A Trustworthy Intelligent Agent Generation Method and System for Enterprise-Wide Digital Marketing

Abstract

The application provides a trusted intelligent agent generation method and system for enterprise whole-process digital marketing, and the method comprises the following steps: acquiring marketing execution process data and user behavior data of an enterprise, and constructing a marketing process graph and a user state machine; constructing a marketing intelligent agent structure based on the marketing process graph and the user state machine, and generating a legal action set; scheduling and executing the legal action set of the marketing intelligent agent structure to generate an actually executed successful action set and a corresponding scheduling score; and structurally recording and auditing and tracking the actually executed successful action set. The application solves the core problems of uncontrollability, uncheckable behavior and unknown state of the intelligent agent of a traditional system.

Description

Technical Field

[0001] This invention belongs to the field of trusted intelligent agent generation, and in particular relates to a trusted intelligent agent generation method and system for enterprise-wide digital marketing. Background Technology

[0002] As enterprises deepen their investment in digital transformation, marketing systems are gradually evolving from manual execution models based on experience and rules to intelligent processes driven by behavioral data and models. However, most current mainstream intelligent marketing platforms focus on automating specific modules, such as pushing content based on tag systems, distributing discounts based on conversion models, or generating recommendations based on user profiles. While they possess some automated decision-making capabilities, they lack the ability to structurally model the entire marketing process and manage intelligent agents at the system level. In enterprise marketing chains with multiple touchpoints and concurrent strategies, systems often fail to dynamically organize intelligent agent behavior according to the structure of task flows and the evolution of user states, leading to frequent occurrences of uncontrolled behavior triggering, content mismatch, and strategy conflicts in the marketing process. Especially when the same user is reached by multiple systems in parallel, current marketing intelligent agents generally lack in terms of behavior generation boundaries, execution control strategies, and behavior result recording. They lack complete task chain definitions and state awareness mechanisms internally, and lack reliable behavior audit channels externally, failing to support enterprises in accurately managing, reconstructing, and reproducing marketing behaviors in a compliant environment. These issues ultimately result in a lack of credibility and flexibility in the deployment of intelligent marketing systems, making it difficult to meet enterprises' real needs for "precise, transparent, and manageable" digital marketing systems. Summary of the Invention

[0003] The purpose of this invention is to propose a method and system for generating trusted intelligent agents for enterprise-wide digital marketing, thereby solving the aforementioned problems.

[0004] To achieve the above objectives, a method for generating a trusted intelligent agent for enterprise-wide digital marketing is provided in a first aspect of the present invention, the method comprising the following steps:

[0005] Acquire enterprise marketing execution process data and user behavior data, and construct a marketing process diagram and a user state machine; wherein, the marketing process diagram contains several process nodes, and the process nodes are bidirectionally mapped to one or more user state machines;

[0006] Based on the aforementioned marketing flowchart and user state machine, a marketing intelligent agent structure is constructed, and a set of legal actions is generated;

[0007] The legitimate action set of the marketing agent structure is scheduled and executed to generate the actual successfully executed action set and the corresponding scheduling score;

[0008] The set of actions that were actually successfully executed is recorded in a structured manner and audited.

[0009] Furthermore, the marketing execution process data consists of process template files and channel strategy data configured in the enterprise marketing platform or automation platform; the user behavior data consists of log structures or tag tables.

[0010] Furthermore, the steps for acquiring enterprise marketing execution process data and user behavior data, and constructing a marketing process diagram and user status include:

[0011] The marketing execution process data is abstracted into a set of process nodes, and each process node represents a specific type of executable task.

[0012] Extract continuous behavior sequences from user behavior data to construct a user state machine and user behavior transition rules;

[0013] Based on the user behavior transfer rules, the process nodes are mapped to one or more user states.

[0014] Furthermore, the process node includes a unique identifier for the task node, the behavioral objective of the task, the set of data fields required to execute the task, the behavioral output format after the task is triggered, and behavioral constraints.

