Activity configuration method and device, electronic equipment, medium and program product

Abstract

The invention provides an activity configuration method and device, electronic equipment, a medium and a program product, relates to application of a large model in operation activity configuration, and can be applied to the technical field of big data, the technical field of artificial intelligence and the field of financial science and technology. The method comprises the following steps: acquiring operation target information of a target activity; obtaining a pre-constructed multi-modal knowledge base, and retrieving and generating a target semantic context from a semantic map of the multi-modal knowledge base based on the operation target information; performing semantic analysis and business reasoning on the target semantic context by using a large language model to generate an initial activity configuration result corresponding to the operation target information; and performing dynamic compliance auditing on the initial activity configuration result, and determining a final activity configuration result based on an auditing result.

Description

TECHNICAL FIELD

[0001] The present application relates to the application of large models in operation activity configuration, relates to the technical field of big data, the technical field of artificial intelligence and the technical field of financial technology, and more particularly to an activity configuration method, device, equipment, medium and program product. BACKGROUND

[0002] The existing operation activity configuration system generally relies on manual entry of parameters by business personnel, resource selection and template binding, and the system only provides limited format checking capability and cannot understand and infer business semantics or logical relationships. Due to the lack of an analysis mechanism for operation targets, the system cannot automatically generate activity structures or parameters according to activity intentions, resulting in the business personnel having to manually compare, select and fill in between multiple fields, templates and resources, which is a tedious and inefficient process and is prone to omissions or inconsistencies. Moreover, the existing solution lacks semantic retrieval capability and automated reasoning capability based on existing knowledge, cannot structurally understand operation targets, and cannot generate executable activity configuration structures, resulting in the configuration process being completely dependent on manual logical judgment and item-by-item filling, which is difficult to meet the higher requirements for intelligentization, automation and scalability in digital operation. SUMMARY

[0003] In view of the above problems, the present application provides an activity configuration method, device, equipment, medium and program product.

[0004] According to a first aspect of the present application, an activity configuration method is provided, the method comprising: obtaining operation target information of a target activity; obtaining a pre-constructed multi-modal knowledge base, retrieving a target semantic context from a semantic graph of the multi-modal knowledge base based on the operation target information; performing semantic analysis and business reasoning on the target semantic context using a large language model to generate an initial activity configuration result corresponding to the operation target information; and performing dynamic compliance auditing on the initial activity configuration result to determine a final activity configuration result based on the auditing result.

[0005] According to an embodiment of the present application, the target semantic context is generated by retrieving from the semantic graph of the multi-modal knowledge base based on the operation target information, comprising: identifying business entities and constraint conditions semantically related to the operation target information from the semantic graph; taking the constraint conditions as path priority weights to perform constraint consistency retrieval on multi-hop association paths between the business entities to obtain a candidate path set; and generating the target semantic context based on an association path in the candidate path set that satisfies the constraint conditions and has the minimum path priority weight.

[0006] According to an embodiment of this application, the step of using a large language model to perform semantic parsing and business reasoning on the target semantic context includes: hierarchically deconstructing the target semantic context to generate a semantic parsing tree; determining the activity type, resource binding strategy, and field filling rules based on the semantic parsing tree to generate an initial activity configuration structure consistent with the business logic; and converting the initial activity configuration structure into an executable configuration data format to form the initial activity configuration result.

[0007] According to an embodiment of this application, the dynamic compliance review of the initial activity configuration result includes: locating the constraint set associated with the initial activity configuration result in the semantic graph based on the target semantic context; and performing a logical consistency check on the constraint set to determine the review result of the initial activity configuration result in terms of time constraint consistency, resource binding, and / or customer coverage conflict.

[0008] According to an embodiment of this application, the method further includes: performing dynamic compliance review on the initial activity configuration result through a multi-layer rule adaptation engine, wherein the multi-layer rule adaptation engine includes: a basic format rule layer for verifying the field length, color value format, and date format in the initial activity configuration result; a business logic rule layer for detecting business logic anomalies based on the semantic graph inference; and a resource usage limit layer for monitoring budget overruns and concurrent resource usage conflicts.

[0009] According to an embodiment of this application, determining the final activity configuration result based on the audit result includes: in response to the audit result indicating that there is a defect in the initial activity configuration result, identifying the target semantic node corresponding to the defect; generating multiple candidate field values ​​based on the contextual relationship of the target semantic node in the semantic graph; using the large language model to perform semantic consistency screening on the multiple candidate field values ​​to determine the completion result; and updating the initial activity configuration result based on the completion result to form the final activity configuration result.

[0010] According to an embodiment of this application, determining the final activity configuration result based on the audit result includes: in response to the audit result indicating that the initial activity configuration result has passed the audit, structurally encapsulating the initial activity configuration result to generate a target configuration file with a compliance identifier; and generating a version signature for the target configuration file and recording audit parameters to form the final activity configuration result.

[0011] According to an embodiment of this application, the method further includes: monitoring operational data during the operation process after the final activity configuration result is put into operation; comparing the operational data with the audit parameters to obtain operational deviation information; and in response to the operational deviation information exceeding the target threshold, associating and storing the operational data and the audit result in the multimodal knowledge base.

[0012] A second aspect of this application provides an activity configuration apparatus, comprising: a data acquisition module for acquiring operational target information of a target activity; a retrieval module for acquiring a pre-built multimodal knowledge base and retrieving a target semantic context from the semantic graph of the multimodal knowledge base based on the operational target information; a parsing and reasoning module for performing semantic parsing and business reasoning on the target semantic context using a large language model to generate an initial activity configuration result corresponding to the operational target information; and a compliance review module for performing dynamic compliance review on the initial activity configuration result and determining the final activity configuration result based on the review result.

[0013] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0014] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0015] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0016] According to embodiments of this application, by constructing a semantic graph in a multimodal knowledge base and performing semantic retrieval, the system can automatically extract entities, constraints, and logical relationships related to the target activity from complex business knowledge, improving the accuracy of understanding operational intentions. Simultaneously, by utilizing a large language model to perform semantic parsing and business reasoning on the target semantic context, structured and parameterized activity configuration results can be automatically generated, significantly reducing the workload and error rate of manual configuration. A dynamic compliance review mechanism enables multi-level verification from format validation to semantic logic consistency, promptly identifying implicit logical risks such as resource conflicts and time mismatches. Furthermore, the system automatically outputs the final activity configuration result based on the review results, achieving an integrated closed loop of configuration generation, logical validation, and compliance assurance, thereby improving the automation level, compliance reliability, and business response efficiency of operational activity configuration. Attached Figure Description

[0017] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0018] Figure 1 The illustrations depict application scenarios of the activity configuration method, apparatus, device, medium, and program product according to embodiments of this application.

[0019] Figure 2 A flowchart illustrating an activity configuration method according to an embodiment of this application is shown schematically;

[0020] Figure 3 The flowchart illustrating a method for retrieving and generating target semantic context in a semantic graph according to some exemplary embodiments of this application is shown in the schematic diagram.

