Marketing intelligent agent operation middle platform system

Abstract

This application relates to the fields of artificial intelligence and digital marketing technology, and provides a marketing intelligence agent operation platform system that supports cross-domain federation and visual orchestration. In this system, the federation center registers marketing intelligence agents; the visual workflow orchestrator displays the capability components of the marketing intelligence agents and generates a marketing process graph based on layout operations; the execution engine obtains global semantic context information based on information from a unified data and knowledge platform; when a local marketing intelligence agent does not meet the task requirements of the current process node in the marketing process graph, it initiates a request to the federation center; upon receiving the request, the federation center generates a temporary access token and identifies the target external marketing intelligence agent; the execution engine generates a cross-domain call request based on the temporary access token and global semantic context information, and invokes the target external marketing intelligence agent; the federation center performs security protection processing on the response data of the target external marketing intelligence agent and sends it to the execution engine to generate marketing response results.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and digital marketing technology, and in particular to a marketing intelligence agent operation platform system. Background Technology

[0002] In response to the increasing complexity of data and the dynamic nature of market competition, the technological architecture of digital marketing has undergone several evolutions. However, these technologies still have structural limitations in achieving autonomy and optimization across the entire marketing campaign process.

[0003] The first stage is marketing automation based on preset rules. The core of this type of system is a workflow engine that triggers a series of marketing tasks, such as email distribution and user tag changes, based on pre-defined, deterministic logical rules (e.g., condition-action rules). The technology at this stage primarily addresses the automation of repetitive tasks in marketing operations. Its fundamental limitation lies in its static and closed logical framework, which cannot dynamically perceive and adaptively adjust to unexpected market changes or complex user behavior patterns. Its intelligence and decision-making capabilities are limited by the completeness and foresight of the rule base, making it difficult to cope with the high-dimensional, non-linear modern marketing environment.

[0004] The second stage focuses on enhancing single-point capabilities based on artificial intelligence models. With the development of machine learning, especially deep learning, a series of intelligent tools applied to vertical marketing processes have emerged. These tools utilize specific algorithmic models to optimize the efficiency of local tasks. For example, they apply natural language generation technology for the automated creation of marketing text content; predictive models to optimize bidding strategies and budget allocation in programmatic advertising systems; collaborative filtering or deep neural network models to build recommendation systems for personalized distribution of products or content to users; and machine learning models such as classification and regression to predict user lifetime value or assess churn risk. This stage of technology improves the efficiency and accuracy of individual marketing execution stages. However, this technology paradigm centered on "point-like intelligence" raises deeper problems at the system level: rigid processes and poor business adaptability. Furthermore, the core intelligent capabilities and data assets of the marketing business platform cannot be securely and efficiently discovered and accessed by other marketing business platforms, resulting in a huge waste of overall enterprise resources and hindering cross-business domain scenario innovation. Summary of the Invention

[0005] Therefore, it is necessary to provide a marketing intelligence agent operation platform system to address the aforementioned technical issues.

[0006] This application provides a marketing intelligence agent operation platform system, which includes a unified data and knowledge platform, a visual workflow orchestrator, an execution engine, and a federation center;

[0007] The Federation Center is used to register marketing agents from different marketing business platforms; each of these marketing business platforms has a corresponding execution engine deployed.

[0008] A visual workflow orchestrator is used to display the capability components of registered marketing agents, responding to user operations on the layout of marketing process nodes based on capability components, and generating a marketing process graph.

[0009] The execution engine is used to obtain global semantic context information based on marketing data and marketing knowledge corresponding to user marketing requests obtained from the unified data and knowledge platform; when executing the current process node of the marketing process graph, if the local marketing agent does not meet the task requirements of the current process node, it initiates a cross-domain call authorization request and capability discovery request to the federation center.

[0010] The federation center is used to generate temporary access tokens and identify target external marketing agents that are compatible with the current process node in response to cross-domain call authorization requests and capability discovery requests.

[0011] The execution engine is used to invoke the target external marketing agent based on a cross-domain invocation request generated by the temporary access token and global semantic context information.

[0012] The federation center is used to perform security protection processing on the response data received from the target external marketing intelligence agent in response to the cross-domain call request and then feed it back to the execution engine; the security-protected response data is used by the execution engine to generate a marketing response result corresponding to the user's marketing request.

[0013] The visual workflow orchestrator of this application can serve as a business interaction portal for the system, enabling users (such as marketing personnel) to independently design and visually construct marketing processes, avoiding the rigidity of marketing processes and better adapting to different marketing businesses. Moreover, the federation center can serve as the top-level extension framework of the system, ensuring the secure, controllable sharing and collaboration of marketing intelligence capabilities in cross-domain environments. Specifically, during cross-domain calls, the response data of the target external marketing intelligence is first processed for security protection by the federation center, thereby enabling collaborative invocation of cross-domain intelligence capabilities while ensuring data security and compliance. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of a marketing intelligence agent operation platform system in one embodiment;

[0015] Figure 2 This is a schematic diagram of a visualization workflow orchestrator in one embodiment;

[0016] Figure 3 This is a schematic diagram of the federal center in one embodiment;

[0017] Figure 4 This is a schematic diagram of the agent capability federation and cross-domain collaboration architecture in one embodiment. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] The related technologies, with their core paradigm of "point-based intelligence," raise deeper issues at the system level, specifically in the following six aspects.

[0020] Systemic fragmentation and data silos: Each functional module is typically built using different technology stacks and runs as an independent application, resulting in systemic functional and data silos. The lack of standardized data interfaces and interaction protocols between heterogeneous systems hinders the real-time integration and knowledge sharing of cross-domain data, making it difficult to achieve holistic data insights and comprehensive decision-making.

[0021] Lack of collaborative mechanisms and end-to-end process integration: Marketing campaigns are inherently complex and lengthy processes involving multiple stages such as market analysis, strategy planning, content generation, channel execution, and performance attribution. In related technologies, there is a lack of a unified collaborative framework and task scheduling center between the various intelligent modules. The continuity of the process relies on manual intervention for data import / export and instruction transmission, leading to workflow disruptions and preventing the formation of efficient "collective intelligence" among multiple intelligent units. The overall system efficiency is not a simple linear sum of the efficiency of each module.

[0022] A lack of autonomous planning and decomposition capabilities for high-level business objectives: Most intelligent tools in related technologies are passive, responsive execution units, relying on explicit and atomic instructions input by human operators. They lack the cognitive framework to autonomously understand, logically decompose, and transform high-level, abstract business objectives (e.g., improving quarterly conversion rates for a specific user group) into a series of executable sub-tasks. The core strategy formulation and decision-making functions in marketing campaigns remain entirely manual.

[0023] The lack of marketing knowledge assetization and iterative closed-loop: The massive amounts of data generated during marketing campaigns contain valuable knowledge about strategy effectiveness, user behavior patterns, and market dynamics, but this knowledge is mostly in unstructured or implicit forms. Related technological systems generally lack mechanisms to effectively extract, structure, and solidify this procedural knowledge into reusable knowledge assets. This prevents the system from learning autonomously from past successes or failures, making it difficult to form a self-iteratory and optimization closed loop of "perception-decision-execution-feedback."

[0024] Rigid processes and poor business adaptability: Even second-generation systems integrating multiple AI capabilities often have rigid and "black-box" built-in marketing strategies and workflows. Marketing personnel, as experts who best understand the market and users, are unable to autonomously configure the system's agent invocation order, data flow, and decision-making logic based on their unique business logic and rapidly changing strategic needs. This technology-driven business model limits marketing flexibility and the speed of innovation.

