Marketing scene-oriented AI agent task scheduling execution method

Abstract

The invention discloses a marketing scene-oriented AI agent task scheduling execution method, and belongs to the technical field of artificial intelligence and digital marketing, and the method comprises the steps: S1, collecting and processing multi-dimensional index data; s2, establishing a task priority evaluation model, calculating comprehensive priority scores of tasks according to the processed multi-dimensional index data, and sorting the scores; s3, constructing a simulation marketing environment, and training the reinforcement learning model to obtain an optimized scheduling strategy; s4, modeling a marketing task, an AI agent and audience elements into nodes, establishing associated edges, and training a graph neural network model to predict task execution time and effect; s5, task allocation is carried out according to the content, and real-time monitoring and adjustment are carried out; according to the marketing scene-oriented AI agent task scheduling execution method provided by the invention, a plurality of AI agents cooperatively work in the marketing scene through an intelligent scheduling algorithm, and various marketing tasks can be rapidly and efficiently completed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence and digital marketing, in particular to an AI agent task scheduling execution method for marketing scenarios. BACKGROUND

[0002] In modern marketing activities, multiple AI agents (such as chat robots, recommendation engines, content generation models, etc.) jointly participate in the activities. However, the existing systems have many problems in task allocation, such as low efficiency, slow response, and overloading of some agents while others are idle, resulting in unreasonable resource allocation. This imbalance not only reduces the timeliness of marketing response and affects user experience, but also causes serious waste of computing resources, which is not conducive to the efficient development of enterprise digital marketing activities and cannot meet the pursuit of high ROI and operational efficiency by enterprises. SUMMARY

[0003] The purpose of the present application is to provide an AI agent task scheduling execution method for marketing scenarios to solve the problems of low efficiency, slow response, and unreasonable resource allocation in existing marketing AI systems, to realize efficient collaboration of multiple AI agents in marketing scenarios, to ensure that important marketing tasks are handled in a timely manner, and to avoid the idling or overloading of resources.

[0004] To achieve the above purpose, the present application provides an AI agent task scheduling execution method for marketing scenarios, comprising the following steps: S1, collecting and processing multi-dimensional index data, including user portrait data, content priority data, and resource state data; S2, establishing a task priority evaluation model, calculating the comprehensive priority score of the task according to the processed multi-dimensional index data and sorting; S3, constructing a simulated marketing environment, determining the state space, action space, and reward signal, training the reinforcement learning model to obtain an optimized scheduling strategy; S4, modeling the marketing task, AI agent, and audience elements as nodes and establishing associated edges, training a graph neural network model to predict task execution time and effectiveness; S5, allocating tasks according to the task priority ranking, reinforcement learning scheduling strategy, and graph neural network prediction results, and real-time monitoring and adjusting.

[0005] Preferably, the user portrait data in step S1 includes collecting the user's browsing records, purchase behavior, search keywords, and dwell time.

[0006] Preferably, step S2 specifically includes: S21, the task priority evaluation module takes the user portrait data, content priority data and resource state data in step S1 as the core dimensions of the input; S22, quantize each dimension indicator input; S23, according to the characteristics of the marketing scene, assign weights to each dimension indicator; S24, calculate the task comprehensive priority score using the weighted sum formula: ; Wherein, is the standardized value of the user portrait data; is the quantized value of the content priority data; is the standardized value of the resource state data; , , are weight coefficients, and ; S25, according to the comprehensive priority score, all tasks are sorted, and the task with the highest score is preferentially entered into the scheduling queue corresponding to the reinforcement learning scheduling strategy training link in S3.

[0007] Preferably, step S3 specifically includes: S31, construct a simulated marketing environment, taking the task comprehensive priority sorting result output in step S2 as input, and construct a simulated environment containing task characteristics, AI agent state and resource constraints. The environment dynamically generates a task flow, and the task attributes include user value coefficient, content urgency and resource demand, and simulates user interaction feedback; S32, determine the core elements of the reinforcement learning model: state space, action space and reward function; S33, use deep Q network architecture and related training mechanism to train the reinforcement learning model, so that it can select the optimal task allocation action strategy in different states in the marketing scene task scheduling; S34, after training convergence, output the optimal action according to the state for the task queue sorted in step S2, form a dynamic scheduling strategy, and ensure that high-priority tasks are preferentially allocated to adaptive agents.