[0021] Figure 4 The illustration shows a schematic diagram of a method for determining the final activity configuration result based on the audit results according to some exemplary embodiments of this application;

[0022] Figure 5 A schematic diagram illustrating the structure of an active configuration device according to an embodiment of this application is shown; and

[0023] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing an active configuration method according to an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0026] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0027] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0028] As digital transformation deepens in industries such as finance, internal business operations are becoming increasingly diversified, frequent, and complex. To achieve precise marketing, resource allocation, and campaign deployment, business personnel typically need to configure operational activities within the management system, including customer segment definition, resource allocation, campaign information, deployment time, and channel strategies. Current technologies generally employ manual configuration, requiring business personnel to manually input parameters, select modules, and bind templates in the operations management platform. For example, in a promotion or customer feedback campaign, operations personnel must manually set the customer segment, select available resources, define campaign priorities, and input display style parameters (such as color values, fonts, and layouts), with the system then performing compliance checks based on preset rules. These compliance checks primarily include static specifications such as field length limits, special character detection, and data format matching to prevent input errors or inconsistencies in basic data.

[0029] However, this management approach, which relies on manual configuration and static rules, has significant limitations. First, in terms of configuration efficiency, the large number of activity parameters and the need for manual entry make the process time-consuming and repetitive, easily leading to errors due to negligence or misfilling. For example, parameter items vary considerably across different activity templates, lacking unified intelligent guidance; personnel must frequently consult documents or historical templates to complete the configuration, significantly reducing operational response speed. Second, regarding risk detection capabilities, traditional static validation rules can only detect explicit errors, such as incorrect field formats or violations of special characters, but cannot identify implicit logical conflicts. For example, when the activity period does not match the resource availability period, or when the same customer group is repeatedly covered by multiple activities, the system cannot proactively detect the problem, potentially leading to resource waste and customer experience confusion. Furthermore, some business logic relies on cross-module data (such as constraints between activities and budgets, resource inventory, and channel status), while static rules cannot achieve dynamic cross-module validation, resulting in blind spots in risk detection.

[0030] Furthermore, existing system rules are typically embedded in the backend logic in the form of configuration files or database tables. When regulatory policies, business strategies, or resource status change, technical personnel often need to manually modify the rule scripts or redeploy the service for them to take effect. This approach not only leads to long rule update cycles but also causes system downtime for maintenance and business interruptions, severely impacting the efficiency of campaign launches and the company's compliance responsiveness. With the accelerating pace of changes in the regulatory environment and market, the maintenance costs and risks of traditional rule systems are constantly rising.

[0031] Furthermore, due to the lack of semantic connections and contextual understanding between rules, existing systems cannot perform reasoning and recommendations based on business semantics. For example, different types of activities (such as coupon distribution, points redemption, and new member acquisition) have different logical constraints, resource dependencies, and display templates. The current system cannot automatically identify the business context and provide appropriate parameter suggestions or template recommendations, resulting in operational configurations that heavily rely on human experience, leading to high learning costs and poor transferability. Moreover, existing systems struggle to accumulate knowledge of historical configuration data and validation results, lacking a data-driven self-learning mechanism and unable to optimize subsequent activity configuration processes through experience feedback.

[0032] Based on this, embodiments of this application provide an activity configuration method, the method comprising: obtaining operational target information of a target activity; obtaining a pre-built multimodal knowledge base, and retrieving and generating a target semantic context from the semantic graph of the multimodal knowledge base based on the operational target information; performing semantic parsing and business reasoning on the target semantic context using a large language model to generate an initial activity configuration result corresponding to the operational target information; and performing dynamic compliance review on the initial activity configuration result, and determining the final activity configuration result based on the review result. According to embodiments of this application, by constructing a semantic graph in a multimodal knowledge base and performing semantic retrieval, the system can automatically extract entities, constraints, and logical relationships related to the target activity from complex business knowledge, improving the accuracy of understanding operational intentions. Simultaneously, by utilizing a large language model to perform semantic parsing and business reasoning on the target semantic context, structured and parameterized activity configuration results can be automatically generated, significantly reducing the workload and error rate of manual configuration. A dynamic compliance review mechanism enables multi-level verification from format validation to semantic logic consistency, promptly identifying implicit logical risks such as resource conflicts and time mismatches. Furthermore, the system automatically outputs the final activity configuration result based on the review results, achieving an integrated closed loop of configuration generation, logical validation, and compliance assurance, thereby improving the automation level, compliance reliability, and business response efficiency of operational activity configuration.

[0033] It should be noted that the activity configuration methods, apparatus, devices, media, and program products defined in this application can be used in the fields of big data technology, artificial intelligence technology, and fintech, and can also be used in various other fields besides big data technology, artificial intelligence technology, and fintech. The application fields of the activity configuration methods, apparatus, devices, media, and program products provided in the embodiments of this application are not limited.

[0034] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0035] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided in this application all offer users corresponding operation entry points, allowing them to choose to agree to or reject the automated decision results; if the user chooses to reject, the process proceeds to expert decision-making. Here, "expert decision-making" refers to the decision-making activities of personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0036] Figure 1 The illustrations depict application scenarios of the activity configuration method, apparatus, device, medium, and program product according to embodiments of this application.

[0037] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0038] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0039] In the embodiments of this application, the first terminal device 101 can be an example of the first device, and the second terminal device 102 and / or the third terminal device 103 can be an example of at least one second device. The first device and the second device can communicate collaboratively through an internal client mechanism to implement the data distribution and rendering logic described in the activity configuration method.

[0040] In some embodiments, the first device and at least one second device may be different display modules, windows or screens on the same computing terminal (such as a host), or multiple physical devices that work together through a network, such as different client instances deployed on a desktop computer, tablet terminal or mobile device respectively.

[0041] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smart mobile terminals, tablet computers, laptop computers, and desktop computers.

[0042] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0043] It should be noted that the activity configuration method provided in this application embodiment can generally be executed by server 105. Correspondingly, the activity configuration device provided in this application embodiment can generally be located in server 105. The activity configuration method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the activity configuration device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0044] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0045] The following will be based on Figure 1 The described scene, through Figures 2 to 4The activity configuration method of the disclosed embodiments will be described in detail.

[0046] Figure 2 A flowchart illustrating an activity configuration method according to an embodiment of this application is shown schematically.

[0047] like Figure 2 As shown, the activity configuration method 200 of this embodiment includes operations S210 to S240.

[0048] In operation S210, obtain the operational target information of the target activity.

[0049] In the embodiments of this application, operational target information can originate from manually input natural language descriptions, operational instructions generated by system scheduling tasks, or activity demand data automatically transmitted by external business system interfaces. For example, in digital operation scenarios in the banking or retail industries, operations personnel can input natural language targets such as "increase the login frequency of high-net-worth customers" or "improve the page conversion rate of specific products" through the operations management platform. The system can identify the target customer groups, target indicators, and constraints, thereby generating structured operational target information.

[0050] In some embodiments, the process of acquiring operational target information can also be integrated with the enterprise's internal business data system, such as a marketing campaign management system or a resource scheduling system. When a user initiates an event planning task in an external system, the system can automatically receive event information, including event name, target user group, budget limit, execution period, and channel strategy, through interface calls, and input it as operational target information into subsequent processing stages. This enables automatic synchronization of operational target information, reduces repetitive manual data entry, and ensures that the target information remains consistent with the status in the business system.

[0051] In some embodiments, operational target information can also be automatically generated through historical data analysis and intelligent recommendation mechanisms. For example, the system can automatically identify potential optimization targets, such as "improving the secondary conversion rate of active users" or "optimizing the interaction rate of low-frequency customer groups," based on the operational performance of historical activities, customer behavior data, and changes in the market environment, using built-in analysis models, and convert these analytical conclusions into actionable operational target information for input into subsequent modules.