[0025] Enterprise-level platform silos and capability reuse bottlenecks: In large enterprises, different business units (e.g., automotive and financial businesses) may build their own marketing business platforms to serve their respective domains. While this achieves capability reuse within each business line, it creates new, higher-dimensional platform silos at the group level. The core intelligent capabilities and data assets of each marketing business platform cannot be securely and efficiently discovered and invoked by other business units, resulting in a huge waste of overall enterprise resources and hindering cross-business domain scenario innovation.

[0026] In summary, existing intelligent marketing technologies not only suffer from deficiencies in internal system collaboration and autonomy, but also have fundamental shortcomings in business-oriented flexibility and cross-organizational capability sharing. A new technical architecture is urgently needed that can empower business personnel to autonomously orchestrate processes while breaking down enterprise-level collaboration barriers.

[0027] The marketing intelligence agent operation platform system provided in this application supports cross-domain federation and visual orchestration. By constructing a systematic, multi-level intelligence agent management and scheduling structure, the system achieves unified governance of internal and external marketing capabilities, visual process orchestration, multi-agent collaborative execution, and cross-business domain capability federation and sharing. This empowers business personnel to independently build marketing strategies, flexibly combine intelligence agent capabilities, and reuse existing intelligence resources across domains, ultimately forming a system-level intelligent operation capability that is orchestratable, collaborative, and evolvable across the entire marketing chain.

[0028] In one exemplary embodiment, such as Figure 1As shown, the marketing intelligence agent operation platform system provided in this application includes a unified data and knowledge platform, a visual workflow orchestrator, an execution engine, and a federation center.

[0029] Among them, the unified data and knowledge platform can serve as the underlying data support platform of the system. Through four core components—data access and parsing module, structured processing module, procedural knowledge extraction component, and knowledge supply interface—it constructs a unified data model and knowledge supply system, providing consistent and traceable data semantics and knowledge context for upper-layer intelligent agents.

[0030] The marketing agent operation platform system may also include a marketing agent market and standardized protocols. The marketing agent market and standardized protocols can serve as the system's capability governance center, consisting of an agent capability declaration and verification module, a unified interface protocol module, and a data exchange protocol module, forming a standardized registration, release, and scheduling management system for multi-source heterogeneous agents.

[0031] A visual workflow orchestrator can serve as a business interaction portal for the system, such as... Figure 2 As shown, the visual workflow orchestrator can include a visual canvas interaction module, a process description model generation module, and a process verification and version management module. Through these three components, business personnel can independently design and visually construct marketing processes.

[0032] The execution engine can be called a multi-agent collaborative execution engine. The execution engine and the full-link monitoring module can serve as the runtime core of the system, including the process scheduling execution engine, the parallel control module, and the full-link monitoring and diagnostic module, forming a complete closed loop of operation from process parsing, agent scheduling to execution monitoring.

[0033] The federated center can adopt an agent-based capability federation and cross-domain collaborative architecture. The federated center can serve as a top-level extension framework for the system, such as... Figure 3 As shown, the Federation Center can include a federated capability registration and discovery mechanism, a cross-domain secure call protocol, and a full-link audit tracing module. Through these three parts, it can form a two-way linkage with the visual workflow orchestrator to ensure the secure, controllable sharing and collaboration of agent capabilities in cross-domain environments.

[0034] Each marketing business platform has a corresponding execution engine and its own marketing intelligence agent.

[0035] The Federation Center is used to register marketing agents from different marketing business platforms.

[0036] Specifically, the federation center can adopt an agent capability federation and cross-domain collaboration architecture. This architecture serves as a top-level control system for enterprise-level intelligent resource sharing, aiming to break down long-standing data barriers and capability silos between different business platforms. This allows the secure, controllable, and auditable flow of high-value agent resources accumulated within each marketing business platform across the entire group organization. By constructing a logical-level federation center, this architecture enables each marketing business platform to access its marketing agent capabilities as services through a lightweight registration mechanism without deep integration. Standardized capability description protocols ensure cross-platform discoverability, matching, and invocation. The federation center can uniformly maintain cross-domain capability catalogs, version information, quality indicators, and service availability status, thus forming a group-level agent capability index system.

[0037] The federated capability registration and discovery mechanism of the intelligent agent capability federation and cross-domain collaborative architecture enables unified management of cross-domain intelligent agents through a capability registration interface and a semantic matching engine. The capability registration interface provides a standardized capability access point for each marketing business platform, receiving and verifying the capability declaration files of cross-domain marketing intelligent agents and registering their metadata in the federated capability directory. The semantic matching engine, based on natural language processing and knowledge graph technology, performs semantic annotation and classification on the registered capabilities, establishing a relationship between capabilities and business scenarios. This allows for the generation of business scenario tags corresponding to the capabilities. The semantic matching engine also supports capability discovery and recommendation based on semantic similarity.

[0038] A visual workflow orchestrator is used to display the capability components of registered marketing agents, responding to user operations on the layout of marketing process nodes based on capability components, and generating a marketing process graph.

[0039] Specifically, a visual workflow orchestrator may include a visual canvas interaction module, a process description model generation module, and a process verification and version management module.

[0040] The visual canvas interaction module provides users with an intuitive process building experience through the collaborative work of the node drag-and-drop controller and the connection manager. The node drag-and-drop controller retrieves a list of registered marketing agents from the agent marketplace and presents their capabilities as visual components in the canvas sidebar. Users can arrange marketing process nodes based on these capability components; specifically, users can drag and drop marketing agent capability nodes onto the canvas workspace, thereby generating a marketing process graph. The connection manager monitors the positional relationships of nodes in the canvas in real time. When a user attempts to connect two nodes, it automatically verifies whether the output structure of the source node matches the input requirements of the target node and provides feedback on the feasibility of the connection through color-coded and hovering prompts.

[0041] The process description model generation module can include a process structure analyzer and a model serialization component. The process structure analyzer performs topological analysis on the node layout in the canvas, identifying the start node, end node, parallel branches, and conditional decision nodes of the process, and constructing a complete process logic structure. The model serialization component transforms the analyzed process structure into an executable process description model for the system. This model can use a standardized JSON-based format to fully record the configuration parameters of each node, the connection relationships between nodes, data transmission paths, and error handling strategies.

[0042] The process validation and version management module can perform automated validation after the user completes the process design, ensuring the process's executability through a dependency checker and a parameter integrity validator. The dependency checker identifies structural issues in the process, such as circular dependencies and missing necessary inputs; the parameter integrity validator checks whether the required parameters of each node are correctly configured. Validated processes are assigned a unique version identifier and stored in the process library, supporting version tracing, comparison, and rollback operations. Through the collaboration of these modules, the visual workflow orchestrator can achieve a seamless transition from visual design to executable processes.

[0043] The execution engine is used to obtain global semantic context information based on marketing data and marketing knowledge corresponding to user marketing requests obtained from the unified data and knowledge platform. When executing the current process node of the marketing process graph, if the local marketing agent does not meet the task requirements of the current process node, it initiates a cross-domain call authorization request and capability discovery request to the federation center.

[0044] Specifically, a user can issue a marketing request within a specific marketing business platform; this request can be termed a user marketing request. The execution engine deployed within this marketing business platform can obtain marketing data and knowledge corresponding to the user's marketing request from a unified data and knowledge platform, encapsulating this data and knowledge into standardized global semantic context information. This global semantic context information can be injected into each scheduled marketing agent instance during process execution, serving as a shared input environment for the marketing agent instances. The execution engine can execute based on a marketing process graph; the marketing process graph can only describe task logic and capability requirement semantics, without forcibly binding to a specific marketing business platform to which the capability belongs. Therefore, when the execution engine reaches the current process node, it can determine whether the local marketing agent meets the task requirements of the current process node; if not, the execution engine can initiate a cross-domain call authorization request and capability discovery request to the federation center. The local marketing agent is the marketing agent owned by the marketing business platform where the execution engine is deployed.