[0015] Furthermore, the steps of constructing a marketing intelligent agent structure based on the marketing flowchart and user state machine, and generating a set of legal actions, include:

[0016] The user's current state machine is used to query the set of process nodes that are allowed to be triggered in that state, to ensure that the currently activated process node belongs to the marketing process diagram; otherwise, the generation of a marketing agent is not allowed.

[0017] Once the status query is completed, the marketing intelligent agent structure is constructed based on the rule graph structure, combined with the process nodes and the user state machine.

[0018] For the aforementioned marketing agent structure, an action scoring function is designed to filter the set of legal actions that meet the conditions and the corresponding marketing agent structure.

[0019] Furthermore, the action scoring function is calculated based on the degree of matching between the current marketing agent's behavior and the current task intent, the intent matching reinforcement factor, and the response inhibition coefficient of the user's state machine; the range of the matching degree is given by the rule engine; the intent matching reinforcement factor is fixed to a positive value; and the response inhibition coefficient represents the user's current behavior density or strategy saturation.

[0020] The marketing agent structure includes behavioral roles mapped from the task's behavioral target string, a set of legal actions, user-level context information, and execution constraint boundaries.

[0021] Furthermore, the scheduling and execution of the legal action set of the marketing agent structure to generate the actually successfully executed action set and the corresponding scheduling score includes the following steps:

[0022] Register the marketing intelligence agent structure with the scheduling and control center, generate a unique identifier for each task node and bind it to the user ID;

[0023] Accept the set of legal actions, calculate the scheduling score of the set of legal actions, and determine whether it is most worthwhile to execute in the current context;

[0024] For the set of legal actions, after calculating the scheduling score, the actions in the set of legal actions are sorted in descending order of score, and the actions with scheduling scores higher than the threshold are selected for deployment and execution; wherein, each action is encapsulated into a task request by the scheduler and calls different enterprise channel service APIs to complete the action push.

[0025] After execution, the system generates a structured execution record for each successful / failed action, resulting in a set of actions that were actually successfully executed.

[0026] Furthermore, the scheduling score is calculated based on the channel availability score, action timeliness value, current channel congestion level, and behavior content similarity penalty factor;

[0027] The channel availability score is derived from marketing execution process data; the action timeliness value is used to measure whether the current action is within an effective execution window; the current channel congestion level indicates the number of tasks currently queued in the channel to which the current action is bound; the behavior content similarity penalty factor is used to measure the similarity between the user's most recently received action and the current action;

[0028] If the behavior content similarity penalty factor is closest to 1, it indicates that the action repetition is high, and the scheduling priority should be reduced to the minimum.

[0029] Furthermore, each action in the set of actions that were actually successfully executed includes a user ID, the identifier of the corresponding action, the execution channel, the reach timestamp, and the execution result;

[0030] Then, the steps for structurally recording and auditing the set of actions that were actually successfully executed include:

[0031] For each action in the set of actions that have been successfully executed, a structured behavior path entry is constructed; wherein, the structured behavior path entry includes user ID, identifier of the corresponding action, unique identifier of task node, corresponding user state machine, marketing agent structure to which it belongs, execution channel, reach timestamp, and execution response code;

[0032] The structured behavior path entries are then converted into structured objects and stored in the behavior log database.

[0033] The system configures a persistence module to write each structured behavior path entry into a log table and generates a behavior path representation in standard JSON format.

[0034] A second aspect of the present invention provides a trusted intelligent agent generation system for enterprise-wide digital marketing, the system comprising:

[0035] The process and state modeling module is used to acquire enterprise marketing execution process data and user behavior data, and construct marketing process diagrams and user state machines; wherein, the marketing process diagram contains several process nodes, and the process nodes are bidirectionally mapped to one or more user state machines;

[0036] The agent generation module is used to construct a marketing agent structure based on the marketing flowchart and user state machine, and generate a set of legal actions;

[0037] The behavior scheduling module is used to schedule and execute the set of legal actions of the marketing agent structure, and generate the set of actions that are actually successfully executed and the corresponding scheduling score;

[0038] The behavior audit module is used to structure and audit the set of actions that were actually successfully executed.