[0052] In some embodiments, operational objective information may also include multi-dimensional elements, such as business intent tags, customer segmentation conditions, a list of constrained resources, and risk control requirements. When receiving this information, the system can simultaneously identify the objective type, time window, key metrics (such as click-through rate and retention rate), and business constraints (such as budget limits and resource type availability).

[0053] In some embodiments, semantic understanding preprocessing can be introduced to improve the quality of operational target information acquisition. Specifically, natural language understanding models can be used to perform intent recognition, named entity recognition, and dependency analysis on the input text, transforming freely input natural language descriptions into a structured parameter set. For example, for the input "Promote high-end wealth management products during the Spring Festival", the system can automatically parse out "Time constraint: Spring Festival", "Target customer group: High-net-worth clients", "Target type: Promoting wealth management products", and "Key indicator: Increased click-through rate or purchase rate".

[0054] In some embodiments, within certain enterprise collaborative environments, the acquisition of operational target information can also be achieved through multi-role collaboration. For example, the marketing department defines marketing objectives, the data department provides customer segmentation results, and the resources department uploads information on available resource pools. The system aggregates inputs from different roles and performs semantic aggregation to form a unified target description.

[0055] In operation S220, a pre-built multimodal knowledge base is obtained, and a target semantic context is generated by retrieving it from the semantic graph of the multimodal knowledge base based on the operational target information.

[0056] In the embodiments of this application, the multimodal knowledge base can comprehensively store structured data (such as business indicator tables, rule lists, and resource lists) and unstructured data (such as historical activity texts, market reports, and operational cases), and establish semantic relationships between different entities through a semantic graph. For example, the semantic graph in the knowledge base can contain core entities such as "activity type—customer profile—resource package—display template—regulatory constraints" and their associated edges, used to express the contextual semantic dependencies between operational elements. After receiving operational target information, the system can extract keywords and intent entities through the natural language module, such as "promoting high-net-worth clients' financial products," and retrieve entity nodes related to "financial products," "high-net-worth clients," and "promotional activities" in the semantic graph, thereby generating a target semantic context corresponding to the target.

[0057] In some embodiments, the semantic graph in the knowledge base can adopt a multi-layered structure, including a business rules layer, a resource semantic layer, and a contextual relationship layer. When performing a retrieval, the system can automatically determine the retrieval scope based on the type of operational target information. For example, when the target information involves "activity cycle optimization," the system prioritizes querying the business rules layer and the contextual relationship layer to obtain the temporal relationship between time constraints, resource dependencies, and historical activities; when the target information involves "template recommendation," the system can prioritize retrieving the associated nodes of visual templates and interaction structures in the resource semantic layer.

[0058] In some embodiments, the multimodal knowledge base can also integrate data from multiple heterogeneous sources, such as an enterprise's internal customer relationship management system, advertising logs, compliance audit records, and external industry knowledge graphs. When retrieving data from the semantic graph of the multimodal knowledge base based on operational goal information, the system can dynamically select subgraph sets from different sources for fusion retrieval based on the domain tags in the operational goal information. For example, in a banking scenario, when the operational goal is "to increase the activity of newly opened customers," the system can combine customer transaction records, marketing script templates, and regulatory policy documents for joint semantic retrieval, thereby generating a target semantic context that includes customer segmentation logic, compliance elements, and resource strategies.

[0059] In some embodiments, the system can also utilize constraint-driven path retrieval algorithms to perform multi-hop queries on the semantic graph. Specifically, constraints extracted from operational target information (such as budget limits, time windows, and customer group restrictions) can be used as path priority weights to perform constraint consistency retrieval on the relationship paths between multiple business entities in the semantic graph, thereby determining the optimal association path that satisfies all constraints.

[0060] In some embodiments, the generation of the target semantic context can also incorporate temporal evolution information. The system maintains temporal attributes for each entity node in the semantic graph, such as activity validity period, resource availability time, and rule effective date. Upon receiving operational target information, the system prioritizes filtering nodes and relationships that match the current business time window, thereby generating a dynamic semantic context that conforms to time constraints. For example, if the operational target is "to increase brand exposure during the 2025 Spring Festival," the system will automatically retrieve the resource combinations with the highest exposure rates and related template nodes from historical Spring Festival activities, thus generating a time-sensitive context result.

[0061] In some embodiments, to improve retrieval accuracy and robustness, the system can introduce an embedded representation learning mechanism into the semantic graph. Specifically, each business entity and relation edge can be represented by semantic features through vectorized embedding. When performing a retrieval, the system first encodes the operational target information into semantic vectors, and then determines the most relevant set of entities and path candidates through similarity calculation, thereby generating the target semantic context. For example, for the input target "improving the conversion rate of financial products for elderly customers", the system can identify group labels (such as "retired customers") that are semantically close to "elderly customers" through semantic embedding calculation, and find related successful activity case nodes in the knowledge graph to form rich contextual associations.

[0062] In some embodiments, the target semantic context generated by the system may include not only business entities and relational structures, but also semantic confidence and weight distribution information. Each relational edge may carry a context matching degree or semantic relevance score for use by the subsequent large language model when performing semantic parsing and business reasoning. For example, after generating the target semantic context, the system may attach parameters such as "customer matching confidence 0.92, budget constraint confidence 0.85, template relevance 0.80" to guide the large model to prioritize high-confidence business elements when generating activity configurations, thereby improving overall reasoning accuracy.

[0063] In operation S230, a large language model is used to perform semantic parsing and business reasoning on the target semantic context to generate an initial activity configuration result corresponding to the operational target information.

[0064] Specifically, the target semantic context can be input into a large language model. The model performs hierarchical parsing of the business entities, constraints, semantic paths, and their relationships contained therein, identifying the core business intent and logical dependencies involved in the target activity. For example, when the target semantic context involves "increasing the login frequency of high-net-worth clients," the large language model can automatically identify semantic elements such as "customer type = high-net-worth," "target indicator = login frequency," and "optional methods = points activities, push notifications, exclusive benefits," and infer the corresponding activity structure template. Subsequently, the large language model can generate initial activity configuration results based on these elements, including structured content such as activity name, execution period, target customer group, available resource binding relationships, and display strategy.

[0065] In some embodiments, the semantic parsing process of the large language model can adopt a hierarchical reasoning structure: In the first layer, semantic role labeling and intent extraction are performed, transforming natural language semantics into a semantic structure of "activity target—customer group—resources—constraints"; in the second layer, business logic reasoning is performed, combining entity relationships in the semantic graph to complete and confirm the constraints between different business elements; in the third layer, configuration generation is performed, mapping the results after logical reasoning to configuration fields that the system can recognize. For example, given the input semantic context "Promoting financial product A, targeting member customers, with an activity period of one month," the system can generate an initial activity configuration result, including the fields: Activity Type = Promotion, Target Product = Financial Product A, Target Customer Group = Member Customers, Time Period = 30 days, Budget Constraint = Default Upper Limit, and Display Template = Financial Recommendation Page.