[0045] The Federation Center is used to generate temporary access tokens and identify target external marketing agents that are compatible with the current process node in response to cross-domain call authorization requests and capability discovery requests.

[0046] Specifically, after receiving a cross-domain call authorization request, the Federation Center can verify the request and generate a temporary access token if the verification is successful. The temporary access token expires after the cross-domain call.

[0047] To ensure that cross-domain calls meet security, compliance, and controllability requirements during transmission between different marketing business platforms, this application constructs a multi-layered, end-to-end access control and data protection system within the intelligent agent capability federation architecture of the federation center. First, the intelligent agent capability federation architecture adopts a zero-trust security concept, treating each cross-domain call request as a potentially untrusted source. It does not rely on any existing network boundary or system-level trust assumptions, but instead achieves real-time authentication through a combination of dynamic keys, short-term authorization credentials, and context verification. The federation center can dynamically generate a temporary access token (also known as a temporary authorization token) valid only within the current cross-domain call period based on multi-dimensional characteristics such as the identity features of the calling marketing business platform, the type of capability requested, historical call behavior, runtime environment, and task sensitivity. This token is destroyed after the cross-domain call ends to prevent credential reuse or theft.

[0048] The federated architecture of the intelligent agent capabilities in the federated center can also include a cross-domain secure invocation protocol. This protocol ensures the security of cross-domain calls through two main modules: dynamic authentication and authorization, and secure data transmission. The dynamic authentication and authorization module logically belongs to the federated center and is used to uniformly manage the identity verification and authorization policies for cross-domain calls. Each marketing business platform has a local execution engine deployed within it, used to parse and execute the marketing process graph constructed by the user through a visual orchestrator, and to act as the initiator of cross-domain intelligent agent calls. When the execution engine of a business platform identifies, during the execution of the marketing process graph, that the intelligent agent capability corresponding to the current process node does not belong to this platform, or that there are no available capabilities in its local capability pool that meet the task semantics, performance, or compliance constraints, the execution engine can initiate a cross-domain call authorization request to the federated center, thereby triggering the workflow of the dynamic authentication and authorization module.

[0049] The dynamic authentication and authorization module is based on a zero-trust security architecture, independently authenticating and authorizing each cross-domain call request without relying on any pre-defined network trust relationships. The module comprehensively verifies the request based on the caller's platform identity, target capability identifier, process context, runtime environment, and security policies. Upon successful verification, it generates a unique temporary access token for the cross-domain call. This temporary access token can be bound to the current call session, its validity period can extend only to the current call cycle, and it expires immediately upon call completion or abnormal interruption. The generated temporary access token can be returned to the execution engine that initiated the call, allowing the engine to carry the token in subsequent cross-domain capability call requests to securely access the target external marketing agent.

[0050] Upon receiving a capability discovery request, the Federation Center can use a semantic matching engine to identify the target external marketing agent that is compatible with the current process node. The Federation Center can then return the callable entry point of the target external marketing agent to the requesting execution engine via a secure call session. This allows cross-domain capabilities to be dynamically leased without exposing underlying implementation details, thereby enabling on-demand collaboration and reuse of agent capabilities between different marketing business platforms.

[0051] The execution engine is used to generate cross-domain call requests based on temporary access tokens and global semantic context information, and to invoke the target external marketing agent.

[0052] Specifically, after receiving a temporary access token and the callable entry point of the target external marketing agent, the execution engine can generate a cross-domain call request based on the temporary access token and global semantic context information. Based on this cross-domain call request and the callable entry point of the target external marketing agent, the execution engine can then invoke the target external marketing agent. Specifically, the execution engine can bind the cross-domain call request and the callable entry point of the target external marketing agent and send them back to the federation center, which then invokes the target external marketing agent. After being invoked, the target external marketing agent can output response data to the cross-domain call request and send the response data to the federation center.

[0053] The Federation Center is used to perform security protection processing on response data received from the target external marketing intelligence agent in response to cross-domain call requests, and then feed the response data back to the execution engine. The security-protected response data is used by the execution engine to generate marketing response results corresponding to the user's marketing request.

[0054] Specifically, when the Federated Center receives response data from the target external marketing agent in response to a cross-domain call request, it can perform security protection processing on the response data and feed the security-protected response data back to the execution engine. As a result, the execution engine can continue processing based on the security-protected response data, thereby generating a marketing response result corresponding to the user's marketing request.

[0055] Regarding data protection, this application introduces a deeply configurable field-level anonymization and usage restriction mechanism. Upon receiving a cross-domain call request, the Federation Center can activate its built-in data policy engine to perform structured analysis on the response data based on the "usage constraint labels" and "sensitivity grading rules" set by the capability provider. It automatically performs security operations such as data pruning, field hiding, hash mapping, obfuscation, or minimal information exposure, resulting in security-protected response data. This ensures that the caller can only access the data necessary to complete the task and will not obtain any sensitive fields unrelated to the task due to the capability call. Furthermore, the Federation Center can construct an independent "security context" for each cross-domain call. This context includes a call chain identifier, access permission scope, data flow path, call validity window, and usage restriction flags, ensuring that the call operates within strict policy boundaries throughout its lifecycle. All data flowing through the Federation Center (including input parameters, context metadata, and agent execution results) must be automatically wrapped into a structured secure container by the policy engine, ensuring that data remains encrypted, securely isolated, and traceable even during cross-network transmission.

[0056] The visual workflow orchestrator of this application can serve as a business interaction portal for the system, enabling users (such as marketing personnel) to independently design and visually construct marketing processes, avoiding the rigidity of marketing processes and better adapting to different marketing businesses. Moreover, the federation center can serve as the top-level extension framework of the system, ensuring the secure, controllable sharing and collaboration of marketing intelligence capabilities in cross-domain environments. Specifically, during cross-domain calls, the response data of the target external marketing intelligence is first processed for security protection by the federation center, thereby enabling collaborative invocation of cross-domain intelligence capabilities while ensuring data security and compliance.

[0057] In one embodiment, the execution engine is configured to: determine that the local marketing agent does not meet the task requirements of the current process node if the capability corresponding to the current process node does not belong to the local marketing business platform, or if the capability pool of the local marketing business platform does not have a matching available marketing agent capability.

[0058] When the execution engine reaches the current process node, it can search the local marketing business platform's capability pool for intelligent agent capabilities that meet the task semantics, performance, and compliance constraints of the current process node. If no available marketing intelligent agent capability is found in the local marketing business platform's internal capability set, or if the intelligent agent capability corresponding to the node to be executed does not belong to the local marketing business platform, the execution engine can determine that the internal capabilities cannot meet the task requirements of the current process node and initiate a capability discovery request to the federation center.

[0059] In one embodiment, the execution engine, when executing the current process node of the marketing process graph, generates a local call request based on global semantic context information when the local marketing agent meets the task requirements of the current process node; the execution engine, based on the local call request, calls the matching local marketing agent to obtain the response data of the local marketing agent, so as to generate a marketing response result corresponding to the user's marketing request.

[0060] When the execution engine reaches the current process node, if it determines that the local marketing agent meets the task requirements of the current process node, it can generate a local call request based on the global semantic context information and call the matching local marketing agent according to the local call request. After being called, the local marketing agent can output the corresponding response data. After receiving the response data, the execution engine can continue to process it, thereby generating a marketing response result corresponding to the user's marketing request.

[0061] In one embodiment, when identifying a target external marketing agent that fits the current process node, the federated center specifically uses:

[0062] Obtain the task semantics of the current process node, and identify several candidate marketing agent capabilities in the federated capability library that meet the similarity conditions with the task semantics; the federated capability library includes registered marketing agent capabilities; among the several candidate marketing agent capabilities, identify the candidate marketing agent capabilities associated with the current business scenario tag; based on the candidate marketing agent capabilities associated with the current business scenario tag, determine the target external marketing agent that is compatible with the current process node.

