied to field names, operators, and threshold ranges in various conditional expressions. Synonymous fields (such as "channel type" and "channel category") are merged into standard field identifiers, and all interval thresholds are converted into standard closed or open interval forms. For action behavior tags, an action type enumeration table and a behavior semantic tag set are used to map action content describing the same action intent but with different expressions in different rules to a unified standard action code, achieving consistent normalization of action tags. For applicable scope metadata, a unified data schema is established based on multi-dimensional constraint attributes such as department / channel / time window, converting multi-source heterogeneous applicable scope descriptions (such as department ID lists, channel whitelists, time period strings, etc.) into structured metadata objects with a consistent format, and automatically filling in and trimming missing or redundant attributes. For historical execution feedback summaries, a fixed set of statistical indicators (such as the hit rate of the past 7 days, average response latency, and failure cause distribution) is set. Various feedback information from the original log data is extracted and populated into a standardized feedback summary template, ensuring that all rule feedback indicators have the same data dimensions and statistical scope. Through the above multi-level data field alignment and format normalization processing, all routing rule semantic elements generated from different development teams, different periods, and heterogeneous system environments are transformed into a standard set of rule elements with a unified data schema, field naming conventions, and numerical format constraints. This effectively eliminates the dimensional differences caused by multi-source heterogeneous data, laying the foundation for subsequent vectorized encoding and feature fusion. This standardization process transforms the unstructured or semi-structured rule elements extracted in the previous step into highly consistent data objects that can be directly used for subsequent embedding calculations, achieving standardization and comparability at the rule semantic modeling input level.

[0085] S2.2: Based on the standard rule element set, use a pre-trained domain-specific semantic embedding model to perform text encoding and numerical discretization operations respectively to generate conditional feature vectors, action one-hot encoding vectors, range constraint tensors and feedback statistical embedding vectors containing contextual semantic information.

[0086] The standard rule element set, after field alignment and format normalization, is input, including conditional expression structural features with a unified schema, action behavior labels, applicable scope metadata, and historical execution feedback summaries. A domain-specific pre-trained semantic embedding model is used to text-encode the conditional expression structural features, mapping information such as field reference paths, operator type identifiers, and threshold interval forms into conditional feature vectors with contextual relevance. One-hot encoding is used for action behavior labels, mapping each type of action (e.g., "route configuration to a node," "downgrade to a fallback interface") into a unique high-dimensional sparse vector, achieving distinguishable representation between action types. For applicable scope metadata, including the list of effective channels, department whitelists, and time window constraints, discretization and tensor encoding operations are performed based on enumeration and interval features respectively, generating scope constraint tensors that can be used for subsequent vector space calculations. For historical execution feedback summary data (e.g., statistical indicators such as the hit rate of the past 7 days, average response latency, and failure attribution main causes), numerical normalization and embedding mapping methods are used to form feedback statistical embedding vectors, providing semantic supplementation for the dynamic representation of the rules. Through the aforementioned continuous processing actions, the four types of standard rule elements are transformed into conditional feature vectors containing contextual semantic information, one-hot action encoding vectors, range constraint tensors, and feedback statistical embedding vectors, respectively. This lays a high-quality data foundation for subsequent multi-head attention mechanism fusion and dimensionality reduction projection. By utilizing pre-trained models and encoding strategies, efficient conversion of unstructured and heterogeneous rule elements into a unified vector space is achieved, significantly improving the comparability of rule logical intent in multi-dimensional space.

[0087] For example, in a real-world multi-channel service aggregation scenario, for a conditional expression containing the field reference "user.region", the operator "==", and the threshold range "East China", a 128-dimensional conditional feature vector is obtained through text encoding using a domain-specific BERT variant model. For the action behavior label "routing configuration to a certain node", an 8-dimensional one-hot encoding method is used, where the 3rd bit is activated and the remaining bits are 0. For the applicable scope metadata, if the effective channel list includes "APP" and "Web", the department whitelist is "Marketing Department", and the time window is "9:00-18:00", then a 12-dimensional range constraint tensor is concatenated after 2-dimensional binary encoding (APP=1, Web=1), department ID index mapping, and time period normalization. For the historical execution feedback summary, such as a hit rate of 0.85 in the past 7 days, an average response latency of 80ms, and a main cause distribution of type 2, a 4-dimensional feedback statistical embedding vector is generated through normalization and embedding table lookup. In this scenario, after the above chain-like derivation process, each rule obtains a complete four-class semantic vector representation, providing a solid data foundation for subsequent fusion computing and fingerprint generation, and effectively supporting the high-accuracy modeling of rule logical intent and conflict detection capabilities under cross-channel and cross-policy configuration.

[0088] S2.3: Perform multi-head attention mechanism fusion calculation on the conditional feature vector, action one-hot encoding vector, range constraint tensor and feedback statistical embedding vector to generate a weighted fusion feature matrix representing the implicit dependency relationship between each semantic element, and establish the contribution weight of different elements in the rule logic intent.

[0089] The conditional feature vectors, one-hot action encoding vectors, range constraint tensors, and feedback statistical embedding vectors obtained after mapping are used as input channels for the multi-head attention mechanism, and parallel feature representation mapping is performed according to their respective dimensional structures. For each semantic element, an independent query, key, and value weight matrix is ​​set, and the original vectors are mapped to a unified attention space through linear transformation to form multiple sets of candidate attention heads. The dot product attention calculation method is used to perform similarity scoring on each set of queries and corresponding keys, and the value vectors are weighted and converged to obtain the local attention distribution of explicit and implicit dependencies between semantic elements. The outputs of all attention heads are concatenated along the channel dimension, and the information interaction between different semantic elements is fused using a fully connected layer to achieve high-order feature coupling across element types. During the fusion process, a learnable normalization coefficient is introduced to dynamically adjust the contribution weight of each semantic element, enabling the model to adaptively adjust the focus according to different rule logic intentions. Finally, the fused weighted feature matrix is ​​used as the output of this step, providing a high-quality representation foundation for the subsequent dimensionality reduction projection network to generate fixed-dimensional rule semantic fingerprints. By using the multi-head attention mechanism fusion calculation method described above, the various semantic elements extracted in the previous step are transformed into a weighted fusion feature matrix that reflects their inherent correlation and contribution distribution, thus realizing explicit modeling of the dependency relationship between key elements in the rule logic intent.

[0090] For example, in a practical application scenario, for a certain routing rule, its conditional feature vector is 64-dimensional, the action one-hot encoding is 16-dimensional, the range constraint tensor is 8-dimensional, and the feedback statistical embedding is 4-dimensional. The number of heads in the multi-head attention mechanism is set to 8, and the Query, Key, and Value within each head are all set to 32 dimensions. In the specific implementation, Q is obtained by linearly transforming each type of input feature. i K i V i (i=1,2,3,4), and calculate the single-head attention output according to the following formula:

[0091]

[0092] Where, d k Let Q be the dimension of the Key vector. iLet K be the query matrix obtained after linear transformation of the i-th type of input features. i V is the key matrix obtained after linear transformation of the i-th type of input features. i The Value matrix is ​​obtained after linear transformation of the i-th type of input features. The outputs of the eight heads are concatenated and then combined through a fully connected layer and a normalization layer to generate a 256×1 weighted fusion feature matrix. In this scenario, due to differences in conditional complexity, action type, and historical performance, different rules exhibit significant non-linear dependencies between their conditional features and action labels. After processing by the multi-head attention mechanism, the important factors influencing policy determination can be effectively highlighted, achieving efficient modeling and differentiation of complex rule logical intent structures. The final weighted fusion feature matrix demonstrates good discriminative power and generalization ability in actual verification, providing a solid data foundation for subsequent fingerprint dimensionality reduction and collision detection.

[0093] S2.4: Based on the weighted fusion feature matrix, a linear transformation and nonlinear activation process are performed using a dimensionality reduction projection network to generate an initial rule semantic fingerprint of fixed dimension, compressing the high-dimensional sparse feature space into a low-dimensional dense semantic representation space.