[0066] In some embodiments, the large language model can perform inference tasks based on prompt templates. The system constructs multi-dimensional prompt word templates at the input layer, including descriptions of operational intent, constraint parameters, and semantic context examples. The large language model then performs context-aware generation based on these templates. For example, when the system detects that the target semantic context belongs to the "activity template recommendation" type, the model will invoke the "template inference prompt template," focusing on the matching degree between page structure, displayed elements, and customer preferences, thereby generating an initial page configuration result with visual style and interaction logic. Conversely, when the semantic context belongs to the "resource scheduling optimization" type, the model switches to the "resource association inference template" to prioritize inferring resource weights, inventory constraints, and time allocation.

[0067] In some embodiments, the semantic parsing and business reasoning processes can also incorporate knowledge enhancement mechanisms to improve the model's business consistency and compliance. Specifically, before executing large language model reasoning, the system associates and matches key entities and constraints in the target semantic context with business rule nodes in the multimodal knowledge base, enabling the large language model to reference structured business knowledge during reasoning. For example, when the initial configuration results generated by the model involve key parameters such as budget allocation and discount amounts, the system can automatically inject regulatory policies or historical activity constraint information based on the knowledge base to ensure that the generated results comply with business specifications and compliance boundaries. This design enables the large language model to not only possess natural language understanding capabilities but also domain-knowledge-based decision-making rationality.

[0068] In some embodiments, to improve the accuracy and stability of the generated results, the system can employ a multi-model collaborative mechanism during the semantic parsing and business reasoning stages. Specifically, the system can set up a two-level structure of a "main inference model" and a "verification model": the main inference model is responsible for generating the initial activity configuration results, while the verification model performs semantic consistency and constraint matching checks on the generated results. When the verification model detects a deviation between the generated results and the target semantic context (such as missing fields, logical conflicts, etc.), the system can trigger an automatic correction or regeneration process. This collaborative reasoning mechanism effectively reduces the "illusion" risk when a large language model directly generates configurations, improving the interpretability and business reliability of the results.

[0069] In some embodiments, the generation of the initial activity configuration result can also incorporate a semantic scoring mechanism. After generating multiple candidate configuration results, the system calculates a comprehensive score based on semantic consistency score, business constraint satisfaction, and context coverage, and selects the scheme with the highest score as the final initial activity configuration result. For example, in cases where the target semantic context is complex (such as involving cross-channel marketing activities), the system may generate multiple different combinations of schemes. The model automatically evaluates the matching degree between each scheme and the target semantic context through semantic scoring, thereby selecting the optimal configuration scheme.

[0070] In some embodiments, the large language model can also perform semantically constrained backward reasoning tasks. When the system detects incomplete or ambiguous descriptions of operational target information, the model can perform supplementary reasoning on the missing content based on the contextual structure of the semantic graph.

[0071] In some embodiments, to adapt to the application needs of multiple industries and languages, the large language model can be combined with a domain fine-tuning mechanism. Specifically, the system introduces domain corpora and rule knowledge into the training phase for different business domains, enabling the model to have multi-scenario reasoning capabilities.

[0072] In operation S240, the initial activity configuration result is dynamically reviewed for compliance, and the final activity configuration result is determined based on the review result.

[0073] In the embodiments of this application, dynamic compliance auditing can be performed collaboratively by semantic graph reasoning and a rule engine: The system first parses key fields in the initial activity configuration results, such as activity period, target customer group, budget amount, resource binding, and display template, and locates the corresponding constraint nodes and their associated rules based on the semantic graph. Subsequently, the system uses the reasoning engine to perform consistency verification on the relationships between these fields, such as determining whether the activity period exceeds the resource validity period, whether the budget exceeds the upper limit, and whether the customer group is repeatedly covered. When potential logical conflicts or constraint mismatches are detected, the system will mark the corresponding risk items and generate an audit report, thereby achieving dynamic verification and preventive control of the configuration results.

[0074] In some embodiments, dynamic compliance auditing can be processed using a multi-layered rule architecture. The system can be configured with a basic format validation layer, a business logic rule layer, and a resource constraint layer. The basic format layer checks the format validity of fields in the activity configuration results, such as whether the date format is correct and whether the color code conforms to the standard. The business logic rule layer focuses on validating the rationality of the configuration logic, such as whether the time period of the same activity conflicts with other activities and whether the marketing channels match the target customer profile. The resource constraint layer monitors resource usage, inventory quotas, and budget allocation to prevent over-limit or resource hogging. Data and rules between each layer support dynamic loading and hot updates. When the system detects changes in business strategies or regulatory standards, it can update the relevant rules immediately without requiring manual downtime maintenance.

[0075] In some embodiments, to achieve a higher level of intelligent review, the system can introduce a dynamic logic detection mechanism driven by semantic reasoning. Specifically, an activity logic constraint graph can be constructed based on the relationship network in the semantic graph, and reasoning algorithms can be used to dynamically traverse and verify the constraint chains in the activity configuration. For example, when the system detects that "activity A depends on resource X, and the availability period of resource X depends on the status of supplier Y," it will further analyze whether the service timeliness and contractual constraints of supplier Y affect the execution of activity A, and then determine whether there is an indirect logical conflict.

[0076] In some embodiments, the dynamic compliance audit process can also incorporate machine learning models to perform pattern recognition and risk prediction on historical activity data. During the audit, in addition to the rules engine, the system also invokes a risk prediction model to extract features and perform similarity analysis on key parameters of the initial activity configuration results, determining whether they are similar to previous activity configuration patterns that had a risk of violation or failure.

[0077] In some embodiments, dynamic compliance audits can also collaborate with manual review processes. After completing the initial automated audit, the system generates audit tasks at different levels based on the detected risk level. For low-risk items, the system can automatically correct and approve them; for medium-risk items, the system can issue a manual review prompt, requiring operations personnel to confirm; and for high-risk items, approval and confirmation from management or the compliance department are required before execution can continue. For example, when the system detects that "the resource validity period is earlier than the activity end date," it can automatically mark the configuration as high-risk and generate an interpretability report explaining the source of the problem, the affected fields, and the possible business consequences, thereby helping auditors make quick decisions.

[0078] In some embodiments, while performing dynamic compliance audits, the system can also record audit parameters, detection processes, and result data, and generate corresponding version identifiers for subsequent traceability and knowledge accumulation. For example, each audit generates an audit log containing an activity identifier, rule version number, audit timestamp, and a summary of the detection results, and stores this log in association with activity nodes in the knowledge base. When subsequent business rules or regulatory requirements are updated, the system can quickly locate the affected activity configuration results based on historical versions, enabling batch review and automatic adjustment.

[0079] In some embodiments, dynamic compliance auditing can also be extended to cross-system collaborative verification. When the activity configuration results involve external resource systems or multi-channel delivery platforms, the system can call the status information and resource list of the external system in real time through the Application Programming Interface (API) and perform comparisons at the semantic layer. For example, the system can verify whether "the ad placements in the configuration actually exist and are available", "whether the authorization status of the channel API is valid", and "whether the budget synchronization of the external platform is completed". When an anomaly is detected, the system will automatically roll back the configuration or trigger an anomaly alert, thereby achieving dynamic compliance control across the entire chain.