[0039] The beneficial technical effects of the present invention are at least as follows:

[0040] This invention proposes a structured marketing agent system method that spans "process modeling—state awareness—agent generation—behavior scheduling—execution recording." Addressing issues in existing systems such as fragmented processes, disconnected states, conflicting behaviors, and unauditable logs, it establishes a multi-stage agent generation mechanism based on task flowcharts and user behavior state machines as inputs. A dynamic behavior filtering method based on context and historical behavior density is also designed, enabling the agent to not only structurally possess behavioral boundary control capabilities but also to have rationality and interference awareness capabilities at the behavior scheduling layer. In the execution phase, this invention significantly improves the quality of behavior selection in multi-agent concurrent scenarios by introducing a composite scheduling function that integrates channel state, time window, and behavior similarity. Furthermore, by recording the structured path of all executed actions, binding process nodes, trigger states, agent origins, and action execution results into auditable logs, each interaction becomes traceable and reconstructable. By organically combining structural modeling, strategy generation, behavior control, and trusted evidence storage, this invention systematically constructs an interpretable, executable, and auditable end-to-end solution for marketing intelligent agents, solving the core problems of uncontrollable, unverifiable, and unknowable states of intelligent agents in traditional systems. Attached Figure Description

[0041] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0042] Figure 1 This is a flowchart of the trusted intelligent agent generation method for enterprise full-process digital marketing according to the present invention. Detailed Implementation

[0043] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0044] like Figure 1 As shown in the embodiment of the present invention, a trusted intelligent agent generation method for enterprise end-to-end digital marketing is provided, the method comprising:

[0045] S1. Obtain enterprise marketing execution process data and user behavior data, and construct a marketing process diagram and user state machine; wherein, the marketing process diagram contains several process nodes, and the process nodes are bidirectionally mapped to one or more user state machines.

[0046] Specifically, this step aims to structurally model the enterprise's marketing execution process and the evolution of user behavior throughout the marketing cycle. By extracting the process structure from the enterprise system and summarizing the historical user behavior in a chronological order, a set of contextual information is formed to control the generation and execution of the marketing agent. The flowchart represents the enterprise's pre-defined task sequence and logical boundaries, while the state machine represents an abstract description of the user's current behavioral stage. This structure provides subsequent agents with clear behavioral intentions, contextual input requirements, and triggering conditions.

[0047] The input consists of two existing data sources within the enterprise marketing system:

[0048] The process template files and channel strategy data configured in the enterprise marketing platform or automation platform (such as Alibaba Cloud Marketing Platform, Salesforce Marketing Cloud, etc.). This data is stored in a structured form and is generally accessed through APIs provided by the platform (such as REST interfaces), or it can be obtained by reading the "process configuration table" and "execution path log" in the system database.

[0049] User behavior data logs are typically stored in an enterprise's data warehouse or user profiling system, in the form of log structures (such as behavior IDs, timestamps, page IDs, etc.) or tag tables. Data is generally collected in real-time through event tracking systems (such as Alibaba DataV, GrowingIO, etc.), or can be accessed from a Kafka streaming consumption module.

[0050] Furthermore, the enterprise's marketing process is first structurally modeled by analyzing the predefined execution paths in its automation platform. The system abstracts marketing activities into a set of process nodes, each representing a specific executable task. For example, node n1 could represent "recommending products to first-time login users," n2 could represent "triggering coupons for users who browse three times without purchasing," and n3 could represent "returning visits from existing users." These nodes are extracted from existing task configuration fields in the system, such as task ID, reach channel, API call, and push content type. The output flow of each node is determined by the enterprise-defined process sequence diagram, typically configured in JSON or YAML format, which includes pre-dependencies, post-jump conditions, and failure fallback mechanisms. The system parses this into a directed graph G. p =(N p E p ), where N p It is a set of nodes in the process, E p It refers to the connection relationships between nodes.

