A method and apparatus for controlling a social intelligent agent swarm

Abstract

A social intelligent agent device and its cluster control method and apparatus are disclosed. The method includes: statistically analyzing different user groups on an internet platform to construct corresponding user feature profiles; based on the internet user feature profiles, cultivating user profiles for intelligent agents deployed on the cluster to obtain intelligent agents with preferences; creating information consumption activity tasks for the intelligent agents based on a B / S architecture; and collecting information consumption data generated by the intelligent agent activities. This invention improves the realism of social intelligent agents and the efficiency of cluster control through intelligent agent profile cultivation and task scheduling optimization. Intelligent agent behavior is more diverse and natural, enabling more realistic simulation of user interactions. The cluster control mechanism is responsive and flexible, with more efficient data synchronization and task allocation, solving problems such as monotonous intelligent agent behavior, slow cluster control, complex data processing, and inability to provide real-time feedback in existing technologies, thereby improving information dissemination efficiency and social platform activity.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, distributed systems, and network communication, and in particular to a social intelligent agent device and its cluster control method and apparatus. Background Technology

[0002] With the rapid development of the internet and social media, information dissemination has gradually entered the Web 3.0 era. The application of big data and intelligent algorithms has greatly improved the efficiency of information acquisition and dissemination. Social platforms such as Weibo, Twitter, Facebook, and Toutiao have become important channels for users to obtain information, share opinions, and participate in interactions. As an application based on artificial intelligence technology, social intelligent agents can simulate the behavior of real users on these platforms, autonomously performing social tasks such as browsing, commenting, liking, and sharing, thereby improving the efficiency of information dissemination and platform activity.

[0003] The lifelike simulation effect of social agents is one of their core advantages. From an agent's perspective, as "agents" for information dissemination, social agents need to highly simulate the behavior of real users to integrate into the platform ecosystem and interact effectively. By mimicking human online behavior, social agents can achieve interactions highly similar to real users, enhancing the effectiveness of information dissemination. However, existing social agents have some significant shortcomings. First, the behavioral patterns of existing social agents are relatively simple, lacking sufficient flexibility and diversity, resulting in low realism. The behavior of agents is often too mechanical, unable to fully simulate complex user behaviors, thus limiting their application scope.

[0004] Secondly, the efficiency and flexibility of social agent cluster control are relatively poor. Currently, most social agent clusters lack rapid response mechanisms and efficient task scheduling capabilities, and cannot adjust task and resource allocation in real time, resulting in low efficiency when performing complex tasks. In addition, data synchronization and communication latency issues still exist, and the agents' real-time feedback capabilities are insufficient, affecting their performance on large-scale platforms.

[0005] Therefore, improving the realism and efficiency of social agents' cluster control are major challenges currently facing the technology. To enhance realism, social agents need to exhibit greater diversity and naturalness in their behavior, thereby achieving a higher level of interactive simulation. Simultaneously, optimizing the task scheduling and data synchronization capabilities of agent clusters, and enhancing their adaptability and flexibility in complex social environments, are also key to improving technical performance. Solving these problems will facilitate the widespread application of social agents in information dissemination. Summary of the Invention

[0006] The present invention aims to overcome the above-mentioned shortcomings of the prior art and provides a social intelligent agent device and its cluster control method and apparatus.

[0007] In a first aspect, the present invention provides a method for controlling a social intelligent agent cluster, comprising:

[0008] S1. Collect statistics on different user groups on the internet platform and construct corresponding user profiles;

[0009] S2. Based on the user profiles of Internet users, user profiles are generated for the agents deployed on the cluster to obtain agents with preferences;

[0010] S3. Create information consumption activity tasks for the agent based on the BS (Browser / Server) architecture;

[0011] S4. Collect information consumption data generated by the activities of intelligent agents.