[0080] According to embodiments of this application, operational target information for target activities can be automatically obtained from natural language input, system task instructions, or multi-source business data, achieving high-precision identification of operational intentions and multi-channel access, significantly improving the response speed and input flexibility of activity configuration; through semantic graph retrieval of a multimodal knowledge base, complex business rules, resource relationships, and historical cases are transformed into computable target semantic context, thereby enabling the system to possess business semantic understanding and contextual reasoning capabilities; by utilizing a large language model to perform semantic parsing and business reasoning on the target semantic context, structured and parameterized initial activity configuration results are automatically generated, reducing the complexity of manual configuration and avoiding configuration deviations caused by experience differences; by introducing a dynamic compliance review mechanism, combined with semantic reasoning and a multi-layer rule engine, dynamic consistency verification is performed on multi-dimensional logic such as time constraints, resource binding, budget limits, and customer coverage, and rule hot updates and human-machine collaborative review are supported, thereby ensuring the comprehensive reliability of configuration results in terms of logical rationality, business compliance, and resource executability.

[0081] The activity configuration method of this application will be specifically described below by way of preferred embodiments.

[0082] In the embodiments of this application, the following method is provided in order to accurately identify business semantic elements related to operational objectives in a multimodal knowledge base and generate structured contextual results.

[0083] Figure 3 The flowchart illustrating a method for retrieving and generating a target semantic context in a semantic graph according to some exemplary embodiments of this application is shown schematically.

[0084] like Figure 3 As shown, the method for retrieving and generating target semantic context in a semantic graph includes operations S310 to S330.

[0085] In operation S310, business entities and constraints related to the semantics of the operational target information are identified from the semantic graph.

[0086] Specifically, the system can first perform semantic parsing of operational target information based on a natural language understanding model to extract key business intentions, entity categories, and constraint parameters. For example, when the operational target information is "to increase the login frequency of high-net-worth customers," the system can identify semantically relevant nodes such as "customer group (high-net-worth customers)," "business indicators (login frequency)," "time constraints (periodic monitoring)," and "resource types (points, coupons)" from the semantic graph. The nodes and relationships in the semantic graph can be represented by a knowledge triple (entity-relationship-attribute) structure, thereby enabling the system to accurately locate the set of business elements related to the target semantics at the semantic level.

[0087] In operation S320, the constraints are used as path priority weights to perform constraint consistency retrieval on multi-hop association paths between the business entities, resulting in a candidate path set.

[0088] Specifically, different constraints (such as time, budget, resource dependence, regulatory requirements, etc.) correspond to different priority weights, and the system prioritizes paths that satisfy the consistency of constraints during the search process. For example, when the constraints are "budget not exceeding 5,000 yuan and activity period less than 30 days", the system will prioritize path branches that satisfy both budget and time constraints during path retrieval, thereby forming a set of candidate paths within a reasonable range of constraints.

[0089] In operation S330, the target semantic context is generated based on the associated path in the candidate path set that satisfies the constraints and has the smallest path priority weight.

[0090] Specifically, the semantic distance and constraint matching degree of each candidate path can be calculated using a path cost function, and the path with the lowest priority weight is selected to generate the target semantic context. The path with the lowest priority weight not only reflects the entity chain that is semantically closest to the operational target information, but also maintains logical consistency at the business constraint level. For example, in a bank marketing scenario, the system may generate a path containing "Activity type = wealth management promotion → Target customer group = high-net-worth customers → Resource binding = VIP coupons → Execution time = weekly → Compliance rules = marketing frequency limit". Finally, this path can be structured into a target semantic context, along with a semantic confidence score, to guide the subsequent parameter filling and logical reasoning of the large language model during the configuration generation phase.

[0091] In some embodiments, the system can also introduce a time evolution and context fusion mechanism during the semantic graph retrieval process to generate a more dynamic semantic context. Specifically, when constructing the semantic graph, the system can add timestamps and semantic weight information to entities and relationships. When the system performs constraint consistency retrieval, it can dynamically filter candidate paths by combining the current business period, seasonal characteristics, or regulatory update status. For example, when the target activity is "increasing brand exposure during the 2025 Spring Festival," it can prioritize retrieving high-exposure path nodes from historical Spring Festival activities and generate target semantic contexts with time weights, making the results more timely and relevant to the context.

[0092] Through the above steps, the system can generate context deeply associated with operational objectives based on the semantic graph of a multimodal knowledge base, providing an interpretable semantic foundation for intelligent configuration. Compared with traditional keyword-based or static rule-based retrieval methods, the embodiments of this application can not only understand natural language intent but also perform multi-hop semantic reasoning under constraints, thereby automatically constructing a semantic context network that satisfies business logic.

[0093] In this embodiment, a method for semantic parsing and business reasoning of the target semantic context can also be utilized using a large language model. By introducing a hierarchical semantic parsing mechanism and a business logic reasoning strategy, the system can transform the context information obtained from semantic graph retrieval into structured activity configuration results, thereby achieving automatic generation of operational activities and ensuring logical consistency.

[0094] In this embodiment, the target semantic context can first be hierarchically deconstructed to generate a semantic parse tree. Specifically, the system takes the target semantic context as input and uses a large language model to extract business entities, relationship types, and attribute tags hierarchically. Each layer of nodes represents a semantic hierarchy, such as the activity target layer, customer group layer, resource layer, and rule layer. By performing semantic dependency analysis on the entities and relationships in the semantic graph, a hierarchical semantic parse tree can be constructed.

[0095] Subsequently, business logic reasoning can be performed based on the semantic parsing tree to determine the activity type, resource binding strategy, and field filling rules. In this stage, the large language model combines domain knowledge with the hierarchical relationship of the semantic context to determine the logical dependencies between different business entities. For example, the system can determine the dependency relationship between "activity type = promotion" and "resource type = coupon" based on the parsing tree, thereby inferring that the resource binding strategy is "coupon bound to financial product A"; further, combining "customer group = high-net-worth customers," the system infers that the field filling rules include "display template selection = high-end financial style," "display priority = high," and "budget cap = 5000 yuan."

[0096] Furthermore, the system transforms the initial activity configuration structure into an executable configuration data format to form the initial activity configuration result. Specifically, the system converts the generated structured configuration content into a standardized data object format based on a preset data template or configuration specification. Each configuration item includes a field identifier, parameter value, logical relationship, and optional constraints, which can then be directly called and executed by the subsequent activity management system.

[0097] In some embodiments, to improve generation accuracy and interpretability, the large language model can incorporate prompt templates for contextual constraints during the semantic parsing and business reasoning stages. The system automatically selects the appropriate template based on different task types, such as an "activity configuration generation template," a "resource strategy reasoning template," or a "field rule matching template," to ensure that the output meets the logical requirements of specific business scenarios. When the model generates configuration results, it can also include semantic confidence scores and rule matching metrics for subsequent compliance review and manual verification.

[0098] In some embodiments, after completing semantic parsing and business reasoning, the system can also record intermediate results (such as parse trees, reasoning paths, and logical weights) during the reasoning process in the log module for interpretability tracking and model evaluation. For example, when there are conflicts in the generated results or when review is required, the system can trace back the semantic parse tree nodes and their reasoning basis to verify whether the reasoning chain meets expectations.

[0099] By introducing a semantic parsing and business reasoning mechanism based on a large language model, the system achieves intelligent conversion from natural language semantics to structured activity configuration. Through hierarchical deconstruction of the semantic parsing tree, the system can accurately depict the logical dependencies between complex semantics; through the automatic generation of business logic reasoning and field filling rules, the system achieves high-dimensional semantic decision-making and business structure mapping; and through standardized output of configuration results, the system ensures the executability and reusability of the generated content.