[0063] After receiving a capability discovery request, the Federation Center can match the task semantics and capability semantics of the current process node based on the semantic matching engine, and filter out a set of candidate marketing agent capabilities by combining the relationship between capabilities and business scenarios. On this basis, it can further sort and evaluate the capability performance indicators, service availability status and access permission rules to determine the most suitable external agent in the current calling scenario, thereby obtaining the target external marketing agent.

[0064] For example, when the task semantics of the current process node are user churn prediction, the semantic matching engine can filter candidate marketing agent capabilities with high semantic similarity to churn prediction from the federated capability library to obtain preliminary candidate marketing agent capabilities, such as user churn prediction capabilities in membership management scenarios or user churn prediction capabilities in e-commerce operation scenarios. Then, combined with the current business scenario tags, capabilities strongly associated with the current business scenario tags are prioritized as final candidate marketing agent capabilities. Further comprehensive evaluation of capability performance indicators, service availability status, and access permission rules is conducted to determine the most suitable external agent, such as the user churn prediction agent in the e-commerce operation scenario. The most suitable external agent is an external agent that, relative to the marketing business platform making the current request, does not belong to its local capability pool.

[0065] The Federation Center can then return a callable entry point for the target external marketing agent to the requesting platform via a secure call session. This allows cross-domain capabilities to be dynamically rented without exposing the underlying implementation details, thereby enabling on-demand collaboration and reuse of agent capabilities between different business platforms.

[0066] The marketing intelligence agent operation platform system provided in this application can be applied to the field of digital marketing. It can uniformly manage and schedule marketing intelligence agents, allow users to visually orchestrate workflows, and support the federated reuse of capabilities across business platforms.

[0067] In one embodiment, the data access and parsing module of the unified data and knowledge platform can work collaboratively with a multi-source data adapter and a metadata automatic identification unit to achieve unified access to heterogeneous data from multiple sources, including customer relationship management systems, user behavior logs, advertising delivery systems, and external data services. The multi-source data adapter provides dedicated connection protocols and parsing logic for different types of data sources, automatically adapting to different access methods such as API interfaces, direct database connections, and file transfers. The metadata automatic identification unit can intelligently identify metadata such as field types, primary key relationships, and timestamp formats of the accessed data. Based on a preset entity-event model, it performs preliminary semantic alignment of the data from dimensions such as user entity identification, business event semantics, time dimension, and behavioral context, uniformly mapping data from different sources to standardized user entity, business event, and marketing feature structures.

[0068] For example, customer records in the customer relationship management system, device or account identifiers in user behavior logs, and audience identifiers in the advertising system are uniformly mapped to the same user entity; different behaviors such as clicks, submissions, and conversions are uniformly categorized into standardized business event types. The aligned and mapped data serves as input to subsequent structured processing modules, supporting operations such as data cleaning, primary key reconstruction, feature generation, and procedural knowledge extraction, thereby providing a consistent semantic foundation for data modeling and agent reasoning within the platform.

[0069] The structured processing module can include a data cleaning unit and a feature management unit. The data cleaning unit performs deduplication, format standardization, outlier handling, and primary key reconstruction on the raw data to ensure data quality meets the requirements of the intelligent agent. The feature management unit can automatically construct marketing features based on preset feature generation rules and maintain a feature version library, recording the generation logic, data source, and version evolution history of each feature. The processed data is stored in a unified format in the feature library and the basic data warehouse, forming a traceable data asset system.

[0070] The procedural knowledge extraction component runs continuously during the marketing process. Through two main functional units—a rule parsing engine and a knowledge fragment generator—it transforms the execution engine's output logs, node input / output content, and related metrics into structured knowledge fragments. The rule parsing engine, based on predefined knowledge extraction rules, identifies key decision points, execution results, and user feedback in the logs. The knowledge fragment generator organizes this information into knowledge objects in a unified format, adding metadata such as timestamps and process identifiers, and stores them in a knowledge base.

[0071] The knowledge supply interface provides a unified knowledge access portal for upper-layer intelligent agents, supporting combined queries based on entity dimensions, event types, or policy contexts. The interface can include a knowledge version selector, automatically choosing the most suitable knowledge version based on the time range and task requirements during query execution, ensuring that the intelligent agent obtains accurate and consistent knowledge context. Through the collaborative work of these modules, the unified data and knowledge platform constructs a complete data support system from data access and processing to knowledge accumulation and supply.

[0072] In one embodiment, the agent capability declaration and verification module of the marketing agent marketplace and standardized protocol can achieve unified description and quality control of agent capabilities through structured declaration files and automated verification processes. The capability declaration file can adopt a unified JSON Schema format to define the agent's input parameter structure, output format specifications, execution resource requirements, version information, and applicable scenario descriptions. The automated verification process can perform full verification on the submitted declaration files, including syntax correctness checks, parameter integrity verification, interface compatibility testing, and version conflict detection, ensuring that only agents that meet the quality standards can be registered in the marketplace.

[0073] The unified interface protocol module defines standardized communication specifications for agent calls and can include two core components: a request body builder and a response body parser. The request body builder automatically generates call requests that conform to the protocol requirements based on capability declarations, uniformly encapsulating capability identifiers, version numbers, operation types, input parameter sets, and runtime context information. The response body parser performs standardized parsing of the results returned by the agent, extracting execution status, output data, and diagnostic information, and verifying the consistency of its structure with the declaration file.

[0074] The data exchange protocol module ensures structural consistency in cross-agent data interaction through standardized object models and semantic mapping rules. This module can define standard data objects in the marketing domain, including user profile objects, marketing content objects, and channel configuration objects, and specify the required, optional, and extended fields for each object. The semantic mapping rule engine can automatically map the private data formats of each agent to the standard object format, eliminating data exchange barriers caused by structural differences. Through the synergistic effect of these mechanisms, the agent market establishes a unified capability governance system, providing a technological foundation for visual orchestration and cross-domain collaboration.

[0075] In one embodiment, the multi-agent collaborative execution engine and the process scheduling execution engine of the end-to-end monitoring module can achieve runtime execution of a visualized process through the collaborative work of a process parser and a node scheduler. The process parser can load the process description model, parse its node information, dependencies, and execution logic, and construct an internal execution plan. The node scheduler can dynamically determine the execution order of nodes based on the dependencies in the execution plan, initiating parallel execution for nodes without dependencies and ensuring the correctness of the execution order for nodes with sequential dependencies.

[0076] The parallel control module can optimize the scheduling of parallel nodes through a resource pool manager and a concurrency coordinator. The resource pool manager can monitor the system's computing resource usage, including CPU, memory, and network bandwidth, and dynamically allocate execution resources according to the resource needs of the agent. The concurrency coordinator can manage the synchronization and communication between parallel execution nodes, ensuring the correct transmission of shared data and the consistency of state, and preventing resource conflicts and data races.

[0077] The end-to-end monitoring module comprises two main components: a real-time metrics collector and an execution tracker. The real-time metrics collector gathers key performance indicators (KPIs) during the execution of each agent node, including response time, resource consumption, execution status, and output summary. The execution tracker constructs a complete call chain record, capturing all critical events during process execution, including node start / end times, input / output data snapshots, anomaly information, and user interaction logs. Monitoring data is aggregated in real-time onto a monitoring dashboard, providing users with a visual representation of process execution and supporting automatic alerts and root cause analysis for execution anomalies. Through the collaboration of these components, the execution engine ensures the reliable operation and transparent monitoring of complex marketing processes.