[0012] Optionally, step S1, which involves statistically analyzing different user groups on the internet platform and constructing corresponding user profiles, includes: performing multi-dimensional data analysis by acquiring users' historical behavior data, access trajectories, content browsing records, and social interaction information. These analysis dimensions may include, but are not limited to, users' interests, occupational information, social circles, purchasing preferences, and device usage habits. Next, based on users' basic attributes such as geographical location, age group, gender, and income level, and combined with significant features from their historical behavior, segmented user profiles are formed. By classifying and aggregating users according to multiple feature dimensions, high-value user groups and potential interest groups can be identified. Finally, user profile features for different user groups are constructed.

[0013] Optionally, step S2 involves cultivating user profiles for the agents deployed on the cluster based on internet user characteristic profiles to obtain agents with preferences. This process is as follows:

[0014] S201. Data Acquisition and Preprocessing: Using user profiles, time-series data is extracted from internet users' swiping, touching, and browsing interactions. Simultaneously, historical user data and user profile data are acquired, serving as input sources for user feature profiling and providing comprehensive data support for subsequent steps. The collected behavioral sequence data is then cleaned and standardized, outliers are removed, and the data is arranged chronologically to form a complete time-series feature dataset. Missing values ​​are imputed using interpolation or other appropriate methods to ensure the continuity and usability of the time-series data.

[0015] S202. Behavioral Sequence Modeling: Time-series feature models are constructed for swipe, touch, and browsing interactions. Specifically, a recurrent neural network (RNN) structure is used to model the above behavioral sequences to capture the temporal features and long-term dependencies of user behavior data. To alleviate the gradient vanishing problem that may occur during RNN network training, a gated recurrent unit (GRU) is further employed. The calculation formula for GRU is as follows:

[0016] Update Gate: r t =σ(W er e t +W hr h t-1 +b r (1)

[0017] Reset Gate: z t =σ(W ez e t +W hz h t-1 +b z (2)

[0018] Candidate hidden state:

[0019] Where e t is the embedding vector of the t-th action, σ is the sigmoid function, and ⊙ is the element-wise multiplication operator.

[0020] To support dynamic updates, this application introduces a dynamic weight adjustment mechanism, which dynamically updates the hidden state through a time decay function and an update increment.

[0021] To reduce the impact of outdated behaviors on the current user profile, the following time decay formula is designed:

[0022] w t =α.e -β·(T-t) (4)

[0023] Where T is the current timestamp, t is the behavior timestamp, and α and β are adjustable parameters used to control the decay rate of the time weight. The time decay weight is used to reduce the impact of historical behavior on the current decision and ensure that the model can prioritize recent behavior.

[0024] Dynamic hidden state update:

[0025]

[0026] Where h t h is currently hidden. t-1 This is the previous hidden state. The candidate hidden state for the current behavior is calculated using the core formula of GRU. t The time decay weight controls the balance between historical behavior and current behavior.

[0027] The current hidden state is jointly determined by the historical hidden state and the current behavioral state, but the contribution of the historical state decays over time. Thus, the model can dynamically adjust the hidden state based on time weights, balancing the influence of short-term and long-term interests.

[0028] S203. Dynamic Update Mechanism for User Profiles: This mechanism comprises three aspects: real-time preference prediction, long-term preference trend updating, and global user profile updating. The core of this mechanism lies in its ability to dynamically adjust user profiles by integrating real-time preference prediction and long-term preference trends. Compared to traditional static profiles, this mechanism can adaptively update user profiles, enhancing the agent's responsiveness to changes in user interests.

[0029] Real-time preference prediction:

[0030] P real-time =softmax(W r ·h T +b r (6)

[0031] Where P real-time W represents the real-time preference vector, which in turn represents the current distribution of user interests. r h represents the weight matrix for real-time preference prediction, used to convert hidden states into preference probabilities. T The hidden state of the latest time step reflects the user's current behavioral characteristics. r This is a bias term used to adjust the baseline value for preference prediction.

[0032] At this point, the intelligent agent can predict the distribution of user preferences for various interest tags in real time based on the latest behavioral data.

[0033] Long-term preference trend update:

[0034]

[0035] This is the updated long-term preference vector. Let P be the long-term preference vector from the previous time step. real-time γ represents the current real-time preference vector, and γ is the long-term preference update weight (ranging from 0 to 1), used to balance real-time preferences with historical trends.