[0051] Furthermore, in order to constrain the execution scope of each node, the present invention defines the following structure for each node:

[0052] ni = <id i intent i input i ,output i ,boundary i > (1)

[0053] Where: id i : A unique identifier for the task node, derived from the task configuration field of the enterprise marketing automation platform; intent i The task's behavioral objective, such as "recalling inactive users"; input i : The set of data fields required to perform the task, such as "User ID", "Browsed Product ID", "User Tag Value", etc., with data sourced from the enterprise's user behavior database; output i The output behavior after this task is triggered, such as "send notification" or "generate a list of recommended products," is implemented through an API call within the marketing system; boundary i Behavioral restrictions, such as "SMS channel only" and "execute at most once per day", are derived from the process configuration constraint fields.

[0054] Furthermore, a user state machine is established to complement the flowchart. The system extracts continuous behavior sequences from user behavior logs, such as: register → browse → click → add to cart → purchase → remain silent. This behavior sequence is mapped to a set of finite states Q according to time windows and behavior rules, for example:

[0055] q1: First visit, no clicks;

[0056] q2: No conversion after more than 2 views;

[0057] q3: Within 3 days of the first purchase;

[0058] q4: Inactive for 30 days.

[0059] These state transitions are described by state transition functions T. For example, when a user is in state q1 and has not made any clicks within 72 hours, they transition to q5 (potential churned user). State definitions and transition rules are configured by the behavior rule engine. Enterprises can set the judgment logic in the management backend, and the rules are persisted in the policy database via the rule engine's DSL. The system calculates the user's current state q daily, either periodically or in real-time. curr For downstream use.

[0060] The state transition relation can be formalized as:

[0061]

[0062] Where: qj : The user's current state, derived from the most recent state machine calculation result; T: User behavior transition rules, such as "not clicking on any product for 3 consecutive days"; q j+1 The new state after the user's state transition.

[0063] To establish context binding, this invention will bind each process node n in the flowchart. i With one or more user states q j Mapping is performed. This mapping is configured by the enterprise rule setter based on the marketing strategy, for example: "When a user is in state q2, i.e., browsing multiple times without purchasing, process node n2 should be triggered, and a limited-time coupon should be pushed." The system will generate mapping rules. j ,n i > It serves as an instruction for "what should be done now" in the subsequent generation of the marketing intelligent agent structure.

[0064] S2. Construct a marketing intelligent agent structure based on the marketing flowchart and user state machine, and generate a set of legal actions.

[0065] Specifically, in this step, the system is based on the marketing process diagram G constructed in the previous step. p =(N p E p ) and user state machine S u = (Q, T), for each user at a specific marketing node n i Its current state q curr Generate a unique, structured marketing intelligence agent A i,j This agent possesses clear behavioral intent, contextual input, compliance boundaries, and dynamic behavior control capabilities. Its generation process integrates process objectives, behavioral timing, input data characteristics, and system execution constraints. In the patented scenario, this step is a crucial hub connecting enterprise tasks and user behavior decisions within the entire system, determining whether the marketing intelligence agent can make the right marketing response to the right user at the right time, in the right way, and to the right audience.

[0066] Furthermore, the generation process of the marketing agent begins with structural logic judgment. The system determines the user's current state q based on this logic. curr Query the set N of process nodes that are allowed to be triggered in this state. q Ensure that the currently active n i Belongs to N q Otherwise, Agent generation is not allowed. This check ensures that the system does not send incorrect policies, such as "send a welcome gift to churned users".

[0067] Once the state validity check is passed, the system will... i and q curr ​Constructing Agent structure A under dual conditions i,j The generation process of this structure is not template replacement, but construction based on a rule graph structure, and each field has a clear source control.