[0078] Regarding data security, this application utilizes a secure data transmission module to provide end-to-end protection for data during cross-domain calls. This module is centrally controlled at the federation layer. When the target external marketing agent completes its capability execution and returns response data, the response data can enter a cross-domain transmission channel constrained by federated security policies. The secure data transmission module can perform field-level security processing on the cross-domain transmitted data based on pre-configured data sensitivity grading rules and usage restriction policies. This includes, but is not limited to, data anonymization, encryption, field pruning, or access control, ensuring that the caller can only obtain the data view necessary to complete the current task. The securely processed response data can be returned to the execution engine of the marketing business platform that initiated the call, allowing the execution engine to continue driving the execution of subsequent process nodes. This achieves collaborative invocation of cross-domain agent capabilities while ensuring data security and compliance.

[0079] The end-to-end audit tracing module enables full traceability of cross-domain calls through a call chain recorder and a security analyzer. The call chain recorder captures all critical events during the cross-domain call process, including detailed logs of capability discovery, authorization verification, request forwarding, and result return. The security analyzer uses machine learning algorithms to perform real-time analysis of the audit logs, identifying abnormal call patterns, potential security threats, and performance bottlenecks, triggering timely security alerts and protective measures. Through the synergistic effect of these mechanisms, the intelligent agent capability federation architecture can achieve cross-business domain intelligent agent capability sharing and collaboration while ensuring security and compliance.

[0080] The system provided in this application achieves a complete closed-loop operation of marketing agents through the coordinated operation of five core modules: a unified data and knowledge platform, a marketing agent marketplace and standardized protocols, a visual workflow orchestrator, a multi-agent collaborative execution engine and end-to-end monitoring module, and an agent capability federation and cross-domain collaborative architecture. Firstly, the system can construct a unified data model and knowledge supply system through the unified data and knowledge platform, providing consistent and traceable data semantics and knowledge context for upper-layer agents, thus eliminating data silos and knowledge fragmentation at their root. Subsequently, the marketing agent marketplace and standardized protocols can establish a unified capability governance system, enabling interconnection and interoperability of multi-source heterogeneous agents through standardized declarations and protocols, solving the problems of system fragmentation and capability reuse.

[0081] Building upon this foundation, a visual workflow orchestrator empowers business personnel to autonomously construct marketing processes. Its intuitive drag-and-drop interface lowers the technical barrier, fundamentally shifting from predefined processes to user-managed orchestration. A multi-agent collaborative execution engine and end-to-end monitoring module ensure reliable operation and transparent monitoring of complex marketing processes, achieving controllability and optimizability through intelligent scheduling and real-time diagnostics. The agent capability federation and cross-domain collaborative architecture break down organizational boundaries, enabling cross-business domain agent capability sharing and collaboration while ensuring security and compliance, forming an enterprise-level intelligent capability network. (Refer to...) Figure 4 , Figure 4 This is a schematic diagram of the architecture of the intelligent agent capability federation and cross-domain collaboration architecture.

[0082] Through the collaborative design of the aforementioned five-layer architecture, the system provided in this application can form a complete, bottom-up intelligent operation system throughout the entire process of marketing strategy design, capability combination, process execution, effect monitoring, and cross-domain collaboration. It can achieve seamless integration from business needs to technical implementation, effectively ensuring the flexibility, efficiency, and innovation capabilities of marketing activities. This system can effectively overcome core problems existing in current marketing systems, such as rigid processes, capability silos, data barriers, and difficulties in cross-domain collaboration. It can improve the agility, collaboration, and intelligence of enterprises in the digital marketing environment, thereby building an evolvable, scalable, and governable next-generation marketing intelligence operation infrastructure.

[0083] This application provides a marketing agent operation platform system that supports cross-domain federation and visual orchestration. By constructing a five-layer collaborative architecture integrating a unified data and knowledge platform, a marketing agent marketplace and standardized protocols, a visual workflow orchestrator, a multi-agent collaborative execution engine and a full-link monitoring module, and an agent capability federation and cross-domain collaborative architecture, this application systematically solves the problems existing in related technologies. The system provided by this application can achieve unified governance and visual process orchestration of internal and external marketing capabilities, ensuring the efficiency and reliability of multi-agent collaborative execution. It also enables secure sharing and collaboration of agent capabilities in cross-business domain environments, thereby significantly overcoming core problems such as system fragmentation, data silos, process rigidity, and platform silos across the entire chain, and improving the agility, collaboration, and intelligence level of enterprise marketing operations.

[0084] The marketing intelligence agent operation platform system provided in this application can adopt a five-layer architecture design, including a unified data and knowledge platform, a marketing intelligence agent marketplace and standardized protocols, a visual workflow orchestrator, a multi-agent collaborative execution engine and a full-link monitoring module, and an intelligence agent capability federation and cross-domain collaborative architecture. Among them, the unified data and knowledge platform is responsible for building a unified data model and knowledge supply system; the marketing intelligence agent marketplace and standardized protocols enable standardized governance of multi-source heterogeneous intelligence agents; the visual workflow orchestrator provides an interactive interface for business personnel to independently build marketing processes; the multi-agent collaborative execution engine and full-link monitoring module ensure the reliable operation and transparent monitoring of complex marketing processes; and the intelligence agent capability federation and cross-domain collaborative architecture enable secure sharing of intelligence agent capabilities across business domains. The five-layer architecture forms a complete intelligent operation closed loop through the interaction of standardized data flow, control flow, and business flow. Each layer is relatively independent yet collaborative, jointly constructing a full-link intelligent marketing guarantee from data preparation, capability governance, process design, execution monitoring to cross-domain collaboration.

[0085] (1) The functions of the unified data and knowledge platform include building the underlying data support platform of the system, forming a unified data model and knowledge supply system through multi-source data access and standardized processing, and providing consistent and traceable data semantics and knowledge context for upper-level intelligent agents.

[0086] The unified data and knowledge platform serves as the underlying infrastructure within the system, supporting data access, structured modeling, knowledge accumulation, and intelligent agent knowledge provision through a relatively fixed service-oriented architecture. Firstly, the unified data and knowledge platform can unify the access of multi-source heterogeneous data from customer relationship management systems, user behavior logs, advertising systems, and external data services through its data access module. During the access process, the built-in parsing component can automatically identify metadata such as field types, primary keys, and timestamp representations. Based on a preset entity-event model, it performs preliminary semantic alignment of the data from dimensions such as user entity identification, business event semantics, time dimension, and behavioral context. This maps data from different sources into a standardized infrastructure of user entities, business events, and marketing features, providing a consistent semantic foundation for subsequent data processing, feature management, and knowledge modeling.

[0087] After model alignment, the structured processing module of the unified data and knowledge platform can perform operations such as data cleaning, format conversion, feature generation, and primary key reconstruction, enabling the data to be stored in a unified format in the feature library and basic data warehouse. Internally, the unified data and knowledge platform maintains a lightweight feature management mechanism to record the generation rules, dependent fields, and version information of each feature. Through this mechanism, the marketing agent can obtain the business semantics, data source, and generation rules of features when performing inference or strategy decisions. Furthermore, in multiple rounds of inference, strategy evaluation, or model updates, it can backtrack historical feature versions and map them to the current standardized semantics, ensuring that the features and data used by the agent remain consistent across different times and task scenarios. For example, the feature "total purchase amount of users in the past 30 days" has the business semantics of the user's total consumption in the past 30 days, the source field is the order table's amount field, and the generation rule is to sum by user ID. Historical versions may only count online orders. During backtracking, the agent can obtain the time range and statistical range of that version of the feature and, in inference or strategy evaluation, only compare the online order portion to maintain semantic consistency across time.