[0036] Long-term preference trends are gradually adjusted based on current forecast results to ensure that the profile reflects users' immediate interests while maintaining long-term stability.

[0037] Global user profile update:

[0038] P global =δ·P real-time +(1-δ)·P long-term (8)

[0039] P global A global user profile is created, integrating short-term interests and long-term trends, which serves as the final basis for the agent's response. δ represents the integration weight (ranging from 0 to 1), adjusting the ratio of short-term preferences to long-term preferences.

[0040] By creating a global profile, the intelligent agent can integrate short-term and long-term interests to provide users with more accurate response services.

[0041] S204. Hierarchical Feature Fusion: This is discussed in two categories: intra-sequence feature extraction (using attention mechanisms) and cross-sequence feature fusion (using cross-view attention mechanisms).

[0042] Intra-sequence feature extraction: An intra-sequence attention mechanism is applied individually to each behavior sequence to calculate the importance weight of each action within the sequence, thereby identifying key actions within the user's behavior sequence. Specifically, the attention score is calculated using the following formula:

[0043]

[0044] in This is the final output state of the feedforward GRU model. This is the t-th output state of the bidirectional GRU model. The attention score a is calculated using this formula. t Able to reflect action h t With the current action The relationship between the two is used to determine important actions. Then, the output of the bidirectional GRU is element-wise multiplied with the corresponding attention score vector to obtain the output of the attention layer within the sequence.

[0045] Cross-sequence feature fusion: Utilizing a cross-sequence attention mechanism to fuse interactive features between different behavioral sequences. (Using a sliding interactive view v) s and browsing interactive view v b For example, cross-view attention mechanism This mechanism can effectively capture the correlation and interaction patterns between asynchronous behavior sequences, and obtain the output of the cross-view attention layer.

[0046] S205. Multi-task Optimization: A multi-task learning framework is designed to improve the ability to identify user behavior patterns by simultaneously predicting short-term and long-term reading intentions. Feature embedding vectors from different behavior sequences are fused through concatenation and then input into the multi-task learning network. Loss functions are defined for the short-term and long-term reading intention prediction tasks, respectively.

[0047] The loss function L for the short-term reading intent prediction task short :

[0048]

[0049] Loss function L for long-term reading intent prediction task long :

[0050]

[0051] Where y is the actual reading label, and N is the total number of samples in the dataset. It is the training set.

[0052] Global loss function L global Defined as a weighted sum of the losses from the two tasks, to achieve multi-task learning:

[0053] L global =λL short +(1-λ)L long (12)

[0054] Here, λ is a hyperparameter used to balance the importance of the two tasks.

[0055] In this way, the model can simultaneously learn to predict both the robot's immediate reading behavior and long-term reading intentions. This not only improves the model's understanding of robot behavior but also enhances its adaptability and accuracy in practical applications. This multi-task learning approach enables the model to capture the complexity and diversity of robot behavior, thus playing a crucial role in information consumption and social interaction.

[0056] S206. Agent Preference Feature Generation and Deployment: This mechanism combines real-time preference prediction results with persistent intent prediction results to form a dynamic update mechanism for user profiles. Through a reverse update mechanism, the generated user profile features are continuously applied to agents on the cluster, ultimately cultivating the agents' preference features. Then, agents with these preferences are deployed to the cluster environment. In this environment, the generated agent preference features are used to respond to user interactions in real time, and the preference features are dynamically adjusted based on changes in user behavior data to improve the agents' responsiveness and adaptability.

[0057] Optionally, step S3, which involves creating information consumption activity tasks for the intelligent agent based on a B / S architecture, includes: creating and managing tasks within a user-friendly web interface, and utilizing the powerful processing capabilities of the server to allocate and execute tasks. Users create new intelligent agent tasks in an intuitive task management interface and select suitable intelligent agents from a predefined pool to perform the task. During task creation, users need to set detailed task parameters, including but not limited to the target platform, the scheduled execution time, the specific description of the task, and the expected results. Furthermore, users can select specific types of intelligent agents to execute tasks based on their specific needs, such as overseas intelligent agents customized for different regions or language environments.