[0008] Preferably, step S32 specifically includes: State space definition: fusion of task priority sorting results in step S2 , real-time load of agent , resource utilization and task context vector , form state vector ; Action space definition: action represents the decision to allocate tasks to agents, and discrete space is defined as: wherein, denotes the sorted task index in step S2; denotes the available agent index; The reward function is set as: ; wherein, is the task execution time on the agent , , , is the weight, is the resource load penalty: ; wherein, is the resource load threshold for judging whether the agent is overloaded.

[0009] Preferably, step S33 is specifically: Deep Q network architecture is adopted, experience replay mechanism is used to store samples, and the target network is used to update the Q value: ; wherein, is the learning rate, is the discount factor, is the main network parameter, is the target network parameter, and the loss function adopts mean square error: The parameters are optimized by gradient descent, the target network parameters are synchronized every N steps, and - Greedy policy balances exploration and utilization, specifically: Initialize exploration rate: set the initial exploration probability and the final exploration probability , and the exploration probability decay rate ; Select action: in each state , generate a random number between 0 and 1, if , randomly select an action from the action space by uniform sampling to explore new task allocation methods; if , select the action that maximizes the Q value predicted by the current Q network, using the learned strategy; Update exploration probability: after each training iteration, update the exploration probability according to the following formula: Reduce, so that the agent uses the learned optimal strategy, while reducing the exploration of new actions.

[0010] Preferably, after the training converges, step S34 ranks the task queue sorted in step S2 according to the state Output optimal action Form a dynamic scheduling strategy to ensure that high-priority tasks are assigned to adaptive agents first.

[0011] Preferably, step S4 specifically includes: S41, integrate nodes and edges into a heterogeneous graph , wherein is the set of all nodes, is the set of all edges; S42, use a graph neural network to aggregate node features, and the feature update formula of each layer is: ; Where, is the feature of node in the first layer, is the neighbor set of node , is the weight of the edge between node and its neighbor node , is the node degree, and are learnable parameters, is an activation function; S43, predict task execution time and effectiveness: For the task-agent pair , the execution time is predicted by the output layer of the graph neural network, and the loss function is the mean square error: ; Where, is the actual execution time, is the number of samples; For the task-agent-audience triple , predict the task conversion rate , and the loss function is cross-entropy: ; Where, is the actual conversion rate.

[0012] Preferably, step S5 constructs an allocation scoring function according to the task priority ranking, reinforcement learning scheduling strategy and graph neural network prediction results, and selects the optimal agent, specifically: For the task and the agent , the score is: ; wherein, is a task comprehensive priority score; is an action value output by the reinforcement learning model; , , is a weight coefficient; to the sorted task queue, thereby selecting the agent with the highest score: wherein, is a set of available agents.

[0013] Therefore, the present application adopts the above-mentioned AI agent task scheduling execution method for a marketing scenario, and effectively solves the problems of low task allocation efficiency, slow response, and unreasonable resource allocation in existing marketing AI systems through multi-dimensional index-driven task priority evaluation, self-adaptive scheduling strategy optimized by reinforcement learning, and intelligent decision-making supported by graph neural networks. It can intelligently coordinate multiple AI agents for efficient operation according to real-time marketing environment and business needs: not only ensures that important marketing tasks are processed in time, but also avoids the idling or overload of resources. In practical applications, it can significantly improve the speed and accuracy of marketing response, improve user experience, and improve the utilization rate of marketing resources, bringing higher ROI and operational efficiency to the enterprise's digital marketing activities.

[0014] The technical solutions of the present application will be further described in detail below through the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a flowchart of the AI agent task scheduling execution method for a marketing scenario of the present application; Figure 2 is a flowchart of the present application for calculating the comprehensive priority score of the task and sorting; Figure 3 is a flowchart of the present application for obtaining an optimized scheduling strategy; Figure 4 is a flowchart of the present application for task execution time and effectiveness prediction. DETAILED DESCRIPTION