[0094] The input data includes a weighted fusion feature matrix calculated using a multi-head attention mechanism. Each row of the matrix corresponds to a weighted combination of multi-dimensional semantic elements such as conditional features, action codes, applicable scope tensors, and feedback statistics of a routing rule in a high-dimensional space. For this weighted fusion feature matrix, a preset dimensionality reduction target dimension k is first selected, and parameters are set based on the size of the rule base and the complexity of subsequent similarity calculations. A linear transformation method is used to define a dimensionality reduction projection weight matrix W, with a shape of d×k, where d is the original feature space dimension and k is the target low-dimensional space dimension. The weighted fusion feature vector x is left-multiplied by W to complete the linear dimensionality reduction mapping. The specific calculation method is as follows:

[0095]

[0096] Where x is the original high-dimensional weighted fusion feature vector, W is the dimensionality-reduced projection weight matrix, and y is the low-dimensional preliminary fingerprint vector after linear transformation. A nonlinear activation function is introduced into the y vector output by the linear transformation to enhance its expressive power under different semantic element combination patterns. Nonlinear functions such as Rectified Linear Unit (ReLU) or Swish are selected and applied to each element in y one by one to achieve the following mapping:

[0097]

[0098] Among them, z iLet represent the i-th component of the initial rule semantic fingerprint, and f(·) be the selected nonlinear activation function, such as the ReLU function or the Swish function. The above linear transformation and nonlinear activation processing are performed on all routing rules, compressing the original high-dimensional sparse multi-source semantic elements into a fixed k-dimensional dense representation space. After completion, the initial rule semantic fingerprint set corresponding to each routing rule is output, providing basic data for subsequent unified normalization and distribution calibration. Through the above-described dimensionality reduction projection network processing method, the high-dimensional weighted fusion feature matrix obtained in the previous step is transformed into a compact and comparable fixed-length initial fingerprint vector, achieving consistency in the logical intent representation of different routing rules in a unified low-dimensional space, laying the foundation for subsequent conflict-aware map construction and similarity inference.

[0099] S2.5: Perform L2 norm normalization and distribution calibration on the initial rule semantic fingerprint to generate the final comparable rule semantic fingerprint set, ensuring that all rule semantic fingerprints have a uniform metric scale in the vector space to support subsequent similarity calculation and conflict detection.

[0100] like Figure 2 As shown, step S3: Based on the logical overlap, conditional mutual exclusion, and action conflict relationships between any two rule semantic fingerprints in the rule semantic fingerprint set, construct a global rule conflict awareness graph containing overlapping edges, mutually exclusive edges, and coupling edges. Specifically, this includes:

[0101] S3.1: Perform high-dimensional spatial distance metric calculation on any two regular semantic fingerprints in the set of regular semantic fingerprints, and use cosine similarity calculation to quantify the geometric proximity of fingerprint vectors in the feature space, so as to generate an initial similarity score matrix for quantifying the geometric proximity between regular semantic fingerprints.

[0102] The input is a set of rule semantic fingerprints that have undergone L2 norm normalization and distribution calibration. Each rule corresponds to a unique fixed-dimensional vector, and all vectors are comparable within the same metric space. For any two fingerprint vectors in this set, a high-dimensional spatial distance metric is used to select pairs of rule semantic fingerprint vectors and perform cosine similarity calculation. Specifically, each rule semantic fingerprint to be compared is denoted as v. i With v j The similarity is quantified using the following cosine similarity formula:

[0103]

[0104] d is the dimension of the semantic fingerprint vector, Sim(v i ,v j ) represents the geometric proximity of the i-th and j-th rules in the feature space, v i,kLet v be the feature value of the semantic fingerprint vector of the i-th rule in the k-th dimension. j,k Let be the feature value of the semantic fingerprint vector of the j-th rule in the k-th dimension. The above calculation process is iterated through all rule combinations in a double-loop manner to generate a symmetric initial similarity scoring matrix. According to the matrix structure, the diagonal elements are always 1, and the remaining elements are between -1 and 1. Positive values ​​represent convergent logical intentions, while negative values ​​reflect divergent logical intentions. Through the above processing method, the standardized set of rule semantic fingerprints is transformed into a two-dimensional scoring matrix representing the similarity of logical intentions among all rules, realizing a quantitative characterization of the potential correlations between rules in a high-dimensional dense feature space.

[0105] S3.2: Based on the high-score rule pairs in the initial similarity scoring matrix, extract the corresponding condition expression structure feature sequence and action behavior label data, perform set inclusion verification and logical negation comparison operations, identify the covering relationship that satisfies the condition full set coverage and the action incompatibility, and the mutually exclusive relationship that satisfies the condition intersection and the action conflicting with each other, so as to generate a primary logical edge set with basic conflict type labels.

[0106] The high-score rule pair refers to the combination of any two rule semantic fingerprints selected from the initial similarity score matrix whose cosine similarity score is higher than a preset threshold. It is directly derived from the traversal and threshold comparison processing of the matrix generated in step S3.1.

[0107] For the initial similarity score matrix generated by high-dimensional spatial distance measurement in the rule semantic fingerprint set, rule pairs with scores higher than a preset threshold are selected as candidate conflict detection objects. For each high-scoring rule pair, its condition expression structural feature sequence and action behavior label data are extracted to form a condition set and an action set to be compared. A set inclusion verification method is used to perform set operations on the condition expression structural feature sequence. By determining whether the condition set of one rule is a proper subset or the entire set of another rule, the hierarchical relationship between the two rules in terms of logical coverage is clarified. For rule pairs that satisfy the condition complete set coverage relationship, further action behavior label comparison processing is performed. Logical compatibility analysis is performed on the action sets of the two rules. If incompatible action outputs, mutual negation, or target resource conflicts are detected, the rule pair is marked as a coverage edge type conflict. For rule pairs that do not satisfy the complete set coverage but have partial intersection, logical negation comparison operation is performed on their action behavior label data. By determining whether the action outputs of the two rules show mutual exclusion, exclusion, or conflict, typical mutually exclusive edge type conflicts are identified. The inclusion relationship between two rule-conditional expression structural feature sequences A and B is quantified using the following formula:

[0108]

[0109] wherein S is an inclusion score, |A| and |B| respectively represent the number of elements in the condition sets of two rules, and |A∩B| is the number of elements in the intersection. When S=1 and the actions are incompatible, the edge is determined as a covering edge; when 0<S<1 and the actions are mutually exclusive, the edge is determined as a mutually exclusive edge. All rule pairs determined to be valid through the above logic judgment are encapsulated into primary logic edge objects with basic conflict type markers (covering, mutually exclusive), and are aggregated to form a primary logic edge set. Through the above chaining processing method, high-score rule pairs in the initial similarity scoring matrix are converted into data structures with clear conflict attributes, realizing automatic identification and structured labeling of basic logic conflict relationships across rules, and laying a data foundation for subsequent dependent link mining and global graph topology construction.

[0110] S3.3: According to the action behavior tags of each rule pair in the primary logic edge set and the input requirement of the conditional expression structure feature of downstream rules, perform directed dependency link topology search processing, track and confirm the coupling relationship forming an implicit data flow closed loop or long-chain dependency, so as to generate a complete logic relationship edge set integrating covering edges, mutually exclusive edges and coupling edges.

[0111] The input is a pre-generated set of primary logical edges, consisting of rule pairs labeled with overlapping and mutually exclusive relationships, and including the action behavior label for each rule and the structural features of the conditional expression of the downstream rule. Based on this input, for each rule pair in the primary logical edge set, the action behavior label and the structural features of the conditional expression of the downstream rule are extracted to establish a one-to-one mapping table between action outputs and conditional inputs. For all action output features, a vectorized comparison method is used to match the vector encoded by the action label with the data field paths and operator types involved in the conditional expressions of the downstream rules, thereby filtering out rule pairs that may have data flow dependencies. For each selected action-condition matching pair, a directed dependency candidate chain is constructed, pointing the upstream rule node to the downstream rule node to form a preliminary directed link. Through topological search processing, all directed dependency candidate chains are recursively traversed to identify and trace paths that form a closed-loop structure (i.e., the data flow returns to itself after multiple hops from a certain node) or a long-chain structure (i.e., the data flow crosses multiple nodes without branch interruption). During the recursive process, for each node sequence on a path, the semantic consistency of actions and conditions, the overlap of applicable scope, and the dependency strength parameters in historical execution feedback are verified among the nodes of each rule semantic fingerprint. Data flow links that are invalid due to business constraint conflicts are filtered out, and only valid coupling links that satisfy implicit dependencies are retained. All valid coupling links are integrated into the original primary logical edge set, and together with covering edges and mutually exclusive edges, a complete logical relationship edge set that integrates covering edges, mutually exclusive edges, and coupling edges is generated. Through the above processing method, the basic conflict type label obtained in the previous step is expanded to include multiple types of logical constraints containing explicit data flow dependencies, realizing a systematic sorting of various constraint relationships in the global rule conflict perception graph.