[0088] Based on this data, the unified data and knowledge platform can provide a procedural knowledge extraction component. This component continuously receives logs, node input / output content, and relevant metrics from the execution engine during the marketing process. Through rule parsing and template processing, it transforms these processes into structured knowledge fragments and stores them in the knowledge base. Each knowledge fragment is associated with a timestamp and task identifier. Event context refers to additional information associated with a specific event, which may include, but is not limited to, the user or entity identifier that triggered the event, the execution strategy or task identifier, the time of the event, and node input / output data. When influencing an agent to perform reasoning or decision-making, it can input a time range, task identifier, and optional entity or event context information into the knowledge base to retrieve corresponding knowledge fragments on demand. This enables knowledge index queries based on time, task, and event context, thereby obtaining knowledge related to the current reasoning or decision-making task.

[0089] Knowledge is provided externally through the knowledge supply interface of the unified data and knowledge platform. This interface supports queries based on entity, event, or strategy context. Upon invocation, the interface automatically selects the corresponding knowledge version based on the time range, task identifier, and event context information provided by the marketing agent. This ensures that the knowledge acquired by the agent during reasoning or strategy execution remains semantically consistent with the current task. Different knowledge versions refer to different instances of the same knowledge object generated under different generation rules, time, or task conditions. By selecting the appropriate knowledge version, the agent obtains field structures and feature definitions consistent with the current task, thus ensuring semantic consistency in reasoning and decision-making. Furthermore, the unified data and knowledge platform maintains a simple data feedback mechanism, automatically writing the execution results and user feedback back to relevant data tables after the marketing process is completed. This data is then used for subsequent knowledge updates or model training, ensuring continuous iteration of the knowledge system.

[0090] Through the above structured implementation, the unified data and knowledge platform can provide the system with a stable, unified, and traceable data and knowledge foundation, enabling the reasoning and decision-making of upper-level intelligent agents to be completed in a consistent data semantics and reliable knowledge context.

[0091] (2) The functions of the marketing agent market and standardization protocol include establishing a unified governance system for agent capabilities, realizing the registration, discovery and interoperability of multi-source heterogeneous agents through standardized declarations and protocols, and providing a standardized underlying communication foundation for visual orchestration and cross-domain collaboration.

[0092] A unified marketing intelligence agent marketplace and standardized protocols can resolve core technical obstacles such as fragmented intelligent capabilities, incompatible interfaces, and inconsistent invocation methods. To this end, marketing intelligence agents can first undergo capability abstraction, with their capabilities uniformly defined through structured declaration files. Capability declarations standardize the input parameters, output formats, execution logic, dependent resources, version numbers, applicable scenarios, and runtime environment requirements of the marketing intelligence agent, ensuring a consistent behavioral model when registered, scheduled, and invoked. The intelligence agent marketplace, acting as the registration and publishing center for marketing intelligence agents, receives capability declarations submitted by developers and performs full verification of the declaration content through a built-in automatic validation mechanism. This validation process includes field integrity checks, parameter type validation, input / output structure matching checks, dependent resource compatibility verification, and backward compatibility analysis for version changes. Based on the types and constraints of each parameter in the declaration, the system can automatically compare whether input fields are missing, whether undefined fields exist, whether types are consistent, and check whether the output structure conforms to the system's preset data exchange protocol format. Meanwhile, regarding the runtime environment, the system verifies whether the model version, inference image, and external service dependencies meet the declared requirements to ensure that the marketing agent can execute stably in the platform's runtime environment. For agent version iterations, the system automatically compares the differences between the old and new versions to determine whether the update will disrupt existing workflows or affect the reliability of other agent call chains, thereby ensuring that the evolution of capabilities will not have a destructive impact on the system.

[0093] To enable marketing agents to interoperate across systems, languages, and technology frameworks, this embodiment defines a unified interface protocol. This protocol stipulates that all marketing agent communication must use structured request and response bodies. The call request body of a marketing agent uniformly includes a capability identifier, version number, operation name, set of input parameters, and runtime context, typically encapsulated in a JSON structure. When the system triggers an agent, it automatically constructs a call request according to the protocol, for example: "{'agent_id':'content_generator','version': '1.2.0', 'operation': 'generate','inputs': { 'topic': 'New Product Launch Event', 'tone': 'Official', 'length': 120},'context': { 'trace_id': 'ab12-89fe-33cd', 'auth_token': 'xxx', 'sensitivity_level': 2}}". This structure is constrained by capability declarations generated by the agent market, enabling the system to consistently pass parameters and context information when invoking any agent.

[0094] After the marketing agent completes its execution, the system returns a response body in a standardized format, which can be represented as JSON. This response includes fields such as execution status, output results, and diagnostic information. For example, a successful execution response could be expressed as: "{ 'status':'success', 'outputs': { 'content': 'Dear user, the new product launched this time…'},'diagnostics': { 'latency_ms': 128, 'resource_usage': { 'cpu': 0.32,'memory': 256}, 'logs': []}}". This structure allows the caller to parse the output without needing to understand the agent's internal implementation and to utilize the diagnostic information for link monitoring, debugging, or performance analysis. The format of the response body is also uniformly constrained by capability declarations and system protocols, ensuring structural consistency across all marketing agents when returning content.

[0095] To ensure that no structural conflicts occur during data interaction of the marketing intelligence agent, the data exchange protocol proposed in this embodiment performs unified semantic modeling for common marketing objects (such as user profiles, marketing content, delivery parameters, channel configurations, etc.), specifying field sets, extended field rules, and variable constraint methods. All data transmitted across modules is first mapped to standardized objects to avoid operational anomalies caused by inconsistent field naming and structural incompatibility between different business systems. In addition, the protocol can specify sensitive data de-identification strategies, field-level encryption methods, and compliance requirements for cross-domain transmission, enabling the marketing intelligence agent's capabilities to be securely shared between different marketing business platforms within the group.

[0096] By establishing an intelligent agent market, an automatic verification mechanism, and a unified communication protocol, this embodiment can reconstruct previously scattered intelligent tools into standard capability units that are registerable, publishable, governable, and schedulable. These units can be consistently referenced by the visual orchestrator and run stably within the execution engine. Simultaneously, this standardized foundation provides a standardized underlying communication structure for subsequent cross-domain federated capability sharing, enabling intelligent agents to be securely and transparently leased and reused in cross-platform scenarios, thereby forming a governable and scalable intelligent capability network.

[0097] (3) The functions of the visual workflow orchestrator include providing business personnel with an interactive portal to build marketing processes independently. Through drag-and-drop interface and automated process generation, it enables seamless transformation from business needs to executable processes, greatly reducing the technical threshold for building marketing strategies.

[0098] Visual workflow orchestrators address the issue of rigid workflows in marketing systems, which cannot be independently configured by business personnel. At the implementation level, a visual workflow orchestrator includes a visual canvas interaction unit, a workflow description model generation unit, and a workflow validation and version management unit. These units respectively handle workflow construction, structured workflow representation, and workflow validity control and lifecycle management. The visual workflow orchestrator provides an interactive approach based on a visual canvas, allowing users to drag and drop desired agents from an agent marketplace and place them as workflow nodes on the canvas. The system automatically identifies the input / output requirements and dependencies of nodes based on the agent's capability declarations in the background, and provides real-time prompts on the connectability between nodes, thereby ensuring the structural correctness of the workflow chain formed during the orchestration process.

[0099] After the user completes the layout of process nodes, the visual workflow orchestrator can automatically generate the corresponding process description model. The process description model is typically represented using a structured flowchart, internally expressing node information, execution order, triggering conditions, error handling paths, and data transfer logic in a unified format. For example, a node in the process description can be represented as a structured object like { "node_id": "generate_copy_01", "agent": "content_generator", "inputs":{ "topic": "${user_input.topic}"}, "next": ["a_b_test_02"]}, but the system treats it as part of the process description rather than program code. Through this structural representation, the visual workflow orchestrator ensures both the readability of the process and enables the downstream execution engine to accurately parse the operating rules of each node.