[0100] In the embodiments of this application, constraint sets can be located in the semantic graph and logical consistency checks can be performed, thereby realizing dynamic verification of configuration results and semantic-level conflict detection.

[0101] In this embodiment, the system first locates the set of constraints associated with the initial activity configuration result in the semantic graph based on the target semantic context. Specifically, the system extracts key fields from the configuration result, such as activity period, budget amount, target customer group, and resource binding relationship, and uses these fields as query entry points to match the corresponding constraint entity nodes and their relationships in the semantic graph. For example, when the configuration result contains "activity period = 30 days", "resource type = coupon", and "customer group type = high-net-worth customers", the system can automatically locate the corresponding constraint nodes in the semantic graph, such as "resource validity period constraint", "customer group coverage rule", and "budget allocation restriction", and generate the set of constraints corresponding to the activity configuration.

[0102] Furthermore, logical consistency checks can be performed on the constraint set to determine the compliance status of the initial activity configuration results across key dimensions such as time constraint consistency, resource binding conflicts, and duplicate customer coverage. Logical consistency checks are implemented through an inference engine, which uses constraint rules in the semantic graph to perform semantic derivation and verification of the configuration results. For example, the system can detect time consistency between the "activity period" and the "resource validity period": when the resource validity period is earlier than the activity end date, the system determines there is a risk of time inconsistency; the system can also detect whether resource binding logic conflicts, such as whether the same coupon is concurrently bound to or reused by multiple activities.

[0103] In some embodiments, the logical consistency verification process can be combined with a dynamic constraint update mechanism. When performing verification, the system can automatically load the latest versions of business rules, regulatory requirements, or resource status information to ensure that the review process remains synchronized with the current business situation. For example, when regulatory authorities update the rules on the upper limit of activity promotion frequency, the system can adjust the corresponding constraint nodes and their logical relationships in the semantic graph in real time, thereby automatically applying the new rules in the next review without manual intervention.

[0104] In some embodiments, the system can also generate an interpretable audit chain during the logical consistency verification process. The audit chain records the reasoning path, matching rules, and judgment results for each constraint verification, supporting auditors to perform manual review or reverse tracing when necessary. For example, when the system detects a "resource binding conflict," the audit chain can display the specific conflict path, such as "Resource R1 is bound to both Activity A and Activity B simultaneously → Conflict dimension: resource uniqueness → Suggested solution: automatically release the binding of Activity B."

[0105] In another embodiment, to improve audit efficiency and intelligence, the system can also combine historical audit data with machine learning models to predict potential risks. When performing logical consistency checks, the system can perform similarity analysis between the current configuration results and historical risk samples to identify configuration combinations with high-risk characteristics in advance. For example, if historical data indicates that a combination of long-cycle activities and high-budget resource allocations is prone to compliance conflicts, the system will automatically mark a high-risk warning when it detects a similar structure.

[0106] This application implements a dynamic compliance review mechanism based on semantic graphs, which can not only detect explicit format errors, but also identify deep business logic conflicts and cross-module constraint issues, thereby effectively improving the compliance intelligence level of activity configuration results, reducing manual review costs, and building a sustainable compliance assurance system.

[0107] In the embodiments of this application, the configuration results can also be verified in multiple dimensions through a hierarchical rule system to ensure that they meet system compliance requirements in terms of data format, business logic, and resource constraints.

[0108] The multi-layer rule adaptation engine can include a basic format rule layer, a business logic rule layer, and a resource consumption restriction layer. The basic format rule layer verifies the basic data compliance of the initial activity configuration results, ensuring that the input data meets the system's processing requirements. At this layer, the system performs validity checks on field lengths, character encodings, color value formats, date formats, etc., in the configuration results. For example, it checks whether the activity name field length exceeds the database definition limit, whether the color value parameter conforms to the standard format, and whether the date field follows the timestamp specification. When non-compliant data is found, the system can automatically trigger a correction mechanism or generate a format error message to avoid subsequent configuration failures due to abnormal input formats.

[0109] The business logic rules layer is used to detect business logic anomalies based on semantic graph reasoning. At this layer, the system utilizes knowledge triples and logical relationships within the semantic graph to verify the logical consistency of initial activity configuration results. For example, when the system detects that the "activity period" exceeds the "resource validity period" or that the "budget allocation" does not match the "channel strategy," it determines it as a logical conflict. Similarly, if the same customer group is simultaneously reached by multiple activities or repeatedly bound to the same type of resources, the system can automatically detect this conflict through contextual reasoning using the semantic graph.

[0110] The resource usage restriction layer is used to monitor budget overruns and concurrent resource conflicts. At this layer, the system analyzes resource allocation information in activity configurations and the global resource pool status in real time, dynamically monitoring budget limits, inventory capacity, distribution channels, and server load. When it detects that budget allocation exceeds the approval threshold or that a resource is concurrently occupied by multiple activities, the system can immediately generate a risk warning and block the submission operation. For example, when two activities simultaneously call the same coupon batch or advertising placement, the system can automatically identify and lock the conflict node through the resource usage graph, thereby preventing duplicate resource allocation from causing execution anomalies. This layer also supports real-time data synchronization with external systems, such as budget management systems or inventory control systems, to achieve cross-module resource consistency verification.

[0111] In some embodiments, the three layers of the multi-layer rule adaptation engine can work in parallel or be executed sequentially according to the priority of the audit tasks. When the system starts an audit task, it can automatically allocate audit paths based on the characteristics of the configuration results and the business type. For example, activities with correct format but complex logic will be prioritized for the business logic rule layer; while activities involving large-scale resource scheduling will directly trigger concurrent verification at the resource consumption limit layer. This layered parallel mechanism can significantly improve audit efficiency, reduce redundant calculations, and increase overall system throughput. During execution, intermediate results can be shared between layers through a message bus to achieve collaborative decision-making and complex risk identification. For example, error correction results from the basic format layer will be directly passed to the logic layer so that the latter can perform inference verification based on the latest data.

[0112] The multi-layer rule adaptation engine constructed through the embodiments of this application realizes full-link compliance control from low-level format specifications to high-level semantic logic and system resource constraints.

[0113] Figure 4 The illustration shows a schematic diagram of a method for determining the final activity configuration result based on the audit results according to some exemplary embodiments of this application.

[0114] like Figure 4 As shown, the method for determining the final activity configuration result based on the audit result may include operations S410 to S440.

[0115] In operation S410, in response to the audit result indicating that there is a defect in the initial activity configuration result, the target semantic node corresponding to the defect is identified.

[0116] Specifically, during the dynamic compliance audit phase, the system can detect issues such as missing fields, incomplete constraint references, and logical inconsistencies. When a defect is detected, the system will locate the corresponding configuration field and its associated logical path based on the error identifier in the audit report.

[0117] In operation S420, multiple candidate field values ​​are generated based on the contextual relationship of the target semantic node in the semantic graph.

[0118] The system extracts potential sources of semantic completion by retrieving the associated paths, parent nodes, and sibling nodes of the target semantic node. For example, when a missing budget field is detected, the system can retrieve the historical activity budget distribution, budget ranges for similar activity types, and the system's preset budget cap strategy from the semantic graph. When a time conflict is detected, the system can generate multiple candidate time ranges from historical activity cycles, resource validity periods, and recommendation cycle templates. While generating candidate field values, the system can also calculate the relevance weight of each candidate to the original semantic context to guide the subsequent filtering process.