[0112] S3.4: Using the complete set of logical relationship edges as connection constraints, each rule semantic fingerprint in the rule semantic fingerprint set is mapped to a graph structure node, and the covering edge, mutually exclusive edge and coupling edge are instantiated as graph structure connection edges with directional and weight attributes, so as to construct a global rule conflict perception graph topology that reflects the potential logical constraints between all rules.

[0113] Using a set of rule semantic fingerprints as input, the node identity of each rule semantic fingerprint in the global logical relationship network is determined. Based on the constructed complete logical relationship edge set, each pair of rule semantic fingerprints with overlapping, mutually exclusive, or coupled relationships is mapped to graph structure connection edges with directional and weight attributes. For rule pairs that satisfy the condition of full set coverage but have incompatible actions, they are instantiated as top-down covering edges, and weight parameters representing confidence levels are assigned to these edges. For rule pairs with overlapping conditions and conflicting actions, they are instantiated as bidirectional or unidirectional mutually exclusive edges, and the repulsion strength weight is set according to the frequency of historical conflict triggers and the degree of action denial. For rule sets that form implicit data flow closed loops or long-chain dependencies, they are instantiated as directed coupling edges, and dependency strength weights are assigned according to indicators such as dependency tightness and upstream and downstream feedback correlation. Through the above mapping process, all rule semantic fingerprint nodes and their corresponding logical relationship connection edges are organized in an orderly manner in a unified graph structure container, forming a global rule conflict awareness graph topology with clear node sets and edge sets and complete attributes. This processing method transforms the complete logical relationship edge set generated in the previous step into graph structure data with directionality, weight, and inter-node constraints, thus realizing explicit spatial modeling of the potential logical constraints among all routing rules.

[0114] S3.5: Perform connectivity verification and isolated node cleaning on the topology of the global rule conflict perception graph, and initialize the weight parameters of the graph structure connection edges according to the conflict trigger frequency statistics in the historical execution feedback summary, so as to generate a global rule conflict perception graph with adaptive evolution capability that can be used for subsequent conflict propagation reasoning.

[0115] like Figure 3 As shown, step S4: Based on the target channel attribute features, retrieve a suitable subset of rules from the global rule conflict perception map to generate an initial candidate rule set and its corresponding candidate semantic fingerprint sub-map. The initial candidate rule set includes general rules and channel-specific rules. Specifically, it includes:

[0116] S4.1: Perform structured parsing of the target channel attribute features to extract channel type identifiers, service level agreement thresholds, historical fault frequency statistics, and the capability profile vector of the connected system, generating a multi-dimensional channel feature query vector for graph retrieval.

[0117] Using the target channel attribute feature set as input data, the original channel attribute data structure is first parsed at the field level. Based on the preset attribute schema, the channel type identifier, Service Level Agreement (SLA) threshold, historical fault frequency statistics, and the original vector of the connected system capability profile are extracted sequentially. For the channel type identifier, it is transformed into a one-dimensional or multi-dimensional discrete category vector using a classification coding method to ensure format compatibility with subsequent feature concatenation. For the Service Level Agreement threshold, numerical normalization is performed to standardize SLA indicators with different dimensions across different channels to a unified range, for example, through linear normalization.

[0118]

[0119] Here, SLA is the original service level indicator, and min and max are the minimum and maximum SLAs of all channels in the rule base, respectively. For historical failure frequency statistics, a sliding time window aggregation method is used to calculate statistical parameters such as the cumulative number of failures and average recovery time within a specified time period, and these are encapsulated into a fixed-length numerical vector. The original vector of the interface system capability profile is compressed into a fixed-length capability representation vector based on multi-dimensional features such as interface stability, throughput, and supported function sets, using principal component analysis (PCA) or embedded network dimensionality reduction methods. The above four types of structured features are concatenated in a preset order to form a multi-dimensional channel feature query vector for subsequent graph retrieval operations. Through this processing method, the original heterogeneous channel attribute information is transformed into a structured multi-dimensional channel feature query vector with a unified measurement scale and high expressive power, realizing efficient indexing and matching for the global rule conflict perception graph node space.

[0120] For example, in a multi-channel aggregation service scenario, the target channel attribute data includes: the channel type identifier is "API," with a corresponding code of [1,0,0]; the original service level agreement threshold score is 85, which is normalized to 0.83; the historical failure frequency statistics show a cumulative total of 4 failures in the past 30 days, with an average recovery time of 120 seconds, which is normalized to [0.13,0.22]; the interface system capability profile consists of 10 sub-items, which are reduced to 3 dimensions using PCA to obtain [0.56,0.41,0.68]. Finally, the multi-dimensional feature query vector concatenation result for this target channel is [1,0,0,0.83,0.13,0.22,0.56,0.41,0.68]. In different application scenarios, such as switching the channel type to "SDK" or "Web," the corresponding code and capability profile will be adjusted accordingly to achieve flexible adaptation to different input parameter conditions under various business scenarios. The above processing steps effectively improve rule retrieval efficiency and significantly enhance the sensitivity and expressiveness of the initial candidate rule set screening stage to differences in business context.

[0121] S4.2: Based on the multi-dimensional channel feature query vector, and based on similarity calculation processing, traverse all rule semantic fingerprint nodes in the global rule conflict perception map, calculate the cosine similarity score between each rule semantic fingerprint node and the multi-dimensional channel feature query vector, and generate an initial matching node set containing node identifiers and similarity values.

[0122] Using the multi-dimensional channel feature query vector output from step S4.1 as input, semantic fingerprint vector data for each rule semantic fingerprint node is extracted for all rule semantic fingerprint nodes in the global rule conflict awareness graph. A graph neural network (GNN) node feature propagation mechanism is employed to embed the multi-dimensional channel feature query vector into a feature space representation isomorphic to the rule semantic fingerprint, ensuring a consistent measurement standard for subsequent similarity calculations. For each rule semantic fingerprint node, a cosine similarity calculation is performed, the specific mathematical expression of which is:

[0123]

[0124] Where q represents the multi-dimensional channel feature query vector, v represents the semantic fingerprint of the rule to be compared, and S is the cosine similarity score between the two. The above formula is used to calculate the similarity score for all rule nodes sequentially, generating a unique node identifier and corresponding similarity value associated with each rule semantic fingerprint node. All nodes and their similarity scores are then summarized to form an initial matching node set, where each element contains a unique node ID and a numerical cosine similarity index. For each data point in the initial matching node set, a preset threshold filtering strategy is used to further retain highly relevant rule nodes exceeding the set threshold, providing foundational data support for subsequent applicability logic verification and candidate set selection.

[0125] By combining graph neural network embedding with cosine similarity calculation, the multi-dimensional channel feature query vector is compared one by one with all rule semantic fingerprint nodes in the global rule conflict perception graph. The output is an initial matching node set containing the identifiers of all candidate rule nodes and their relevance to the target channel attributes. This achieves efficient, scalable, and context-sensitive rule retrieval capabilities for complex multi-channel scenarios.

[0126] S4.3: Perform logical verification on the applicable scope metadata associated with each rule semantic fingerprint node in the initial matching node set to filter out invalid nodes that do not match the effective channel list or are not covered by the department whitelist, and generate a subset of valid candidate rule nodes that have been filtered by applicability constraints.

[0127] S4.4: Based on the subset of valid candidate rule nodes, perform graph adjacency edge backtracking operation to extract all covering edges, mutually exclusive edges, and coupled edges whose starting and ending points both belong to the subset of valid candidate rule nodes, as well as their edge weights, and construct local topological structure data that reflects the potential logical relationships between candidate rules.