[0068] The behavior generation part adopts an "intent-state" matching structure. This invention introduces a creative correction strategy specifically for the scenario described in this patent: a response saturation adjustment factor α(q) is introduced into the user behavior state weight. curr This coefficient is used to suppress negative behavioral effects caused by frequent user interactions. For example, if a user has been activated by 3 nodes within 48 hours, this coefficient will reduce the action weight to prevent the generation of disruptive actions.

[0069] The specific action scoring function is as follows:

[0070] score(a k )=β·relevance(a k intent i )-λ·α(q curr (3)

[0071] Where: score(a k ): Candidate behavior action a k The score in the current Agent; relevance(a k intent i ): The degree to which this action matches the current task intent. The value range is given by the rule engine and can be configured manually or fitted by a statistical model; β: Intent matching reinforcement factor, fixed as a positive value; α(q) curr ): For state q curr The response inhibition coefficient represents the current behavior density or policy saturation of the user, and is derived from the system behavior count table; λ: the weight of the penalty item, which controls the intensity of the penalty.

[0072] The system calculates scores for all candidate actions using this function, and selects the set of actions with scores greater than a threshold θ to form the agent's actions. i,j The threshold is configured uniformly by the system policy.

[0073] Other fields generated by the Agent come from the following processing:

[0074] role i,j By intent i It is generated by mapping and matching a preset behavioral intent dictionary, such as "recommendation", "reminder", and "conversion";

[0075] context i,j Then from input iDetermine the field name and extract specific values ​​from the data platform by user ID, such as product browsing history, most recent payment time, etc.

[0076] boundary i Maintain the existing practices without making changes, ensuring that the agent's behavior does not exceed the limits allowed by the system.

[0077] The complete structure is defined as follows:

[0078] A i,j = <role i,j actions i,j ,context i,j ,boundary i > (4)

[0079] Among them: role i,j From intent i The behavior roles of string mapping are controlled by a manually preset dictionary; actions i,j : The set of legal actions selected by a scoring function, with filtering criteria including intent matching and state constraints; context i,j User-level context information is extracted from real-time profiling platforms (such as Doris and ClickHouse); boundary i : Execution limit boundaries, derived from the task configuration file, no reasoning required.

[0080] The innovation of this agent generation logic lies in its introduction of a state response inhibition factor as a behavior regulation mechanism, enabling intelligent adjustment of the user reach pace without using predictive models. This mechanism, combining state, action density, and policy rules, allows the generated agent to more naturally achieve behavior control in enterprise marketing scenarios, avoiding problems such as excessive intervention and policy conflicts, thus improving the system's practical usability and intelligence.

[0081] Final output: For process node n i With state q curr Jointly generated intelligent agent structure A i,j Actions that have been filtered and controlled i,j It can then directly proceed to the subsequent scheduling and execution process.

[0082] S3. Schedule and execute the set of legal actions of the marketing agent structure to generate the set of actions that were actually successfully executed and the corresponding scheduling score.

[0083] Specifically, in this step, the system will generate the structured intelligent agent A in step S2. i,j = <role i,j actionsi,j ,context i,j ,boundary i >Actually deployed to the marketing execution system, and its actions i,j Perform scheduling and execution.

[0084] Furthermore, the deployment phase will first include A i,j Register with the scheduling control center, generate a unique task ID, and bind it to the user ID. This operation is completed by the task scheduling engine (deployed in the marketing middle platform). The system then initiates the behavior scheduling process, processing each action a... k ∈actions i,j It performs operations such as channel identification, scheduling timing assessment, frequency conflict filtering, and historical behavior overlap detection.

[0085] The key innovation of this step lies in proposing a composite scheduling priority scoring function S″(a) that combines user state response strength, channel timeliness, and behavioral content similarity. k This scoring function not only determines whether an action can be executed, but more importantly, whether it is the most worthwhile action to execute in the current context, thereby ensuring system resource efficiency and user experience.