[0100] The visual workflow orchestrator can automatically check parameter matching, data flow integrity, and the presence of missing required inputs based on the capability interfaces provided by the agent declarations during process construction. When potential conflicts or omissions are detected, the system will provide immediate prompts on the interface, preventing users from creating unexecutable processes. Once the process is built, the visual workflow orchestrator can submit the entire marketing process graph to the execution engine and generate versionable process instances, enabling users to save, reuse, debug, and rollback the marketing processes.

[0101] This visual orchestration approach allows business users to independently build complex marketing automation processes without needing to understand the underlying model calling methods, interface formats, or internal system logic. It improves the usability of agent capabilities and the efficiency of marketing strategy iteration. Through this module, the system can achieve a fundamental shift from predefined processes to user-managed processes.

[0102] (4) The functions of the multi-agent collaborative execution engine and the full-link monitoring module include ensuring the reliable operation and transparent monitoring of complex marketing processes, realizing the controllability and optimizability of process execution through intelligent scheduling and real-time diagnosis, and forming a continuous improvement mechanism driven by execution feedback.

[0103] The multi-agent collaborative execution engine is responsible for transforming the flowchart generated by the visual orchestrator into a runnable scheduling structure. During execution, it combines the data semantics and knowledge context provided by the unified data and knowledge platform to collaboratively schedule and merge the results of multiple marketing agents in the marketing flowchart. After loading the marketing flowchart, the execution engine can construct a schedulable execution graph based on the dependencies between nodes. At runtime, it retrieves the user profile, historical behavior, business rules, and strategy knowledge corresponding to the current marketing task from the data and knowledge platform, injecting this as global context information into the execution environment of each marketing agent.

[0104] During execution, the execution engine can automatically generate call requests based on the marketing agent's capability declarations. Before the call, it performs type validation and value checks on the input parameters and maps the data generated in the process to standardized semantic objects from the data platform, ensuring that data from different sources has a consistent data structure and semantic meaning before entering the agent. Each marketing agent can understand the current marketing objectives, user status, and business constraints based on a unified knowledge context, thus ensuring semantic consistency in the intermediate results generated by different marketing agents. For multiple nodes requiring parallel execution, the execution engine can create independent running instances and achieve concurrent calls under the constraints of a shared context. For nodes that depend on previous results, it completes sequential execution and state transfer based on inheriting the output semantics of the previous nodes. If the process includes marketing agents from an external business platform, the execution engine can complete authorization verification through a cross-domain capability collaboration protocol and transmit standardized global semantic context information, ensuring consistent understanding of the marketing task across different marketing agents.

[0105] For example, when the system receives a marketing request of "generating a new product promotion plan for high-value users", the execution engine first obtains the portrait features, historical purchase behaviors, and existing marketing strategy constraints of the target user group from the unified data and knowledge middleware, and encapsulates the above data and knowledge into standardized global semantic context information. The global semantic context information is injected into each scheduled marketing agent instance during the process execution as its shared input environment during execution. The user segmentation agent in the process completes user screening based on the global semantic context information. The content generation agent generates marketing copywriting on the premise of inheriting user preferences and brand specifications. The channel recommendation agent gives placement suggestions by combining historical conversion data and business rules. Finally, the result summary node integrates the outputs of multiple agents to generate a structured marketing response result and returns it to the upper-layer application. Through the unified data semantics and knowledge context, multiple agents can form a coherent, consistent, and business-goal-compliant final marketing output during the collaborative execution process.

[0106] The full-link monitoring module supporting the execution engine can be used to record the key events and metrics during the entire process operation. This module automatically captures the node ID, call chain identifier, execution time consumption, return status, and partial structural summaries of input and output before and after each agent call, and forms a complete execution path for these information in chronological order. The monitoring data is collected in real time to the system's monitoring center, enabling users to track and trace the process execution. If the system detects that the time consumption of a certain node is abnormal, there are frequent errors, or the return value structure does not conform to the declared format, the monitoring module can immediately trigger an alarm and mark the abnormal state of the node for the operation and maintenance personnel to handle in a timely manner. After the process execution ends, the monitoring module can write the execution results and diagnostic information into the knowledge middleware for subsequent optimization of the policy model, scheduling strategy, or agent declaration, thereby forming a continuous improvement mechanism driven by execution feedback.

[0107] By combining the collaborative execution engine and the full-link monitoring module, while maintaining high flexibility and orchestratability, it is possible to achieve runtime observability and controllability, enabling the collaborative process between agents to not only be automatically executed but also be continuously monitored, analyzed, and optimized.

[0108] (5)The functions of the agent capability federation and cross-domain collaboration architecture include breaking through organizational boundaries to build an enterprise-level intelligent capability network, and realizing the sharing and collaboration of agent capabilities across business domains while ensuring security and compliance, maximizing the utilization efficiency of enterprise intelligent resources.

[0109] The Intelligent Agent Capability Federation and Cross-Domain Collaboration Architecture, as a top-level control system for enterprise-level intelligent resource sharing, aims to break down the long-standing data barriers and capability silos between different business platforms, enabling the secure, controllable, and auditable flow of a large number of high-value intelligent agent resources accumulated within each marketing business platform across the group organization. This architecture constructs a logical-level federation center, allowing each marketing business platform to access its own marketing intelligent agent capabilities as services through a lightweight registration mechanism without deep direct integration. Standardized capability description protocols enable cross-platform discoverability, matching, and invocation. The federation center uniformly maintains cross-domain capability catalogs, version information, quality indicators, and service availability status, thus forming a group-level intelligent agent capability index system.

[0110] To ensure that cross-domain calls meet security, compliance, and controllability requirements during transmission between different marketing business platforms, this embodiment constructs a multi-layered, end-to-end access control and data protection system within the intelligent agent capability federation architecture of the federation center. Regarding data protection, this embodiment introduces deeply configurable field-level anonymization and usage restriction mechanisms.

[0111] To further meet auditing and compliance requirements, this architecture establishes a full-chain call tracking mechanism for cross-domain calls at the federation layer. Each stage of the call (such as capability discovery, authorization issuance, data anonymization, capability invocation, and result return) is recorded traceably in the call ledger. This call ledger can be used not only for post-event security auditing but also as input to risk control strategy models to identify abnormal call behavior, such as high-frequency calls, unauthorized attempts, or requests from abnormal sources. When necessary, the system can dynamically interrupt call sessions based on real-time audit signals, thereby preventing the further spread of potential attacks or data leakage risks. Through these multi-layered security mechanisms, this embodiment enables the secure flow of cross-domain intelligent agent capabilities within an enterprise group, ensuring that the reuse of intelligent capabilities is both highly efficient and collaborative, while also meeting stringent compliance and data protection requirements.

[0112] In the marketing intelligence agent operation platform system provided in this embodiment, the unified data and knowledge platform can be used to uniformly access, standardize, and extract knowledge from multi-source heterogeneous data, and build a unified data model and knowledge supply system; the marketing intelligence agent market and standardization protocol can be used to realize the unified registration, verification, and standardized management of intelligence agent capabilities, and define unified interface protocols and data exchange specifications; the visual workflow orchestrator can be used to provide a visual interactive interface, support the construction of marketing processes through drag and drop, and generate an executable process description model; the multi-agent collaborative execution engine and full-link monitoring module can be used to execute the visual workflow, realize the collaborative scheduling of multi-agents and full-link operation monitoring; the intelligence agent capability federation and cross-domain collaborative architecture can be used to realize the registration, discovery, and secure invocation of intelligence agent capabilities across business domains, and support cross-domain collaborative marketing.

[0113] The unified data and knowledge platform includes: a data access module for connecting to multi-source heterogeneous data sources and performing data standardization and alignment based on an entity-event model; a structured processing module for performing data cleaning, feature generation, and primary key reconstruction operations, and maintaining feature version management; a procedural knowledge extraction component for converting the marketing execution process into structured knowledge fragments and storing them; and a knowledge supply interface for providing unified knowledge access services to upper-layer intelligent agents.