[0128] Using a subset of valid candidate rule nodes filtered by applicability constraints as input, for each node, all logical connection edges directly connected to that node in the global rule conflict awareness graph are retrieved. For each valid candidate rule node, its adjacency list is traversed sequentially, and covering edges, mutually exclusive edges, and coupled edges are extracted according to edge type labels, with the weight parameters of each connection edge recorded. For covering edges, based on the inclusion relationship of the structural features of the rule condition expressions and the incompatibility of action behavior labels, the covering logic between the source node and the target node is mapped to a directed connection, and the inclusion confidence weight assigned during the graph initialization phase is read synchronously. For mutually exclusive edges, after confirming the mutual exclusion relationship by analyzing the intersection of rule condition expressions and the negation attribute of action behavior labels, the corresponding exclusion strength weight is extracted and a bidirectional or directed connection is established. For coupled edges, a directed topological mapping of the dependency relationship is achieved by tracing the data dependency links between the upstream rule action output features and the downstream rule condition input features, and the dependency tightness value is attached as a weight attribute. After backtracking all adjacent edges, the extracted covering edges, mutually exclusive edges, and coupled edges, along with their corresponding weight parameters, are uniformly organized into local topological structure data, including node sets, connection relationship matrices, and weight tensors. Through the aforementioned graph adjacent edge backtracking and structured aggregation processing, the potential logical constraints between subsets of effective candidate rule nodes are output in an explicit data structure form. This provides basic data support for subsequent assembly of candidate rule sets and generation of candidate semantic fingerprint subgraphs, achieving efficient extraction and visual modeling of multidimensional logical constraints between candidate rules.

[0129] S4.5: Based on the subset of valid candidate rule nodes and the local topology data, assemble an initial candidate rule set containing general rules and channel-specific rules, and simultaneously generate a candidate semantic fingerprint subgraph that maps the internal logical relationships of the initial candidate rule set, as the input object for conflict propagation processing.

[0130] Based on the subset of valid candidate rule nodes obtained through applicability constraint screening and their corresponding local topology data, metadata of the associated rule objects is extracted for each candidate rule node. General rules and channel-specific rules are distinguished and categorized according to information such as channel identifiers and department whitelists in the rule applicability metadata. Index mapping tables are established for the categorized general rules and channel-specific rules to ensure efficient data referencing and deduplication during subsequent candidate rule set assembly. Combining the covering edge, mutually exclusive edge, and coupling edge attributes in the local topology data, the connection relationship in the semantic fingerprint space is analyzed for each pair of logically related candidate rule nodes. The associated rule objects are aggregated in an ordered manner to form an initial candidate rule set reflecting internal logical dependencies. Simultaneously, a candidate semantic fingerprint subgraph is constructed, using all valid candidate rule nodes as subgraph nodes and all covering edges, mutually exclusive edges, and coupling edges in the local topology data whose start and end points belong to the subset of valid candidate rule nodes as subgraph connecting edges. Each edge is assigned a weight parameter from the global graph, achieving a structured expression of the internal logical constraints of the candidate rule set. By performing depth-first traversal to verify connectivity and remove isolated nodes from the candidate semantic fingerprint subgraph, it is ensured that the final input subgraph for conflict propagation processing contains only core candidate rule combinations with practical reasoning value and satisfying contextual constraints. Through this process, the valid candidate rule nodes and their logical relationships obtained in the previous step are transformed into an initial candidate rule set containing general rules and channel-specific rules, as well as a high-quality candidate semantic fingerprint subgraph mapping the internal logical relationships of this initial candidate rule set. This provides an accurate and structured data foundation for subsequent conflict propagation processing, enabling efficient and accurate identification and management of potential conflict paths during policy configuration.

[0131] Step S5: Perform conflict propagation processing on the candidate semantic fingerprint subgraph, identify covering edges, mutually exclusive edges, and coupling edges within the candidate semantic fingerprint subgraph, and generate a set of potential conflict paths labeled with conflict type, impact breadth, and arbitration criteria. Specifically, this includes:

[0132] S5.1: Perform a full traversal of the node pairs in the candidate semantic fingerprint subgraph, and use semantic vector space distance calculation to quantify the logical similarity between any two rule semantic fingerprints, so as to generate an initial association matrix containing high similarity rule pairs.

[0133] S5.2: Based on the high similarity rule pairs in the initial association matrix, execute the condition set inclusion verification and action behavior compatibility comparison logic to identify the covering edges that satisfy the complete inclusion relationship and the mutually exclusive edges that satisfy the condition intersection and action negation relationship, so as to generate a primary conflict edge set with basic conflict attribute tags.

[0134] S5.3: Based on the action output features in the primary conflict edge set and the condition input features of the downstream rules, perform topology search processing for implicit dependency links, track and confirm the coupling edges that form closed loops or long chain dependencies, so as to generate a complete conflict-aware subgraph structure that integrates covering edges, mutually exclusive edges and coupling edges.

[0135] In this embodiment, the action output feature refers to a quantitative description of the operation expected to be executed by each rule when the conditions are met. Specifically, it includes metadata such as action behavior type, target channel identifier, and execution parameters. The action output feature is directly extracted from the structured encoding of the definition of each rule node in the candidate semantic fingerprint subgraph and serves as a direct basis for determining the impact of rule execution on the external system or data state.

[0136] The conditional input features of the downstream rules refer to a structured description of the preconditions upon which a rule depends for being triggered, specifically including elements such as field reference paths, threshold ranges, and dependent triggering conditions. These conditional input features of the downstream rules are also obtained by structured parsing of the definition of rule nodes in the candidate semantic fingerprint subgraph, and are used to determine whether their triggering depends on the execution results of other rule actions.

[0137] For the marked covering and mutually exclusive edges in the primary conflict edge set, the input objects are the action output features of each pair of rule nodes and the conditional input features of the downstream rules. Based on the nodes and edges of the candidate semantic fingerprint subgraph, the action output labels of all rule nodes are first structured and encoded to extract metadata information such as action behavior type, target channel identifier, and execution parameters. The conditional input features of the downstream rule nodes are then structured and parsed in the same way to obtain elements such as field reference paths, threshold ranges, and dependent triggering conditions. Based on the above two types of feature data, a directed graph traversal process is used to determine whether the action output of each node pair connected by a primary conflict edge (including covering and mutually exclusive edges) has a parameter mapping relationship or data flow dependency with the conditional inputs of other rules. For node pairs that satisfy the direct or indirect triggering relationship between the action output and the conditional input, they are marked as potential coupling candidate edges in the subgraph. The process continues to recursively trace along the identified coupling candidate edges, using a depth-first search (DFS) and loop detection mechanism to expand the dependency links layer by layer within the local subgraph. For each path from the upstream rule to the downstream rule, a closed-loop determination and long-chain length statistics are performed. If the starting point and ending point of a path coincide, the path is confirmed to form an implicit dependency loop. If the path length is greater than a preset threshold and there are no branch breaks, the path is identified as a long-chain dependency. All paths that meet the conditions for loop closure or long chain dependency are added to the conflict-aware subgraph as directed coupling edges, and each coupling edge is assigned a dependency strength weight. This weight can be calculated based on parameters such as the linkage hit rate and average response latency in the historical execution feedback summary using the following weighting formula:

[0138]

[0139] Among them, w i,j r represents the coupling edge dependency strength weight from rule i to rule j. i,j d is the normalized value of the linkage hit rate. i,j This represents the normalized value for the average response delay. Through the above processing method, the initial set of conflicting edges is transformed into a complete conflict-aware subgraph structure containing three types of logical connections: covering edges, mutually exclusive edges, and coupled edges confirmed by topological search. This achieves explicit modeling of the complex logical constraints within the candidate rule set.