[0086] Furthermore, the composite scoring function is defined as follows:

[0087] S″(a k )=η1 · C(a k )+η2 · V(a k )-η3 · L(a k )-η4 · R(a k ,u) (5)

[0088] Where: C(a) k ): Channel availability score, derived from the enterprise channel status monitoring module. For example, the higher the remaining quota of the SMS channel, the higher C(a) score. k The larger V(a) is; k ): The timeliness value of an action, used to measure a k Whether it is in a valid execution window depends on n i Time strategy configuration; L(a k ): The current congestion level of the channel, representing a k The higher the number of tasks currently queued on the bound channel, the higher the scheduling risk; R(a kη1, η2, η3, η4: Weighting factors, uniformly set by the enterprise strategy system, varying according to task type (e.g., activation, notification); S″(a k () is the scheduling score.

[0089] Specifically, R(a) k The key innovation of this step is to solve the problem of "redundant content conflicts in multi-agent behavior". Its definition is as follows:

[0090]

[0091] Wherein: H u This represents the set of m most recent actions accepted by user u, maintained by the system in the user log database; sim(a k R(a′) represents the semantic similarity score, which can be calculated by extracting keywords from the behavioral content and then using TF-IDF or vector cosine similarity; if R(a′) = ... k If u) is close to 1, it indicates that the repetition of actions is high, and the scheduling priority should be reduced.

[0092] After the system calculates S″(a) k After that, actions i,j Actions are sorted in descending order of score, and those with scores above a threshold θ are selected for deployment. Each action is encapsulated into a task request by the scheduler, which calls different enterprise channel service APIs to complete the behavior push, such as calling the SMS service platform, APP push SDK, and mini-program subscription interface.

[0093] After execution, the system generates a structured execution record for each successful / failed action, including task ID, user ID, action type, channel, scheduling time, response code, etc. All these structured records will be used for trusted evidence storage in step four.

[0094] Final output: The set of actions that were actually executed successfully (actions) exec It includes user dimensions, channel dimensions, execution time, and response status; the scheduling score S″(a) for each action. k (and the final sorting index, used for behavior auditing and strategy optimization analysis.)

[0095] S4. The set of actions that were actually successfully executed is recorded in a structured manner and audited.

[0096] Specifically, the core task of this step is to process the marketing actions that have been successfully scheduled and executed in the previous step. execStructured recording is performed, and an auditable set of behavioral paths is constructed with "process node - user state - agent - behavior action" as the dimensions. This behavior path is not only used for logging marketing activities, but also for subsequent accountability, process reproduction, effectiveness evaluation, and strategy optimization. Unlike traditional systems that record "reach logs," this patent proposes a context-enhanced behavior path structure that can retain the agent's generation conditions, triggering states, behavioral semantics, and scheduling history, providing stronger interpretability and compliance support capabilities.

[0097] This step strictly uses the two outputs from step three:

[0098] Actions exec Each a k It includes the user IDu, the identifier of the corresponding action, the execution channel, the arrival timestamp, and the execution result, which comes from the execution scheduling system;

[0099] The corresponding action belongs to the agent structure A i,j During the deployment phase, each action is uniquely mapped to the process node n upon which it was generated. i With user state q curr .

[0100] Furthermore, the system provides each a k ∈actions exec Constructing structured behavioral paths k Each behavior path contains the following fields:

[0101] User ID: u;

[0102] Action identifier: a k ;

[0103] Related process node: id i ;

[0104] Corresponding user status: q curr ;

[0105] Belonging Agent:

[0106] Execution channel: ch k ;

[0107] Reaching timestamp: t k ;

[0108] Execution response code: resp k ;

[0109] To ensure the path has full-chain traceability within the system, the system stores the behavior path as a structured object and places it in the behavior log database. The formal definition is as follows:

[0110]

[0111] in: The unique ID assigned by the system when the Agent structure is generated; id i Process node n from Agent i ;q curr It is the execution state recorded in the user behavior state system; a k It is a structured action object, including "type + template + content reference"; ch k It refers to the channel type, such as "SMS" or "APP_PUSH"; t k It is a Unix timestamp or ISO time standard format; resp k It is the channel return code, indicating whether the execution was successful (such as 200, 503, etc.).