[0058] Optionally, step S4, collecting information consumption data generated by the agent's activities, includes: recording various types of information data received by the agent during its access to internet media platforms, such as the information display on homepages, recommendation pages, category pages, and search pages, including content type, source, publication time, and interaction volume (likes, comments, reposts, etc.). For content that is viewed in detail, collect comment-related data, including comment content, author, time, and interaction volume. The agent controller is responsible for receiving and storing this data, and recording metadata based on the collection behavior, such as collection time, source page, and classification criteria, for subsequent analysis and querying. After collecting the information consumption data generated by the agent's activities, the key to the control method lies in the verification of agent feedback and task completion. The standard for task completion includes not only the completeness of data collection, but also an assessment of whether the agent has achieved the predetermined goal according to the task requirements. For example, if the task is to browse a certain type of content, the agent needs to effectively browse according to the task objectives (such as a specific content type, time period, interaction volume, etc.) and provide feedback on the task execution results, including whether the predetermined interaction volume or browsing depth has been achieved. By comprehensively analyzing the process and feedback data of the intelligent agent executing tasks, the system can automatically determine whether the task has been completed, further adjust task parameters or reassign tasks to ensure the accuracy and efficiency of task execution, thereby achieving comprehensive control over the intelligent agents in the cluster.

[0059] Secondly, this invention provides a social intelligent agent device, including an intelligent agent module, a web page automation robot based on Autojs or Jxbrowser, distributed on various public and private cloud hosts or mobile phones. It is an automated program capable of autonomously consuming information and engaging in social behaviors such as browsing, reading, following, commenting, and liking on designated social media platforms based on assigned task information, and also possesses certain text classification, content monitoring, and image recognition capabilities. A task allocation module assigns tasks to one or a group of specific social intelligent agent programs according to execution time and related attributes. A task collection module collects relevant task feedback and data. An automated program handles related errors during task execution. A message broker module includes a task queue unit and a task feedback unit, decoupling the relationship between the intelligent agent and the task allocator. Different types of tasks are distributed through a single-point queue or batch-distributed topics, and social intelligent agent feedback is provided through a queue. A data storage module stores social intelligent agent task information, feedback information, browsing behavior information, social intelligent agent state information, intelligent agent profile information, etc.

[0060] After a user publishes a task, the task information is stored in the task information table of the data storage module. The task allocation module reads the task information and sends the task to the task queue unit in the message broker module. The agent module retrieves the task from the task queue unit and executes it. After execution, it sends the task result to the feedback queue unit in the message broker module. The task collection module reads the task feedback information from the feedback queue unit and stores the feedback information, browsing behavior information, profile information, social agent information, and other data in the data storage module. Furthermore, the agent module performs a separate backup of all information in the data storage module.

[0061] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in either the first or second aspect above.

[0062] The advantages of this invention are: by cultivating intelligent agent profiles and optimizing task scheduling, the realism of social intelligent agents and the efficiency of cluster control are improved; the behavior of intelligent agents is more diverse and natural, and can more realistically simulate user interaction; the cluster control mechanism is responsive and flexible, and data synchronization and task allocation are more efficient, solving the problems of single intelligent agent behavior, slow cluster control, complex data processing, and inability to provide real-time feedback in the prior art, thereby improving the efficiency of information dissemination and the activity of social platforms. Attached Figure Description

[0063] The accompanying drawings, which are included to provide a further understanding of this application and constitute a part of this application, illustrate exemplary embodiments of the application and are used to explain the application, but do not constitute an undue limitation of the application. In the drawings:

[0064] Figure 1 This is a flowchart of the method of the present invention;

[0065] Figure 2 This is a structural diagram of the social intelligent agent device of the present invention;

[0066] Figure 3 This is a flowchart of the task allocator logic in the social intelligent agent device of the present invention;

[0067] Figure 4 This is a flowchart of the task collector logic in the social intelligent agent device of the present invention. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0069] Example 1

[0070] like Figure 1 This embodiment relates to a social intelligent agent cluster control method, including the following steps:

[0071] Step 101: Analyze different user groups on the internet platform and construct corresponding user profiles. In this embodiment, through refined analysis and segmentation of user group characteristics, the behavioral habits and preferences of different user groups are effectively identified, providing a solid foundation for subsequent data mining and targeted recommendations. Specifically, the system first collects multi-dimensional data generated by users on the platform, including but not limited to users' basic attributes, historical operational behaviors, and interaction data. Through this data, a comprehensive and systematic understanding of users' usage and characteristics on the platform can be achieved.