[0140] For example, in a multi-channel aggregation service system, eight candidate rules selected under a channel strategy configuration scenario have action output labels including "route configuration to a certain node," "downgrade to a fallback interface," and "write to alarm log," while downstream rule input features include fields such as "preceding route successful," "fallback interface available," and "alarm log exists." The system, through structured coding, discovers that the action output of rule #3, "downgrade to a fallback interface," corresponds precisely to the condition input of rule #5, "fallback interface available," forming a direct data flow dependency. DFS recursive tracing further reveals that rule #5 has an indirect triggering relationship with rule #7, "write to alarm log," thus forming a long chain dependency path of length 3. In historical execution feedback, the hit rate of rule linkage from #3 to #5 to #7 is 0.82, and the average response latency, after normalization, is 0.36. Therefore, its dependency strength weight is calculated using the formula above:

[0141]

[0142] Ultimately, this step outputs a complete conflict-aware subgraph that integrates all coverage, mutual exclusion, and coupling logical connections, providing a solid data foundation for subsequent multi-hop path propagation and dynamic priority calibration, and significantly improving the ability to automatically identify and enforce constraints on implicit dependency links in complex policy scenarios.

[0143] S5.4: Perform multi-hop path diffusion propagation processing on the complete conflict-aware subgraph structure to simulate the transmission process of conflict signals between nodes and calculate the influence radius of each conflict path, so as to generate a set of conflict influence breadth indicators that quantifies the scope of conflict.

[0144] S5.5: Combining the aforementioned set of conflict impact breadth indicators with the preset channel service level agreement arbitration strategy, perform conflict nature determination and arbitration rule matching operations on each identified conflict path to generate a final set of potential conflict paths marked with conflict type, impact breadth, and arbitration basis.

[0145] In this embodiment, the channel service level agreement (SLA) arbitration strategy is a predefined set of rules used to guide conflict arbitration decisions. Its core logic is to determine whether accuracy or availability should be prioritized in conflict arbitration based on a comparison between the target channel's SLA threshold and a preset threshold. This channel SLA arbitration strategy originates from an inductive analysis of channel SLA requirements and system optimization goals, such as precise routing or high-availability routing, under different business scenarios, and is solidified into a set of executable judgment rules.

[0146] Using the conflict impact breadth index set in the candidate semantic fingerprint subgraph and the preset channel service level agreement (SLA) arbitration strategy as input, each identified conflict path is processed for conflict nature determination and arbitration rule matching. First, for each conflict path, its associated conflict type markers are extracted, including covering edges, mutually exclusive edges, and coupling edges, and the influence radius value of the path in the conflict impact breadth index set is obtained. For covering edges, by analyzing conditional inclusion relationships and action incompatibility, they are determined to be "forced priority covering type" conflicts, and the impact breadth value is used as a weight reference for priority level adjustment. For mutually exclusive edges, dynamic arbitration is performed based on action conflict attributes and SLA threshold parameters, according to the following decision rules: when the target channel SLA is higher than the preset threshold, accuracy is prioritized, and high-confidence action rules are placed with higher priority; when the SLA is lower than the threshold, availability is prioritized, and rules with fast response speed or high hit rate are placed with higher priority. This decision logic can be formally expressed as:

[0147]

[0148] Where P is the final priority sorting position, S i For service level weighting, W i To influence the breadth coefficient, I iThe confidence score is assigned to each rule. For coupled edges, by tracing the dependency loop structure, all rules forming implicit dependency chains are packaged into atomic scheduling units, and all rules within each unit are assigned a consistent priority anchor label, while the maximum influence breadth is used as the arbitration benchmark. After determining the nature of the conflict, automatic arbitration rule matching is performed on each conflict path: for covering conflicts, mandatory hierarchical constraints are introduced; for mutually exclusive conflicts, confidence labels and recommended order are output through arbitration; for coupled conflicts, the internal order is solidified in atomic groups. Finally, each potential conflict path is assigned a clear conflict type label, a specific quantified influence breadth value, and an automatically generated arbitration basis label, forming a structured set of potential conflict paths, providing a decision-making basis for subsequent priority reordering. Through the above multi-dimensional chain processing method, the candidate semantic fingerprint subgraph and its diffusion propagation results generated in the previous step are transformed into a set of potential conflict paths with quantitative attributes and automatic arbitration capabilities, realizing intelligent conflict identification and processing under complex logical constraints.

[0149] For example, in a multi-channel aggregation service system, for candidate rule sets A, B, and C, A and B have complete condition coverage but incompatible actions, and are therefore classified as forced priority coverage type; B and C have partial condition intersection and mutually negating actions, and are classified as mutually exclusive type; A and C form a long chain dependency through action outputs and condition inputs, and are classified as coupled type. The system uses parameters such as channel SLA=9.5 (out of 10), B hit rate 0.82, and C average response latency 30ms, and uses the formula:

[0150]

[0151] The calculations show that path B should be given higher priority. In the final output, the AB path is marked as "coverage / breadth of impact 2 / mandatory hierarchy", the BC path is marked as "mutually exclusive / breadth of impact 1 / SLA arbitration - accuracy priority", and the AC link is marked as "coupled / breadth of impact 3 / atomic group anchoring". The application effect is that the subsequent policy configuration stage can automatically adopt the above arbitration results, which can significantly improve the accuracy of configuration and system consistency, and effectively avoid logical contradictions or policy failures caused by human error.

[0152] Step S6: Based on the conflict types in the potential conflict path set and the policy objective weights, rearrange the rule priority sequences in the initial candidate rule set to generate a preliminary priority sequence that has been context-sensitively calibrated. Specifically, this includes:

[0153] S6.1: Obtain mutually exclusive edge data from the potential conflict path set and service level agreement indicators from the target channel attribute characteristics. Perform priority adjudication on mutually exclusive rule pairs with conflicting action relationships through dynamic arbitration based on service level agreement indicators to generate mutual exclusive rule priority determination results with arbitration confidence labels.

[0154] S6.2: Receive the priority determination result of the mutual exclusion rule and the coverage edge data in the potential conflict path set, perform a forced hierarchical mapping operation based on the condition inclusion range, set the priority benchmark value of the covering rule to be higher than the priority benchmark value of the covered rule, so as to generate a rule priority skeleton sequence containing forced hierarchical constraint relationship.

[0155] Using the priority determination results of mutually exclusive rules and the overlapping edge data in the potential conflict path set as input, all rule pairs marked as overlapping relationships are selected as the processing objects for forced hierarchical mapping. For each pair of rules with overlapping edges, its condition expression structure feature sequence is extracted. The condition set is then checked for subset and superset relationships through set inclusion determination to determine whether the condition of the covered rule is completely contained within the condition of the covering rule. For rule pairs that satisfy the above condition inclusion, their current priority baseline value parameter is extracted, and a priority adjustment operation queue is constructed. This operation queue is traversed sequentially. For each pair (covering rule A, covered rule B), the following forced hierarchical constraint is set: the priority baseline value of A is greater than the priority baseline value of B. Specifically, the following linear promotion formula can be used for adjustment:

[0156]

[0157] Where P represents the rule's priority baseline value, A is the identifier of the overriding rule, B is the identifier of the overridden rule, and δ is the system-preset minimum priority interval. If multiple chained transit relationships exist in the operation queue, a recursive approach is used to perform batch hierarchical mapping on all relevant rules, dynamically maintaining the uniqueness and monotonicity of the adjusted priority baseline values. After all mandatory hierarchical constraints have been processed, all adjusted rules and their corresponding new priority baseline values ​​are assembled into a rule priority skeleton sequence containing the mandatory hierarchical constraint relationships.

[0158] By using the above-mentioned forced hierarchical mapping processing method based on the conditional inclusion range, the coverage edges identified in the previous step are transformed into explicit and irreversible priority ordering constraints. This achieves hard order control required by logical inclusion relationships within the candidate rule set, providing a solid data framework for subsequent coupled link packaging and global multi-objective rearrangement.

[0159] S6.3: Extract the associated coupling edge data and dependency strength parameters from the historical execution feedback summary in the rule priority skeleton sequence, and merge the set of coupling rules forming implicit dependency chains into indivisible priority scheduling units through atomic priority group packaging processing to generate a list of coupled rule atomic groups with internal order locking characteristics.