[0114] The marketing intelligence agent market and standardization protocol includes: a capability declaration and verification module, used to define intelligence agent capabilities through structured declaration files and perform automated verification; a unified interface protocol module, used to define standardized communication specifications for intelligence agent calls, including request body construction and response body parsing; and a data exchange protocol module, used to define standard data objects and semantic mapping rules in the marketing domain to ensure structural consistency of data interaction.

[0115] The visual workflow orchestrator includes: a visual canvas interaction module, which provides a drag-and-drop interface and supports the layout and connection of agent nodes; a process description model generation module, which converts the visual process into an executable structured process description; and a process verification and version management module, which verifies the executability of the process and supports version management.

[0116] The multi-agent collaborative execution engine and end-to-end monitoring module include: a process scheduling execution engine, used to parse the process description model and schedule the execution of agent nodes; a parallel control module, used to manage the resource allocation and execution coordination of parallel nodes; and an end-to-end monitoring module, used to collect execution metrics in real time and build call chain records.

[0117] The intelligent agent capability federation and cross-domain collaboration architecture includes: a federated capability registration and discovery mechanism, used to achieve unified registration and semantic matching discovery of cross-domain intelligent agents; a cross-domain secure call protocol, used to ensure call security through dynamic authentication and authorization and secure data transmission; and a full-link audit and tracing module, used to record the cross-domain call process and perform security analysis.

[0118] The cross-domain secure call protocol adopts a zero-trust security architecture, generates a temporary access token for each cross-domain call, and automatically performs field-level security processing based on the data sensitivity level.

[0119] The marketing intelligence operation platform system also includes a process version management mechanism, which supports saving, comparing, and rolling back versions of marketing processes.

[0120] This application also provides a marketing agent operation method that supports cross-domain federation and visual orchestration, including the following steps: building a unified data model and knowledge supply system through a unified data and knowledge platform; achieving unified governance of agent capabilities through a marketing agent market and standardized protocols; achieving visual construction of marketing processes through a visual workflow orchestrator; executing and monitoring marketing processes through a multi-agent collaborative execution engine and a full-link monitoring module; and achieving cross-business domain agent capability sharing through agent capability federation and cross-domain collaborative architecture.

[0121] The sharing of intelligent agent capabilities across business domains through intelligent agent capability federation and cross-domain collaborative architecture includes: unified registration and catalog management of cross-domain intelligent agent capabilities through a federation center; cross-domain capability discovery and recommendation based on semantic matching; ensuring cross-domain call security through a dynamic authentication and authorization mechanism; and protecting sensitive information through field-level data anonymization and encryption.

[0122] This application provides a marketing agent operation platform system supporting cross-domain federation and visual orchestration, belonging to the fields of artificial intelligence and digital marketing technology. The system comprises five modules: a unified data and knowledge platform for construction and collaborative operation; a marketing agent marketplace and standardized protocols; a visual workflow orchestrator; a multi-agent collaborative execution engine and end-to-end monitoring module; and an agent capability federation and cross-domain collaborative architecture. It achieves unified governance and knowledge supply of multi-source data through the unified data and knowledge platform; standardizes the registration and unified scheduling of agent capabilities through the marketing agent marketplace and standardized protocols; enables the visual construction of business processes through the visual workflow orchestrator; ensures the reliable operation of complex processes through the multi-agent collaborative execution engine; and achieves secure sharing of capabilities across business domains through the agent capability federation architecture. This embodiment enables unified governance and visual orchestration of enterprise marketing capabilities, ensuring efficient and reliable multi-agent collaborative execution, and breaking through business barriers while ensuring security and compliance, thereby forming an orchestratable, collaborative, and evolvable intelligent marketing operation system.

[0123] In one embodiment, this application provides a unified data and knowledge platform, which is the unified data and knowledge platform included in the marketing intelligence agent operation platform system described in the foregoing embodiments.

[0124] In one embodiment, this application provides a visual workflow orchestrator, which is the visual workflow orchestrator included in the marketing intelligence operation platform system described in the foregoing embodiments.

[0125] In one embodiment, this application provides an execution engine, which is the execution engine included in the marketing intelligence agent operation platform system described in the foregoing embodiments.

[0126] In one embodiment, this application provides a federated center, which is the federated center included in the marketing intelligence agent operation platform system described in the foregoing embodiments.

[0127] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0128] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this application. The above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A marketing intelligence agent operation platform system, characterized in that, The system includes a unified data and knowledge platform, a visual workflow orchestrator, an execution engine, and a federation center. The Federation Center is used to register marketing agents from different marketing business platforms; each of these marketing business platforms has a corresponding execution engine deployed. A visual workflow orchestrator is used to display the capability components of registered marketing agents, responding to user operations on the layout of marketing process nodes based on capability components, and generating a marketing process graph. The execution engine is used to obtain global semantic context information based on marketing data and marketing knowledge corresponding to user marketing requests obtained from the unified data and knowledge platform; when executing the current process node of the marketing process graph, if the local marketing agent does not meet the task requirements of the current process node, it initiates a cross-domain call authorization request and capability discovery request to the federation center. The federation center is used to generate temporary access tokens and identify target external marketing agents that are compatible with the current process node in response to cross-domain call authorization requests and capability discovery requests. The execution engine is used to invoke the target external marketing agent based on a cross-domain invocation request generated by the temporary access token and global semantic context information. The federation center is used to perform security protection processing on the response data received from the target external marketing intelligence agent in response to the cross-domain call request and then feed it back to the execution engine. The response data, after security protection processing, is used by the execution engine to generate a marketing response result corresponding to the user's marketing request.

2. The system according to claim 1, characterized in that, The execution engine is used for: If the capability corresponding to the current process node does not belong to the local marketing business platform, or if the capability pool of the local marketing business platform does not have a matching available marketing agent capability, then it is determined that the local marketing agent does not meet the task requirements of the current process node.

3. The system according to claim 1, characterized in that, The execution engine is used to generate a local call request based on the global semantic context information when the local marketing agent satisfies the task requirements of the current process node when executing the current process node of the marketing process graph. The execution engine is used to invoke the matching local marketing agent according to the local call request, obtain the response data of the local marketing agent, and generate a marketing response result corresponding to the user's marketing request.

4. The system according to claim 1, characterized in that, When the federation center identifies a target external marketing agent that is compatible with the current process node, it is specifically used for: Obtain the task semantics of the current process node, and determine several candidate marketing agent capabilities in the federated capability library that meet the similarity conditions with the task semantics; the federated capability library includes registered marketing agent capabilities. Among several candidate marketing agent capabilities, identify the candidate marketing agent capabilities that are associated with the current business scenario tags; Based on the capabilities of candidate marketing agents associated with the current business scenario tags, determine the target external marketing agent that is compatible with the current process node.

5. The system according to claim 1, characterized in that, The temporary access token expires after a cross-domain call.

6. The system according to claim 1, characterized in that, The local marketing agent is the marketing agent owned by the marketing business platform that deploys the execution engine.

7. A unified data and knowledge platform, characterized in that, The unified data and knowledge platform is the unified data and knowledge platform included in the marketing intelligence agent operation platform system as described in any one of claims 1 to 6.

8. A visual workflow orchestrator, characterized in that, The visual workflow orchestrator is the visual workflow orchestrator included in the marketing intelligence operation platform system according to any one of claims 1 to 6.

9. An execution engine, characterized in that, The execution engine is the execution engine included in the marketing intelligence agent operation platform system according to any one of claims 1 to 6.

10. A federal center, characterized in that, The federated center is the federated center included in the marketing intelligence agent operation platform system as described in any one of claims 1 to 6.

Images