[0119] In operation S430, the large language model is used to perform semantic consistency filtering on the multiple candidate field values ​​to determine the completion result.

[0120] Specifically, the large language model can comprehensively consider the contextual consistency, constraint satisfaction, and logical rationality of candidate field values ​​with the target semantic node, calculate semantic matching scores, and select the optimal candidate. For example, when multiple budget candidate values ​​(such as "3000 yuan", "5000 yuan", "unlimited") may all satisfy the constraints, the large language model can select the one that best matches the semantics as the completion result based on activity type, resource level, and historical strategies.

[0121] In operation S440, the initial activity configuration result is updated based on the completion result to form the final activity configuration result.

[0122] The update process can include three steps: field completion, parameter replacement, and logical dependency adjustment. The system first inserts the selected completion results into the corresponding configuration fields. If the field has logical dependencies on other configuration items (such as budget and resource allocation, cycle and channel strategy), the system can automatically trigger dependency updates to ensure the consistency of the overall configuration logic. After the update is complete, the system generates a new configuration version and re-executes a quick review on the updated part to verify the effectiveness and consistency of the fix. If all constraints pass verification, the updated configuration result is marked as "compliant and passed" and can directly proceed to the deployment or execution phase.

[0123] In some embodiments, to enhance the accuracy and stability of automatic repair, the system can incorporate a multi-round screening mechanism based on confidence levels. In the first round of screening, the large language model generates multiple candidate completion schemes and calculates confidence scores. In the second round, the system cross-validates the candidate results by combining semantic graph inference results with business rule verification, thereby eliminating schemes that do not meet constraints or have logical anomalies. For example, in a scenario where resource binding fields are missing, the system may generate multiple candidate resource packages, but only resources that match the current activity type and are not occupied by other activities will be retained.

[0124] In some embodiments, the system can also record intermediate data during the completion process to form a traceable repair path. Each repair operation records the defect type, candidate generation path, filtering weight, and updated fields, forming a structured repair log. When a similar problem occurs again in subsequent configurations, the system can refer to this log to directly call historical repair solutions, achieving self-learning and optimization. For example, when the system detects the "budget missing" problem multiple times, it can automatically generate a recommended completion value range based on historical repair records, thereby further improving repair speed and accuracy.

[0125] According to the embodiments of this application, by combining the contextual relationship modeling of semantic graphs with the semantic filtering capabilities of large language models, automatic identification, candidate generation, and intelligent completion of defective initial activity configuration results are achieved, thereby effectively improving the adaptive capability, stability, and intelligent closed-loop level of the activity configuration system, and enhancing the overall operational automation and compliance reliability.

[0126] In the embodiments of this application, when the audit result indicates that the initial activity configuration result has passed the dynamic compliance audit, the process of determining the final activity configuration result based on the audit result can be executed. Specifically, the system first performs structured encapsulation on the audited initial activity configuration result, organizing the configuration content hierarchically and mapping fields according to a preset data template to form a standardized configuration file that can be directly called. This configuration file includes basic activity information, resource binding relationships, parameterized configuration fields, and logical dependency structures, and is accompanied by a unified metadata description to ensure consistent parsing and portability in different business systems or execution environments. At the same time, the system generates a compliance identifier for the configuration file during the encapsulation process to indicate that it has passed the full verification of the rule engine and semantic graph, ensuring that the configuration data is under control in terms of logic, format, and compliance.

[0127] After completing the structured encapsulation, the system further generates a version signature for the target configuration file and records audit parameters to form the final activity configuration result. The version signature can be generated using a hash digest or digital signature algorithm and is used to uniquely identify the configuration file version and its content integrity; the audit parameters include audit time, rule version, risk score, and execution node information, which are used to support subsequent traceability and version comparison. When the system detects configuration adjustments or rule changes in the future, it can quickly locate the differences and perform automatic review based on the signature and parameter records.

[0128] Through the above encapsulation and signature mechanisms, the system achieves standardized output and traceable management of configuration results, ensuring the consistency, security, compliance and verifiability of configuration content throughout its lifecycle, thereby forming a final activity configuration result that can be directly deployed and executed.

[0129] In the embodiments of this application, after the final activity configuration results are put into operation, the system enters the operation monitoring and feedback phase to achieve integrated closed-loop management of activity execution effects and review results. During operation, the system continuously monitors key operational data, including activity execution status, resource consumption, user response indicators, and system performance parameters. This operational data can be collected in real time through business logs, event tracking, data tracking analysis, or external interfaces, and is categorized and aggregated according to activity identifiers.

[0130] After collecting operational data, the system compares the data with previously recorded audit parameters to obtain operational deviation information. Audit parameters may include configuration version, constraints, risk scores, budget limits, and resource allocation strategies. The system analyzes the differences between the operational results and the audit baseline through a deviation calculation module. For example, it determines whether budget utilization exceeds a set threshold, whether resource consumption is completed ahead of schedule, whether the activity conversion rate is lower than expected, or whether the customer coverage deviates from the target customer group. Through semantic layer matching and statistical deviation calculation, the system can quantify the degree of deviation between operation and configuration, forming interpretable operational deviation information for dynamically evaluating configuration execution quality.

[0131] In another embodiment, when operational deviation information exceeds a target threshold, the system associates and stores the operational data with the corresponding review results and writes it into a multimodal knowledge base. This association storage process not only preserves the correspondence between operational performance and configuration content but also records the semantic context and constraints that led to the deviation, thereby enabling the knowledge base to dynamically accumulate operational experience. For example, when the system detects "abnormal use of coupon resources in high-concurrency scenarios" or "excessive budget allocation leading to a decrease in profitability," it binds such operational deviations with relevant review rules, semantic nodes, and configuration versions into the database for subsequent use in large language model inference and rule optimization.

[0132] Through this closed-loop feedback mechanism, the system achieves full-cycle self-learning capability from configuration review to operation monitoring and knowledge feedback, thereby improving the accuracy and adaptability of the model in subsequent activity generation and risk prediction.

[0133] Corresponding to the above-described activity configuration method, embodiments of this application also provide an activity configuration device.

[0134] Figure 5 A schematic block diagram of an active configuration device according to an embodiment of this application is shown.

[0135] like Figure 5 As shown, the activity configuration device 500 in this embodiment includes a data acquisition module 510, a retrieval module 520, a parsing and reasoning module 530, and a compliance audit module 540.

[0136] The data acquisition module 510 can be used to acquire operational target information for the target activity. In one embodiment, the data acquisition module 510 can be used to perform the operation S210 described above, which will not be repeated here.

[0137] The retrieval module 520 can be used to acquire a pre-built multimodal knowledge base and retrieve target semantic context from the semantic graph of the multimodal knowledge base based on the operational target information. In one embodiment, the retrieval module 520 can be used to perform the operation S220 described above, which will not be repeated here.

[0138] The parsing and reasoning module 530 can be used to perform semantic parsing and business reasoning on the target semantic context using a large language model, and generate an initial activity configuration result corresponding to the operational target information. In one embodiment, the parsing and reasoning module 530 can be used to execute the operation S230 described above, which will not be repeated here.