[0160] The input data includes the generated rule priority skeleton sequence, the coupling edge data associated with each rule node in the sequence, and the dependency strength parameters between each rule node recorded in the historical execution feedback summary. For each rule in the rule priority skeleton sequence, all its directly coupled edges in the global conflict-aware graph are retrieved, and the downstream or upstream rule nodes connected to it are extracted, thereby constructing a preliminary set of coupling relationship links. For the above set of coupling relationship links, combined with the dependency strength parameters in the historical execution feedback summary, high-strength coupling edges with dependency strength greater than a preset threshold θ are filtered out through threshold judgment processing. The coupling edges that meet the conditions are topologically sorted to form multiple implicit dependency chains. For each implicit dependency chain, atomic priority grouping processing is used to merge all rule nodes on the same dependency chain into an indivisible priority scheduling unit, and a unique identifier is assigned to this unit to achieve internal order locking. The atomic grouping weight of each dependency chain is quantified using the following formula:

[0161]

[0162] Among them, s i w represents the dependency strength of the i-th coupled edge within the group, N is the number of coupled edges within the group, and w group Grouping weights are assigned. All atomic groups and their weight information are aggregated to generate a list of coupled rule atomic groups with internal order locking properties. This list serves as a crucial input parameter for the subsequent multi-objective weighted scoring and reordering engine. Through this processing method, the rule priority skeleton sequence and coupling relationship data obtained in the previous step are transformed into structured, indivisible scheduling units, achieving consistent constraints and efficient management of priority order under complex implicit dependencies.

[0163] S6.4: Integrate the mutual exclusion rule priority determination results, rule priority skeleton sequence and coupled rule atomic grouping list, and combine the business continuity coefficient and response delay coefficient in the strategy target weight to run the multi-objective weighted scoring and reordering engine to perform global order optimization calculation on the candidate rule set, so as to generate a preliminary rule priority sequence that has been context-sensitively calibrated.

[0164] Using the priority determination results of mutually exclusive rules, the rule priority skeleton sequence, and the atomic grouping list of coupled rules as input, and combining the business continuity coefficient and response latency coefficient in the strategy objective weights, a global order optimization calculation is performed on the candidate rule set. For each pair of mutually exclusive rules, a confidence weight is assigned according to its priority determination result, and this weight is used as the initial parameter for subsequent weighted scoring. For each level of relationship in the rule priority skeleton sequence, the mandatory order constraints corresponding to the covering edges are extracted, and the sorting relationship between the covering and covered parties is input into the optimization engine in the form of hard constraints. For each group in the atomic grouping list of coupled rules, its internal order is locked to ensure that each rule node in the same group maintains an indivisible relative position in the final sorting. For each candidate rule, based on its role attributes in the above three types of structures (such as whether it is an arbitration node, whether it is at the beginning or end of the dependency chain, and whether it is a high-confidence determination object), a multi-objective weighted scoring function is constructed. The business continuity coefficient and response latency coefficient are assigned to different objective dimensions, and a conflict impact breadth index is introduced as a conflict reconciliation term. The following multi-objective weighted scoring formula is adopted:

[0165]

[0166] Among them, C i L is the business continuity contribution factor for the i-th rule. i P is the contribution factor for the response delay of the i-th rule. i Let α be the breadth of influence of the i-th rule in the conflict path, β be the business continuity coefficient, β be the response delay coefficient, and γ be the breadth of conflict impact. Based on the above scoring function, all candidate rules are globally scored and ranked to generate a preliminary rule priority sequence. During the ranking process, order anomalies caused by hard constraints, dependency locking, or confidence conflicts are dynamically detected and corrected. For ranking results that violate mandatory hierarchical relationships or coupled grouping order, recursive adjustment is used to automatically backtrack to the most recent valid state. Through a multi-objective weighted scoring and re-ranking engine, the multi-dimensional structured constraints formed in the previous step are organically integrated with the policy objective weights, realizing context-sensitive global order optimization of the candidate rule set, and outputting a calibrated preliminary rule priority sequence, providing a foundation for subsequent consistency verification and boundary locking.

[0167] S7: Receive the user-defined highest non-adjustable rule anchor point and lowest basic rule anchor point, combine them with the coupling edge constraints in the global rule conflict awareness graph, perform rule boundary locking on the preliminary priority sequence, and adjust the positions of unlocked floating rules according to the monotonically increasing overall priority sequence without logical breaks, generating the final rule priority sequence. Specifically, this includes:

[0168] S7.1: Obtain the highest non-adjustable rule anchor point identifier and the lowest basic rule anchor point identifier input by the user through the configuration interface, and use identity authentication and format verification methods to verify the legality of the anchor point identifiers in order to generate a set of policy boundary constraints containing valid rule reference pointers.

[0169] The system receives a preliminary rule priority sequence calibrated with context sensitivity and the highest non-adjustable rule anchor identifier and the lowest basic rule anchor identifier entered by the user in the configuration interface. These two user inputs serve as the initial data source for policy boundary constraints. Identity authentication is performed on the highest non-adjustable rule anchor identifier and the lowest basic rule anchor identifier entered by the user. Multi-factor authentication based on access tokens and session credentials verifies the current user's permission level, ensuring they have the management authority to configure and lock the specified rule anchors. After authentication, format validation is performed on the two anchor identifiers. Regular expression matching and unique hash retrieval mechanisms are used to verify whether the input identifiers conform to the system's preset rule naming conventions (such as length, character set, prefix convention, etc.), and the corresponding unique rule entity reference is retrieved from the global routing rule base. For valid anchor identifiers that pass authentication and format validation, a data structure containing pointers to the highest non-adjustable rule and the lowest basic rule is constructed. This structure is incorporated into the policy boundary constraint set, and each pointer is assigned a unique index number to support subsequent dependency retrieval. If any anchor point fails to pass any of the above steps, the exception handling process is automatically triggered, returning detailed error types and correction suggestions, and recording them in the system security audit log, thus ensuring the traceability and security of the operation process. Through chained processing of identity authentication and format verification, user input is transformed into a set of policy boundary constraints that have been verified for legitimacy and possess uniqueness, thereby achieving data entry control for subsequent priority locking and dependency modeling.

[0170] S7.2: Based on the valid rule reference pointers in the policy boundary constraint set, retrieve the corresponding rule semantic fingerprint nodes in the global rule conflict perception graph, and extract all coupling edge data directly connected to the nodes to construct a local coupling link topology containing strong dependencies.

[0171] The input data includes a set of policy boundary constraints processed by identity authentication and format validation. This set contains valid rule reference pointers corresponding to the highest unadjustable rule anchor point identifier and the lowest basic rule anchor point identifier set by the user. Using these rule reference pointers as indexes, the corresponding rule semantic fingerprint nodes are retrieved from the data structure of the global rule conflict awareness graph, ensuring node uniqueness and node attribute integrity. For each retrieved rule semantic fingerprint node, its adjacency list or graph structure is traversed to identify direct connections, filtering out all edge sets of type coupling edges, excluding overlay edges and mutually exclusive edges, and retaining only connection information reflecting implicit dependencies or sequential coupling relationships. Attribute unpacking is performed on the above coupling edge data to extract the upstream and downstream rule node identifiers, dependency strength parameters, and historical feedback weight indicators associated with each coupling edge, forming a structured tuple set describing local dependency links. A Directed Acyclic Graph (DAG) topology is employed, starting with the highest non-adjustable rule anchor node. All downstream rule nodes directly or indirectly reachable through coupling edges are recursively incorporated into the local link topology. During each hop of recursion, link depth, dependency paths, and critical path weight distribution are recorded in real-time, enabling a complete extraction of the entire local dependency network. Through this processing, the critical nodes specified by valid rule reference pointers in the policy boundary constraint set and all their direct coupling relationships are transformed into a locally coupled link topology with hierarchical order, dependency strength, and path attribute labels. This achieves explicit modeling of strong dependencies within priority intervals, providing a high-quality data foundation for subsequent transitive closure calculation and rigid priority constraint chain generation.

[0172] S7.3: Perform transitive closure calculation on the locally coupled link topology to identify all implicit dependency paths from the highest non-adjustable rule anchor point to the lowest basic rule anchor point, so as to generate a rigid priority constraint chain marked with mandatory order relations.