[0016] The following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0017] Please refer to Figures 1-4 , the AI agent task scheduling execution method for marketing scenarios, comprising the following steps: S1, collect multi-dimensional index data and process, multi-dimensional index data includes user portrait data, content priority data and resource state data; User portrait data is collected through the existing user behavior tracking system of the enterprise, including user browsing records, purchase behavior, search keywords, dwell time and other data on the platform such as website and APP. Clean these data, remove invalid information (such as duplicate records, incorrect format data). Then feature extraction, convert user behavior into quantifiable features, such as user's consumption ability (judged according to historical purchase amount), interest preference (determined according to browsing and search content), activity level (calculated according to login frequency and dwell time), etc. Build a complete user portrait database; Content priority data is set by marketing personnel according to the target and strategy of marketing activities. For example, for promotion activities related content, determine the priority according to the time urgency (remaining time to the end of the activity), expected conversion rate and other factors; For regular marketing content, determine the priority according to its importance to brand promotion, user retention and other aspects. Convert these evaluation criteria into specific quantitative indicators, score the marketing content through the system, and determine the priority value of each content task; Resource state data is collected in real time by deploying monitoring programs on each AI agent and the server running the agent, including the number of tasks currently processed by the agent, task processing progress, average processing time, and CPU utilization, GPU utilization, memory occupancy, network bandwidth usage and other resource state data of the server. The collected data is transmitted to the central data processing center in real time for summary and formatting processing for subsequent analysis and use.

[0018] S2, establish a task priority evaluation model, calculate the comprehensive priority score of the task according to the processed multi-dimensional index data and sort, including: S21, the task priority evaluation module takes the user portrait data, content priority data and resource state data in step S1 as the core dimensions of input; S22, quantitatively process each dimension index of input; For user portrait data, convert user features into 0-1 standardized user value coefficient (such as high value user coefficient 0.8-1.0); Content priority data directly uses the priority value (such as 1-10 points) determined in step S1; Resource state data is quantified by calculating resource matching index (such as the inverse of the current load rate of the agent).

[0019] S23, according to the marketing scene characteristics, each dimension index is assigned a weight, and the weight can be dynamically adjusted according to the business target; S24, the weighted sum formula is used to calculate the task comprehensive priority score: ; Wherein, is the standardized value of the user portrait data; is the quantitative value of the content priority data; is the standardized value of the resource state data; 、 、 All are weight coefficients, and ; S25, according to the comprehensive priority score, all tasks are sorted, and the task with the highest score is preferentially entered into the scheduling queue corresponding to the reinforcement learning scheduling strategy training link of S3.

[0020] S3, construct a simulated marketing environment, determine the state space, action space and reward signal, train the reinforcement learning model to obtain an optimized scheduling strategy, specifically including: S31, construct a simulated marketing environment, taking the task comprehensive priority sorting result output in step S2 as input, and construct a simulated environment containing task features (such as user portrait labels, content priority scores), AI agent state (such as current load, historical processing success rate) and resource constraints (such as server CPU / GPU utilization threshold). The environment dynamically generates a task flow, and the task attributes include user value coefficient (from the user portrait data of S2), content urgency (from the priority score of S2) and resource demand (such as calculation time estimation), and simulates user interaction feedback; S32, determine the core elements of the reinforcement learning model: state space, action space and reward function, specifically: State space definition: fusion of task priority sorting result in step S2 , agent real-time load , resource utilization and task context vector , form state vector ; Action space definition: action represents the decision of assigning tasks to agents, and the discrete space is defined as: , represents assigning the th task to the th agent, wherein, represents the task index sorted in step S2; represents the available agent index; Set the reward function: ; in, For the task In the intelligent body The execution time on 、 、 is the weight, Penalty for resource load: ; in, The resource load threshold is used to determine whether the agent is overloaded; when the actual load of the agent is If the threshold is exceeded, the resource load penalty will be triggered, prompting the model to reduce the number of tasks assigned to the overloaded agent; if ≤threshold, the penalty term is 0, allowing normal task allocation to balance the agent load and optimize resource utilization.

[0021] S33. Use the deep Q-network architecture and related training mechanisms to train the reinforcement learning model to select the optimal task allocation action strategy under different states in the marketing scenario task scheduling. Specifically: Adopting deep Q network architecture, storing through experience replay mechanism Sample, use the target network to update the Q value: ; in, is the learning rate, is the discount factor, are the main network parameters, is the target network parameter, and the loss function uses the mean square error: Optimize the parameters by gradient descent, synchronize the target network parameters every N steps, and use - Greedy strategy balances exploration and exploitation, specifically: Initialize exploration probability: set the initial exploration probability and the final exploration probability , and the exploration probability decay rate ; Select actions: At each state Generate a random number between [0,1] ,like , then randomly select an action , uniformly sample from the action space to explore new task allocation methods; if , then select the action that maximizes the Q value predicted by the current Q network. , using the learned strategy; Update exploration probability: After each training iteration, update the exploration probability according to the following formula : ; As the training progresses, gradually decrease, so that the agent increasingly utilizes the learned optimal strategy and reduces exploration of new actions.