[0139] The compliance audit module 540 can be used to perform dynamic compliance audits on the initial activity configuration results and determine the final activity configuration results based on the audit results. In one embodiment, the compliance audit module 540 can be used to perform the operation S240 described above, which will not be repeated here.

[0140] According to an embodiment of this application, the retrieval module 520 can also be used to identify business entities and constraints related to the semantics of the operational target information from the semantic graph; use the constraints as path priority weights to perform constraint consistency retrieval on multi-hop association paths between the business entities to obtain a candidate path set; and generate the target semantic context based on the association path in the candidate path set that satisfies the constraints and has the smallest path priority weight.

[0141] According to an embodiment of this application, the parsing and reasoning module 530 can also be used to hierarchically deconstruct the target semantic context to generate a semantic parsing tree; determine the activity type, resource binding strategy and field filling rules based on the semantic parsing tree to generate an initial activity configuration structure consistent with the business logic; and convert the initial activity configuration structure into an executable configuration data format to form the initial activity configuration result.

[0142] According to an embodiment of this application, the compliance audit module 540 can also be used to locate the set of constraints associated with the initial activity configuration result in the semantic graph based on the target semantic context; and to perform logical consistency verification on the set of constraints to determine the audit result of the initial activity configuration result in terms of time constraint consistency, resource binding and / or customer coverage conflict.

[0143] According to an embodiment of this application, the compliance audit module 540 can also be used in the basic format rules layer to verify the field length, color value format and date format in the initial activity configuration result; the business logic rules layer to detect business logic anomalies based on the semantic graph reasoning; and the resource usage limit layer to monitor budget overruns and concurrent resource usage conflicts.

[0144] According to an embodiment of this application, the compliance audit module 540 can also be used to respond to the audit result indicating that there is a defect in the initial activity configuration result, identify the target semantic node corresponding to the defect; generate multiple candidate field values ​​based on the contextual relationship of the target semantic node in the semantic graph; use the large language model to perform semantic consistency screening on the multiple candidate field values ​​to determine the completion result; and update the initial activity configuration result based on the completion result to form the final activity configuration result.

[0145] According to an embodiment of this application, the compliance audit module 540 can also be used to respond to the audit result indicating that the initial activity configuration result has passed the audit, to perform structured encapsulation of the initial activity configuration result, to generate a target configuration file with a compliance identifier; and to generate a version signature for the target configuration file and record audit parameters to form the final activity configuration result.

[0146] According to an embodiment of this application, the compliance audit module 540 can also be used to monitor the operation data during the operation process after the final activity configuration result is put into operation; compare the operation data with the audit parameters to obtain operation deviation information; and in response to the operation deviation information exceeding the target threshold, associate and store the operation data and the audit result in the multimodal knowledge base.

[0147] According to embodiments of this application, any multiple modules among the data acquisition module 510, retrieval module 520, parsing and reasoning module 530, and compliance audit module 540 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the data acquisition module 510, retrieval module 520, parsing and reasoning module 530, and compliance audit module 540 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the data acquisition module 510, retrieval module 520, parsing and reasoning module 530, and compliance audit module 540 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0148] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing an active configuration method according to an embodiment of this application.

[0149] like Figure 6 As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0150] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0151] According to embodiments of this application, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0152] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0153] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0154] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the activity configuration method provided in the embodiments of this application.

[0155] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0156] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0157] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0158] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0159] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0160] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0161] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. An activity configuration method, characterized in that, The method includes: Obtain operational target information for the target activity; Obtain a pre-built multimodal knowledge base, and generate target semantic context from the semantic graph of the multimodal knowledge base based on the operational target information; Using a large language model, semantic parsing and business reasoning are performed on the target semantic context to generate an initial activity configuration result corresponding to the operational target information; and The initial activity configuration results are subject to dynamic compliance review, and the final activity configuration results are determined based on the review results.

2. The method according to claim 1, characterized in that, The step of retrieving and generating target semantic context from the semantic graph of the multimodal knowledge base based on the operational target information includes: Identify business entities and constraints semantically related to the operational target information from the semantic graph; Using the constraints as path priority weights, constraint consistency retrieval is performed on multi-hop association paths between the business entities to obtain a candidate path set; and The target semantic context is generated based on the associated path in the candidate path set that satisfies the constraints and has the smallest path priority weight.

3. The method according to claim 1, characterized in that, The step of using a large language model to perform semantic parsing and business reasoning on the target semantic context includes: The target semantic context is hierarchically deconstructed to generate a semantic parsing tree; Based on the semantic parsing tree, the activity type, resource binding strategy, and field filling rules are determined, generating an initial activity configuration structure consistent with the business logic; and The initial activity configuration structure is converted into an executable configuration data format to form the initial activity configuration result.

4. The method according to claim 1, characterized in that, The dynamic compliance review of the initial activity configuration results includes: Based on the target semantic context, locate the set of constraints associated with the initial activity configuration result in the semantic graph; and Perform a logical consistency check on the constraint set to determine the audit results of the initial activity configuration results in terms of time constraint consistency, resource binding, and / or customer coverage conflicts.

5. The method according to claim 1, characterized in that, The method further includes: performing dynamic compliance review on the initial activity configuration result through a multi-layer rule adaptation engine, wherein the multi-layer rule adaptation engine includes: The basic format rules layer is used to validate the field length, color value format, and date format in the initial activity configuration result; The business logic rule layer is used to detect business logic anomalies based on the semantic graph inference; and The resource usage limit layer is used to monitor budget overruns and concurrent resource usage conflicts.

6. The method according to any one of claims 1 to 5, characterized in that, The process of determining the final activity configuration result based on the audit results includes: In response to the audit result indicating a defect in the initial activity configuration result, the target semantic node corresponding to the defect is identified; Multiple candidate field values ​​are generated based on the contextual relationship of the target semantic node in the semantic graph; The large language model is used to perform semantic consistency filtering on the multiple candidate field values ​​to determine the completion result; as well as The initial activity configuration result is updated based on the completion result to form the final activity configuration result.

7. The method according to any one of claims 1 to 5, characterized in that, The process of determining the final activity configuration result based on the audit results includes: In response to the audit result indicating that the initial activity configuration result has passed the audit, the initial activity configuration result is structurally encapsulated to generate a target configuration file with a compliance identifier; as well as A version signature is generated for the target configuration file and the audit parameters are recorded to form the final activity configuration result.

8. The method according to claim 7, characterized in that, The method further includes: After the final activity configuration result is put into operation, monitor the operation data during the operation process; The operational data is compared with the audit parameters to obtain operational deviation information; as well as In response to the operational deviation information exceeding the target threshold, the operational data and the audit result are associated and stored in the multimodal knowledge base.

9. An active configuration device, characterized in that, The device includes: The data acquisition module is used to: acquire operational target information for the target activity; The retrieval module is used to: acquire a pre-built multimodal knowledge base, and retrieve and generate target semantic context from the semantic graph of the multimodal knowledge base based on the operational target information; The parsing and reasoning module is used to: perform semantic parsing and business reasoning on the target semantic context using a large language model, and generate an initial activity configuration result corresponding to the operational target information; and The compliance review module is used to: perform dynamic compliance review on the initial activity configuration results, and determine the final activity configuration results based on the review results.

10. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 8.

12. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 8.

Images