[0112] Furthermore, the system configures a persistence module (such as a distributed table like Doris or HBase) to store each row... k Each record in the log table is uniquely bound to a user, agent, process node, and action. This table supports searching by any field, enabling:

[0113] View all historical reach paths of a user;

[0114] Track conversion rates by process node;

[0115] Evaluate the actual output of the agent according to the agent dimension;

[0116] Aggregation behavior quality and success rate by channel.

[0117] To improve the readability of the path model in the auditing system, the system also generates a behavior path representation in standard JSON format:

[0118] {

[0119] "user_id":"u001245",

[0120] "agent_id":"A_n32_u001245",

[0121] "flow_node":"n32",

[0122] "user_state":"q_5",

[0123] "action_type":"SMS_COUPON",

[0124] "channel":"SMS",

[0125] "exec_time":"2025-08-04T14:03:23Z",

[0126] "response":200

[0127] }

[0128] The above indicates that user u001245, at process node n32 and state q5, successfully sent an SMS coupon via AgentA_n32_u001245.

[0129] This invention also provides a trusted intelligent agent generation system for enterprise-wide digital marketing, the system comprising:

[0130] The process and state modeling module is used to acquire enterprise marketing execution process data and user behavior data, and construct marketing process diagrams and user state machines; wherein, the marketing process diagram contains several process nodes, and the process nodes are bidirectionally mapped to one or more user state machines;

[0131] The agent generation module is used to construct a marketing agent structure based on the marketing flowchart and user state machine, and generate a set of legal actions;

[0132] The behavior scheduling module is used to schedule and execute the set of legal actions of the marketing agent structure, and generate the set of actions that are actually successfully executed and the corresponding scheduling score;

[0133] The behavior audit module is used to structure and audit the set of actions that were actually successfully executed.

[0134] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0135] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of apparatuses or units may be electrical, mechanical, or other forms.

[0136] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0137] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for generating a trusted intelligent agent for enterprise whole-process digital marketing, characterized in that, The method comprises the following steps: Obtain marketing execution process data and user behavior data of an enterprise, and construct a marketing process graph and a user state machine; wherein the marketing process graph comprises a plurality of process nodes, and the process nodes are bidirectionally mapped with one or more user state machines; Build a marketing intelligent agent structure based on the marketing process graph and the user state machine, and generate a legal action set; Schedule and execute the legal action set of the marketing intelligent agent structure to generate an actually executed successful action set and a corresponding scheduling score; Structurally record and audit trace the actually executed successful action set; The execution step of building a marketing intelligent agent structure based on the marketing process graph and the user state machine, and generating a legal action set comprises: According to the current state machine of the user, query the set of process nodes allowed to be triggered by the state to ensure that the currently activated process node belongs to the marketing process graph, otherwise the marketing intelligent agent is not allowed to be generated; Once the state query is passed, based on the rule graph structure, the marketing intelligent agent structure is constructed in combination with the process node and the user state machine; Design an action scoring function for the marketing intelligent agent structure to screen a legal action set that meets the conditions and a corresponding marketing intelligent agent structure; The action scoring function is calculated based on the matching degree of the behavior action of the current marketing intelligent agent structure and the current task intention, the intention matching reinforcement factor, and the response inhibition coefficient of the user state machine; the matching degree is given by a rule engine; the intention matching reinforcement factor is fixed as a positive value; and the response inhibition coefficient represents the current behavior density or strategy saturation of the user; The marketing intelligent agent structure comprises a behavior role mapped from a behavior target string of a task, a legal action set, user-level context information, and an execution limit boundary; The execution step of scheduling and executing the legal action set of the marketing intelligent agent structure to generate an actually executed successful action set and a corresponding scheduling score comprises: Register the marketing intelligent agent structure to a scheduling control center to generate a task node unique identifier and bind it with a user ID; Accept the legal action set, calculate the scheduling score of the legal action set to determine whether it is most worthy of execution under the current context; For the legal action set, after calculating the scheduling score, arrange the behavior actions of the legal action set in descending order of score, and select behavior actions with a scheduling score higher than a threshold value for deployment and execution; wherein each action is encapsulated as a task request by a scheduler, and different enterprise channel service APIs are called to complete behavior pushing; After execution, the system generates a structured execution record for each successful / failed behavior action to generate an actually executed successful action set; The scheduling score is calculated according to a channel availability score, an action time value, a current channel congestion degree, and a behavior content similarity penalty factor; The channel availability score is derived from marketing execution process data; the action timeliness value is used to measure whether the current behavior action is in an effective execution window; the channel current congestion degree represents the number of queued tasks of the channel to which the current behavior action is bound; and the behavior content similarity penalty factor is used to measure the similarity between the behavior action recently received by the user and the current behavior action. If the behavior content similarity penalty factor is closest to 1, it indicates that the action repetition degree is high, and the scheduling priority should be reduced to the lowest.