[0072] As an optional implementation method in this embodiment, firstly, the attribute distribution information of platform users is the core foundation for user profile construction. To accurately construct user feature profiles, the system uses automated data collection and data mining techniques to obtain users' geographical distribution attributes, age group divisions, gender classifications, activity time attributes, and content preference attributes. Specifically, geographical distribution attributes can be identified based on users' IP addresses or other geographical information, helping the platform understand the activity levels and content needs of users in different regions; age attributes are divided into different age groups based on users' registration information or social profiles, such as minors, youth, middle-aged, and elderly; in terms of gender attributes, the system not only distinguishes between traditional male and female users but can also introduce mixed gender options based on the composition of the platform's user groups to more accurately reflect the actual user situation.

[0073] As an optional implementation in this embodiment, in addition to basic demographic characteristics, the activity time attribute of platform users is also a key analytical indicator. The system constructs an activity time distribution model and an activity interval distribution model by analyzing users' historical behavioral data. The activity time distribution model records users' activity levels at different times, such as morning, afternoon, and evening, helping to identify peak usage periods. The activity interval distribution model records the interval between each user's access to the platform, helping the system identify the user's usage frequency and activity cycle.

[0074] As an optional implementation method in this embodiment, obtaining content preference attributes provides richer dimensions for constructing user profiles. In content preference attributes, the system extracts information categories that users are interested in by analyzing the content they browse, click, comment on, and share. These categories can be divided according to predefined classification standards on the platform, such as news, entertainment, and technology. Each user's content preferences are ranked according to their activity level in these different categories, forming a user's content preference profile. Furthermore, the system further refines content classification by extracting keywords that users focus on during browsing and interaction. Keywords not only include content-related identifiers but may also involve information such as geographical location and gender preferences, making the profile construction more comprehensive and multi-dimensional.

[0075] In this embodiment, regarding user content consumption behavior, it's not only necessary to record user clicks, but also to track metrics such as dwell time and interaction frequency for each type of information, to more comprehensively reflect the user's actual interest in that type of information. For information viewed in detail, the system will further refine the recording, obtaining behavioral data such as comments, reposts, and likes, thereby analyzing the user's depth of content engagement. This data is crucial in building user profiles, helping the platform identify highly engaged user groups and potential core user groups, providing support for subsequent targeted marketing and advertising.

[0076] Step 102: Based on the internet user characteristic profile, cultivate user profiles for the agents deployed on the cluster to obtain agents with preferences. In this embodiment, cultivating user profiles for agents involves developing their interests through specific preferred reading behaviors or other specific interactions with the recommendation system, thereby obtaining agents with preferences.

[0077] As an optional implementation of this embodiment, user profiling is performed on the intelligent agent to obtain an intelligent agent with preferences, including:

[0078] Data Acquisition and Preprocessing: Time-series data is extracted from internet users' swiping, touching, and browsing interactions using user profiles. Simultaneously, historical user data and user configuration file data are acquired, serving as input sources for user feature profiling and providing comprehensive data support for subsequent steps. The collected behavioral sequence data is then cleaned and standardized, outliers are removed, and the data is arranged chronologically to form a complete time-series feature dataset. Missing values ​​are imputed using interpolation or other appropriate methods to ensure the continuity and usability of the time-series data.

[0079] Behavioral sequence modeling: Time series feature models are constructed for swipe, touch, and browsing interactions. Specifically, a recurrent neural network (RNN) structure is used to model these behavioral sequences to capture the temporal features and long-term dependencies of user behavior data. To mitigate the vanishing gradient problem that may occur during RNN training, gated recurrent units (GRUs) are further employed. The calculation formula for GRUs is as follows:

[0080] Update Gate: r t =σ(W er e t +W hr h t-1 +b r (1)

[0081] Reset Gate: z t=σ(W ez e t +W hz h t-1 +b z (2)

[0082] Candidate hidden state:

[0083] Where e t is the embedding vector of the t-th action, σ is the sigmoid function, and ⊙ is the element-wise multiplication operator.