[0173] Based on the set of valid rule reference pointers obtained after identity authentication and format verification, the input consists of the corresponding rule semantic fingerprint nodes of the highest non-adjustable rule anchor and the lowest basic rule anchor in the global rule conflict perception graph, along with their directly connected coupling edge data, forming a locally coupled link topology containing strong dependencies. Directed graph transitive closure computation is performed on all nodes and their connecting edges in the locally coupled link topology. A path traversal algorithm based on depth-first search (DFS) or breadth-first search (BFS) is used, starting from the highest non-adjustable rule anchor node, recursively searching all reachable paths, and completely enumerating each directed path from that anchor to the lowest basic rule anchor. For each reachable path, the coupling edge attributes between nodes on the path are analyzed, the dependency strength is determined based on the edge weight parameters, and edges with dependency strength greater than a preset threshold are marked and treated as indivisible rigid order constraints. All enumerated reachable paths are deduplicated and merged. If multiple different paths cover the same set of nodes, these node sets are merged into a single rigid priority constraint chain to ensure that all implicit dependencies are completely captured without redundancy. The identified rigid priority constraint chains are sequentially encoded, with each chain starting at the highest unadjustable rule and ending at the lowest basic rule, arranged sequentially according to dependency direction, forming a data structure that explicitly records mandatory order relationships. Through the aforementioned transitive closure calculation and dependency determination processing, the locally coupled link topology generated in the previous step is transformed into rigid priority constraint chains marked with mandatory order relationships. This achieves explicit constraint expression of implicit dependency paths during rule priority adjustment, effectively preventing dependency chain breaks and business logic errors caused by manual sorting or automatic optimization. S7.4: The rigid priority constraint chains are mapped to a context-sensitive calibrated preliminary priority sequence, and interval locking and boundary detection processing are performed. The rule sorting positions that violate the rigid priority constraint chains are automatically corrected to generate an intermediate locked priority sequence that satisfies the graph coupling constraints.

[0174] Using the rigid priority constraint chain obtained from transitive closure calculation as input, we obtain all rule entries and their current sorting positions in the context-sensitive calibrated preliminary priority sequence. For each rule in the preliminary priority sequence, we compare its node order in the rigid priority constraint chain to form a mapping relationship between rule index and constraint chain order. Based on this mapping relationship, we set interval locking boundaries, using the position of the rule corresponding to the highest non-adjustable rule anchor point in the sequence as the upper bound and the position of the rule corresponding to the lowest basic rule anchor point in the sequence as the lower bound, to determine the priority interval to be locked. For all rule nodes on the rigid constraint chain within this interval, we perform out-of-bounds detection: if a rule's sorting position is found to violate the internal order of the constraint chain (i.e., a subsequent node appears before a preceding node), we automatically adjust the rule to the correct relative position that meets the requirements of the rigid constraint chain, and update the positions of other related rules within the interval to maintain overall order consistency. For floating rules outside the interval, no adjustment is performed; only their original sorting state is retained. Through the above interval locking and out-of-bounds detection processing, the initial priority sequence is transformed into an intermediate locking priority sequence that strictly satisfies the graph coupling constraints and whose internal order completely conforms to the requirements of the rigid priority constraint chain, thereby achieving mandatory protection of key dependencies and user-specified anchor points.

[0175] S7.5: Based on the relative positional relationship of each rule in the intermediate locking priority sequence, adjust the range limitation of the unlocked floating rules to eliminate the logical breaks caused by boundary locking, so as to output the final rule priority sequence.

[0176] Based on the relative positions of rules in the intermediate locking priority sequence, the sequence is first structured and analyzed to identify all rule intervals locked by boundary anchors and coupling dependency chains. Unlocked floating rules are then marked as fine-tuning targets. For each floating rule, its associated nodes in the global rule conflict awareness graph are retrieved, and the constraints of overlapping edges, mutually exclusive edges, and coupling edges with locked rules are extracted. Based on this, the upper and lower bounds of the adjustable priority interval for that floating rule in the current sequence are dynamically calculated. All floating rules are recursively traversed according to their original priority order, limiting the adjustable range of each floating rule to within the rigid constraint chains and locked intervals, ensuring that the overall priority sequence after fine-tuning is monotonically increasing or decreasing without logical breaks.

[0177] By constructing a set of interval intersections and using the following interval clipping formula, its final adjustable range is precisely limited:

[0178]

[0179] Among them, L i U iLet represent the minimum and maximum adjustable priority values ​​of the i-th floating rule obtained from the graph constraints, respectively, which are the closed intervals that are actually allowed to be adjusted. After performing the above interval pruning operation on all floating rules, linear interpolation or local weighted smoothing methods are then used to correct the priority continuity of the remaining uncovered micro-fault regions to prevent logical sorting jumps caused by boundary locking.

[0180] For example, in a multi-channel aggregation service scenario, the intermediate priority sequence contains 10 routing rules. Rules 2 and 9 are set by the user as the highest non-adjustable anchor point and the lowest basic anchor point, respectively, and rules 4, 5, and 6 form a strongly coupled dependency chain and are locked as a whole. The system automatically identifies rules 3, 7, and 8 as unlocked floating rules. Taking rule 3 as an example, its upper bound in the graph is defined by anchor point 2, and its lower bound by the coupling group of rule 4, as obtained through retrieval. After applying the interval pruning formula, rule 3 is only allowed to be fine-tuned between 21 and 30. Similarly, rules 7 and 8 obtain their respective adjustable closed intervals [61, 67] and [68, 75] based on the locked intervals before and after them. Subsequently, the system uses linear interpolation to fill the sorting gaps caused by the anchor points, assigning rules 3, 7, and 8 values ​​of 25, 64, and 72 respectively, achieving a monotonically increasing overall priority sequence without logical jumps. In actual operation, this final priority sequence effectively ensures the consistency of routing strategies for critical business channels and the ability to avoid conflicts, significantly improving the accuracy of strategy configuration and the ease of maintenance.

[0181] Step S8: Manual review. Strategy personnel or reviewers verify the business rationality of the final rule priority sequence, focusing on whether conflict resolution meets expectations and whether anchor point settings are accurate. After confirmation, the rule is issued for execution or rolled back for adjustment.

[0182] Strategy personnel conduct mandatory reviews of the final rule priority sequence to verify the business rationality of the conflict resolution strategy. They focus on checking whether conflicts involving coverage, mutual exclusion, and coupling have been properly resolved through priority rearrangement and boundary locking operations. This ensures that rules specific to key channels are not unexpectedly downgraded, mutual exclusion conditions do not cause matching gaps, and conflicting actions do not disrupt interface calls. Simultaneously, they review the accuracy and completeness of the highest non-adjustable rule anchor point and the lowest basic rule anchor point set by the user. They confirm that the mandatory top-level priority strategy and the fallback execution baseline strategy meet business expectations, and that the floating rule sequence between the two anchor points is monotonically increasing without logical gaps. If deviations are found during the review, reviewers can directly adjust the floating rule order or re-specify the anchor point position in the visual interface, and the system will automatically recalculate the affected local sequence. For irresolvable circular dependencies or mutual exclusion rule pairs, reviewers can either forcibly retain one and record the decision reason, or remove the entire conflicting rule set. Upon successful review, electronic signatures and audit logs are submitted, triggering execution; otherwise, the process reverts to the candidate set generation or weight adjustment stage for recalculation. This step serves as the final quality gate for strategy configuration, compensating for the algorithm's shortcomings in implicit business preferences and fuzzy decision-making in complex scenarios, and ensuring that the routing rule sequence has both logical consistency and applicability to real business.

[0183] In one embodiment, a multi-channel aggregation service routing policy configuration system is provided for executing the steps of a multi-channel aggregation service routing policy configuration method described in any of the above embodiments.

[0184] This application provides a multi-channel aggregation service routing policy configuration method, which can be applied to, for example... Figure 5 In the application environment shown, terminal 102 communicates with server 104 via a network. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices, and server 104 can be a standalone server or a server cluster consisting of multiple servers.

[0185] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores data generated during the implementation of a multi-channel aggregation service routing policy configuration method. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements a multi-channel aggregation service routing policy configuration method.