2. The enterprise-oriented whole-process digital marketing trusted intelligent agent generation method of claim 1, wherein, The marketing execution process data is a process template file and channel strategy data configured in an enterprise marketing platform or an automation platform; and the user behavior data is a log structure or a label table.

3. The enterprise-oriented whole-process digital marketing trusted intelligent agent generation method of claim 1, wherein, The marketing execution process data and the user behavior data of the enterprise are obtained, and a marketing process graph and a user state are constructed, and the execution steps include: The marketing execution process data is abstracted into a group of process nodes, and each process node represents a specific executable task; Continuous behavior sequences are extracted from the user behavior data to construct a user state machine and user behavior transition rules; The process nodes are mapped to one or more user states based on the user behavior transition rules.

4. The enterprise-oriented whole-process digital marketing trusted intelligent agent generation method of claim 3, wherein, The process nodes include a task node unique identifier, a behavior target of the task, a set of data fields required for executing the task, a behavior output form after the task is triggered, and a behavior restriction condition.

5. The enterprise-oriented whole-process digital marketing trusted intelligent agent generation method of claim 1, wherein, Each action behavior in the set of actually executed successful actions includes a user ID, an identifier of a corresponding behavior action, an execution channel, a touch time stamp, and an execution result. The set of actually executed successful actions is recorded and audited and tracked in a structured manner, and the execution steps include: A structured behavior path entry is constructed for each action behavior in the set of actually executed successful actions; the structured behavior path entry includes a user ID, an identifier of a corresponding behavior action, a task node unique identifier, a corresponding user state machine, a marketing intelligent agent structure to which the user belongs, an execution channel, a touch time stamp, and an execution response code; The structured behavior path entry is structured as a structured object and is written into a behavior log database; The system writes each structured behavior path entry into a log table through a configuration persistence module, and generates a behavior path representation in a standard JSON format.

6. The system for performing the method of claim 1 for the trusted intelligent agent generation for enterprise-wide digital marketing, characterized in that, The system includes: A process and state modeling module for obtaining marketing execution process data and user behavior data of an enterprise, and constructing a marketing process graph and a user state machine; the marketing process graph includes a plurality of process nodes, and the process nodes are bidirectionally mapped to one or more user state machines; An intelligent agent generation module for constructing a marketing intelligent agent structure based on the marketing process graph and the user state machine, and generating a legal action set; A behavior scheduling module for scheduling and executing the legal action set of the marketing intelligent agent structure, generating a set of actually executed successful actions, and generating a corresponding scheduling score; A behavior auditing module for recording and auditing the set of actually executed successful actions in a structured manner.

Images