[0022] S34, after the training converges, the task queue sorted in step S2 is outputted according to the state optimal action , forming a dynamic scheduling strategy to ensure that high-priority tasks are assigned to adaptive agents first.

[0023] S4, model the marketing tasks, AI agents, and audience elements as nodes and establish associated edges, train a graph neural network model to predict task execution time and effectiveness, specifically including: S41, integrate nodes and edges into a heterogeneous graph , wherein is the set of all nodes, is the set of all edges; S42, use a graph neural network to aggregate node features, and the feature update formula of each layer is: ; wherein, is the feature of node in the first layer, is the neighbor set of node , is the weight of the edge between node and its neighbor node , is the node degree, and are learnable parameters, is an activation function; S43, predict task execution time and effectiveness: For the task-agent pair , the execution time is predicted by the output layer of the graph neural network, and the loss function is the mean square error: ; wherein, is the actual execution time, is the number of samples; For the task-agent-audience triple , the task conversion rate is predicted, and the loss function is cross-entropy: ; wherein, For the actual conversion rate.

[0024] S5, task allocation according to task priority ranking, reinforcement learning scheduling strategy and graph neural network prediction results, and real-time monitoring and adjustment, specifically: For tasks and agents , the score is: ; Wherein, is the comprehensive priority score of the task; is the action value output by the reinforcement learning model; , , are weight coefficients; For the sorted task queue, the agent with the highest score is selected: Wherein, is the set of available agents (which must meet the resource constraints: , is the increased load rate of the task .

[0025] Real-time monitoring and adjustment: The monitoring indicators include: The deviation between the actual execution time and the prediction of the graph neural network: ; The real-time load rate of the agent: (Updated every 5 seconds); The task timeout rate: The number of tasks that have timed out / total number of tasks.

[0026] Dynamic adjustment strategy: If (deviation exceeds 20%), trigger incremental training of the graph neural network model (update the last layer parameters with new data); If (load is too high), reassign the uncompleted tasks of the agent according to the reinforcement learning strategy of S3: Wherein, is the set of agents with load rate .

[0027] Therefore, the application adopts the above-mentioned marketing scene-oriented AI agent task scheduling execution method, through multi-dimensional index driven task priority evaluation, the importance and urgency of the task can be comprehensively and accurately judged, and important tasks can be ensured to be processed in priority; the adaptive scheduling strategy optimized by reinforcement learning enables the task allocation to be adjusted in real time with the change of the environment, improving the efficiency of resource utilization under complex marketing activities; the intelligent decision supported by the graph neural network can capture the potential association patterns between tasks and between tasks and agents, realizing more accurate task allocation.

[0028] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit them, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that: the technical solutions of the present application can still be modified or replaced by equivalents, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. The AI ​​agent task scheduling execution method for marketing scenarios is characterized by: The following steps are involved: S1. Collect and process multi-dimensional indicator data, including user profile data, content priority data, and resource status data; S2. Establish a task priority evaluation model to calculate and rank the comprehensive priority scores of tasks based on the processed multi-dimensional indicator data; S3. Build a simulated marketing environment, determine the state space, action space, and reward signal, and train the reinforcement learning model to obtain an optimized scheduling strategy. S4. Model marketing tasks, AI agents, and audience elements as nodes and establish associated edges, and train a graph neural network model to predict task execution time and effectiveness. S5. Assign tasks based on task priority, reinforcement learning scheduling strategy, and graph neural network prediction results, and monitor and adjust in real time.

2. The AI ​​agent task scheduling execution method for marketing scenarios according to claim 1 is characterized by: The user portrait data in step S1 includes collecting the user's browsing history, purchasing behavior, search keywords and stay time.

3. The AI ​​agent task scheduling execution method for marketing scenarios according to claim 1 is characterized in that: Step S2 specifically includes: S21, the task priority assessment module uses the user portrait data, content priority data and resource status data in step S1 as the core dimensions of input; S22, quantify the input indicators of each dimension; S23. Assign weights to each dimension indicator based on the characteristics of the marketing scenario; S24. Calculate the task's comprehensive priority score using the weighted sum formula: ； in, It is the standardized value of user portrait data; It is a quantitative value of the content priority data; is the standardized value of resource status data; 、 、 are weight coefficients, and ; S25. Sort all tasks according to their comprehensive priority scores. Tasks with the highest scores are given priority to enter the scheduling queue corresponding to the reinforcement learning scheduling strategy training phase in S3.