[0186] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0187] S1: Obtain the target channel attribute characteristics and strategy target weight of the channel strategy to be configured, and extract the semantic elements of each rule in the routing rule base;

[0188] S2: Based on the semantic elements, map each routing rule to a fixed-dimensional rule semantic fingerprint to generate a rule semantic fingerprint set;

[0189] S3: Based on the logical overlap, conditional mutual exclusion, and action conflict relationships between any two rule semantic fingerprints in the rule semantic fingerprint set, construct a global rule conflict perception map containing overlay edges, mutual exclusion edges, and coupling edges;

[0190] S4: Based on the target channel attribute features, retrieve the appropriate rule subset in the global rule conflict perception map to generate an initial candidate rule set and its corresponding candidate semantic fingerprint sub-map;

[0191] S5: Perform conflict propagation processing on the candidate semantic fingerprint subgraph, identify covering edges, mutually exclusive edges and coupling edges in the candidate semantic fingerprint subgraph, and generate a set of potential conflict paths marked with conflict type, impact breadth and arbitration basis;

[0192] S6: Based on the conflict type in the potential conflict path set and the strategy target weight, rearrange the rule priority sequence in the initial candidate rule set to generate a preliminary priority sequence;

[0193] S7: Receive the highest non-adjustable rule anchor point and the lowest basic rule anchor point set by the user, combine the coupling edge constraints in the global rule conflict perception graph, lock the rule boundaries of the preliminary priority sequence, and adjust the positions of the unlocked floating rules according to the monotonically increasing overall priority sequence without logical breaks, to generate the final rule priority sequence.

Claims

1. A method for configuring routing strategies for multi-channel aggregation services, characterized in that, Construct a routing rule library that includes general routing rules and channel-specific routing rules. Divide routing policies into two categories: general policies and channel policies. General policies can only use general routing rules from the rule library and cannot use any channel-specific routing rules. Channel policies can use general routing rules from the rule library and channel-specific routing rules corresponding to the channel.

2. The multi-channel aggregation service routing policy configuration method according to claim 1, characterized in that, The channel strategy configuration method specifically includes: S1: Obtain the target channel attribute characteristics and strategy target weight of the channel strategy to be configured, and extract the semantic elements of each rule in the routing rule base; S2: Based on the semantic elements, map each routing rule to a fixed-dimensional rule semantic fingerprint to generate a rule semantic fingerprint set; S3: Based on the logical overlap, conditional mutual exclusion, and action conflict relationships between any two rule semantic fingerprints in the rule semantic fingerprint set, construct a global rule conflict perception map containing overlay edges, mutual exclusion edges, and coupling edges; S4: Based on the target channel attribute features, retrieve the appropriate rule subset in the global rule conflict perception map to generate an initial candidate rule set and its corresponding candidate semantic fingerprint sub-map; S5: Perform conflict propagation processing on the candidate semantic fingerprint subgraph, identify covering edges, mutually exclusive edges and coupling edges in the candidate semantic fingerprint subgraph, and generate a set of potential conflict paths marked with conflict type, impact breadth and arbitration basis; S6: Based on the conflict type in the potential conflict path set and the strategy target weight, rearrange the rule priority sequence in the initial candidate rule set to generate a preliminary priority sequence; S7: Receive the highest non-adjustable rule anchor point and the lowest basic rule anchor point set by the user, combine the coupling edge constraints in the global rule conflict perception graph, lock the rule boundaries of the preliminary priority sequence, and adjust the positions of the unlocked floating rules according to the monotonically increasing overall priority sequence without logical breaks, to generate the final rule priority sequence.

3. The multi-channel aggregation service routing policy configuration method according to claim 2, characterized in that, The target channel attribute characteristics to be configured can be obtained using natural language input.

4. The multi-channel aggregation service routing policy configuration method according to claim 2, characterized in that, Following S7, the following also includes: S8: Manual review, where strategy personnel or reviewers verify the business rationality of the final rule priority sequence, focusing on whether conflict resolution meets expectations and whether anchor point settings are accurate. Once confirmed to be correct, the rule is issued for execution or rolled back for adjustment.

5. The multi-channel aggregation service routing policy configuration method according to claim 2, characterized in that, Step S3 specifically includes: Perform high-dimensional spatial distance metric calculation on any two regular semantic fingerprints in the set of regular semantic fingerprints to generate an initial similarity score matrix; Based on the high-score rule pairs in the initial similarity scoring matrix, set inclusion verification and logical negation comparison operations are performed to identify covering relationships that satisfy the condition of full set coverage and incompatible actions, as well as mutually exclusive relationships that satisfy the condition of intersection and conflicting actions, and to generate a primary logical edge set. Based on the action behavior labels of each rule pair in the primary logical edge set and the condition expression structure feature input requirements of the downstream rules, a directed dependency link topology search process is performed to track and confirm the coupling relationship that forms an implicit data flow closed loop or long chain dependency, and to generate a complete logical relationship edge set that integrates the covering edge, mutually exclusive edge and coupling edge. Using the complete logical relationship edge set as connection constraints, each rule semantic fingerprint in the rule semantic fingerprint set is mapped to a graph structure node, and the covering edge, mutual exclusion edge and coupling edge are instantiated as graph structure connection edges with directional and weight attributes to construct a global rule conflict perception graph topology. The global rule conflict awareness graph topology is subjected to connectivity verification and isolated node cleaning. The weight parameters of the graph structure connection edges are initialized based on the conflict trigger frequency statistics in the historical execution feedback summary to generate the global rule conflict awareness graph.

6. The multi-channel aggregation service routing policy configuration method according to claim 2, characterized in that, Step S4 specifically includes: The target channel attribute features are structured and parsed to extract channel type identifiers, service level agreement thresholds, historical fault frequency statistics, and system capability profile vectors, generating multi-dimensional channel feature query vectors for graph retrieval. Based on the multi-dimensional channel feature query vector, and based on similarity calculation processing, all rule semantic fingerprint nodes are traversed in the global rule conflict perception map, and the cosine similarity score between each rule semantic fingerprint node and the multi-dimensional channel feature query vector is calculated to generate an initial matching node set containing node identifiers and similarity values. Logical verification is performed on the applicable scope metadata associated with each rule semantic fingerprint node in the initial matching node set to generate a subset of valid candidate rule nodes; Based on the subset of valid candidate rule nodes, a graph adjacency edge backtracking operation is performed to extract all covering edges, mutually exclusive edges, and coupled edges whose starting and ending points both belong to the subset of valid candidate rule nodes, along with their edge weights, and to construct local topology data. Based on the subset of valid candidate rule nodes and the local topology data, an initial candidate rule set containing general rules and channel-specific rules is assembled, and the candidate semantic fingerprint subgraph that maps the internal logical relationships of the initial candidate rule set is generated simultaneously.

7. The multi-channel aggregation service routing policy configuration method according to claim 2, characterized in that, Step S5 specifically includes: The node pairs in the candidate semantic fingerprint subgraph are fully traversed, and the logical similarity between any two rule semantic fingerprints is quantified by semantic vector space distance calculation, generating an initial association matrix containing highly similar rule pairs. Based on the high similarity rule pairs in the initial association matrix, the condition set inclusion verification and action behavior compatibility comparison logic are executed to identify the covering edges that satisfy the complete inclusion relationship and the mutually exclusive edges that satisfy the condition intersection and action negation relationship, thereby generating a primary conflict edge set. Based on the action output features in the primary conflict edge set and the condition input features of the downstream rules, perform topology search processing for implicitly dependent links, track and confirm the coupling edges that form closed loops or long chain dependencies, and generate a complete conflict-aware subgraph structure that integrates the covering edges, mutually exclusive edges and coupling edges. Multi-hop path diffusion propagation processing is performed on the complete conflict-aware subgraph structure to simulate the transmission process of conflict signals between nodes and calculate the influence radius of each conflict path, generating a set of conflict influence breadth indexes. By combining the aforementioned set of conflict impact breadth indicators with the preset channel service level agreement arbitration strategy, a conflict nature determination and arbitration rule matching operation is performed on each identified conflict path to generate the final set of potential conflict paths marked with conflict type, impact breadth, and arbitration basis.

8. A multi-channel aggregation service routing policy configuration system, used to execute the steps of the multi-channel aggregation service routing policy configuration method according to any one of claims 1-7.

9. A computer device, characterized in that, include: A processor, a memory, and a computer program stored in the memory and executable on the processor; the computer program, when executed by the processor, implements the steps of a multi-channel aggregation service routing policy configuration method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the steps of a multi-channel aggregation service routing policy configuration method as described in any one of claims 1 to 7.

Images