4. The AI ​​agent task scheduling execution method for marketing scenarios according to claim 1 is characterized in that: Step S3 specifically includes: S31. Construct a simulated marketing environment. Using the comprehensive task priority ranking result output from step S2 as input, construct a simulated environment that includes task characteristics, AI agent status, and resource constraints. The environment dynamically generates a task flow. Task attributes include user value coefficient, content urgency, and resource requirements, and simulates user interaction feedback. S32. Identify the core elements of the reinforcement learning model: state space, action space, and reward function; S33. Use the deep Q-network architecture and related training mechanisms to train the reinforcement learning model to select the optimal task allocation action strategy under different states in the marketing scenario task scheduling; S34. After the training converges, the task queue sorted in step S2 is output with the optimal action according to the state to form a dynamic scheduling strategy to ensure that high-priority tasks are assigned to the adaptive intelligent agent first.

5. The AI ​​agent task scheduling execution method for marketing scenarios according to claim 4 is characterized in that: Step S32 is specifically as follows: State space definition: Fusion step S2 task priority sorting results , Real-time load of intelligent agents , resource utilization and task context vector , forming the state vector ; Action space definition: Actions represent decisions that assign tasks to agents, and the discrete space is defined as: ,in, represents the task index after sorting in step S2; Represents the available agent index; Set up the reward function: ; in, For the task In the intelligent body The execution time on 、 、 is the weight, Penalty for resource load: ； in, is the resource load threshold, which is used to determine whether the agent is overloaded.

6. The AI ​​agent task scheduling execution method for marketing scenarios according to claim 5 is characterized in that: Step S33 is specifically as follows: Adopting deep Q network architecture, storing through experience replay mechanism Sample, use the target network to update the Q value: ； in, is the learning rate, is the discount factor, are the main network parameters, is the target network parameter, and the loss function uses the mean square error: Optimize the parameters by gradient descent, synchronize the target network parameters every N steps, and use - Greedy strategy balances exploration and exploitation, specifically: Initialize exploration probability: set the initial exploration probability and the final exploration probability , and the exploration probability decay rate ; Select actions: At each state Generate a random number between [0,1] ,like , then randomly select an action , uniformly sample from the action space to explore new task allocation methods; if , then select the action that maximizes the Q value predicted by the current Q network. , using the learned strategy; Update exploration probability: After each training iteration, update the exploration probability according to the following formula : ； As the training progresses, Reducing , allows the agent to use the optimal strategy that has been learned and reduce the exploration of new actions.

7. The AI ​​agent task scheduling execution method for marketing scenarios according to claim 6 is characterized by: After the training in step S34 converges, the task queue sorted in step S2 is sorted according to the status Output optimal action , forming a dynamic scheduling strategy to ensure that high-priority tasks are assigned to adaptive agents first.

8. The AI ​​agent task scheduling execution method for marketing scenarios according to claim 7 is characterized in that: Step S4 specifically includes: S41. Integrate nodes and edges into a heterogeneous graph ,in is the set of all nodes, is the set of all edges; S42. Use graph neural network to aggregate node features. The feature update formula for each layer is: ； in, For nodes In the Layer characteristics, For nodes The neighbor set of For nodes and its neighboring nodes The weight of the edge between is the node degree, and is a learnable parameter, is the activation function; S43. Predict task execution time and effectiveness: For task-agent pairs , predicting execution time through the graph neural network output layer , the loss function is the mean square error: ； in, is the actual execution time, is the number of samples; For the task-agent-audience triplet Predicting task conversion rate , the loss function is cross entropy: ； in, is the actual conversion rate.

9. The AI ​​agent task scheduling execution method for marketing scenarios according to claim 8 is characterized in that: Step S5 constructs an allocation scoring function based on task priority ranking, reinforcement learning scheduling strategy, and graph neural network prediction results to select the optimal agent. Specifically: Task and agents ,score for: ； in, Give tasks a comprehensive priority score; The action value output by the reinforcement learning model; 、 、 is the weight coefficient; For the sorted task queue, select the agent with the highest score: ,in, is the set of available agents.

Images