[0084] To support dynamic updates, this application introduces a dynamic weight adjustment mechanism, which dynamically updates the hidden state through a time decay function and an update increment.

[0085] To reduce the impact of outdated behaviors on the current user profile, the following time decay formula is designed:

[0086] w t =α.e -β·(T-t) (4)

[0087] Where T is the current timestamp, t is the behavior timestamp, and α and β are adjustable parameters used to control the decay rate of the time weight. The time decay weight is used to reduce the impact of historical behavior on the current decision and ensure that the model can prioritize recent behavior.

[0088] Dynamic hidden state update:

[0089]

[0090] Where h t h is currently hidden. t-1 This is the previous hidden state. The candidate hidden state for the current behavior is calculated using the core formula of GRU. t The time decay weight controls the balance between historical behavior and current behavior.

[0091] The current hidden state is jointly determined by the historical hidden state and the current behavioral state, but the contribution of the historical state decays over time. Thus, the model can dynamically adjust the hidden state based on time weights, balancing the influence of short-term and long-term interests.

[0092] The dynamic update mechanism for user profiles comprises three aspects: real-time preference prediction, long-term preference trend updates, and global user profile updates. The core of this mechanism lies in its ability to dynamically adjust user profiles by integrating real-time preference predictions and long-term preference trends. Compared to traditional static profiles, this mechanism can adaptively update user profiles, enhancing the agent's responsiveness to changes in user interests.

[0093] Real-time preference prediction:

[0094] P real-time =softmax(W r ·h T +b r (6)

[0095] Where P real-time W represents the real-time preference vector, which in turn represents the current distribution of user interests. r h represents the weight matrix for real-time preference prediction, used to convert hidden states into preference probabilities. T The hidden state of the latest time step reflects the user's current behavioral characteristics. r This is a bias term used to adjust the baseline value for preference prediction.

[0096] At this point, the intelligent agent can predict the distribution of user preferences for various interest tags in real time based on the latest behavioral data.

[0097] Long-term preference trend update:

[0098]

[0099] This is the updated long-term preference vector. Let P be the long-term preference vector from the previous time step. real-time γ represents the current real-time preference vector, and γ is the long-term preference update weight (ranging from 0 to 1), used to balance real-time preferences with historical trends.

[0100] Long-term preference trends are gradually adjusted based on current forecast results to ensure that the profile reflects users' immediate interests while maintaining long-term stability.

[0101] Global user profile update:

[0102] P global =δ·P real-time +(1-δ)·P long-term (8)

[0103] P global A global user profile is created, integrating short-term interests and long-term trends, which serves as the final basis for the agent's response. δ represents the integration weight (ranging from 0 to 1), adjusting the ratio of short-term preferences to long-term preferences.

[0104] By creating a global profile, the intelligent agent can integrate short-term and long-term interests to provide users with more accurate response services.

[0105] Hierarchical feature fusion: This is discussed in two categories: one is intra-sequence feature extraction (using attention mechanisms), and the other is cross-sequence feature fusion (using cross-view attention mechanisms).

[0106] Intra-sequence feature extraction: An intra-sequence attention mechanism is applied individually to each behavior sequence to calculate the importance weight of each action within the sequence, thereby identifying key actions within the user's behavior sequence. Specifically, the attention score is calculated using the following formula:

[0107]

[0108] in This is the final output state of the feedforward GRU model. This is the t-th output state of the bidirectional GRU model. The attention score a is calculated using this formula. t Able to reflect action h t With the current action The relationship between the two is used to determine important actions. Then, the output of the bidirectional GRU is element-wise multiplied with the corresponding attention score vector to obtain the output of the attention layer within the sequence.

[0109] Cross-sequence feature fusion: Utilizing a cross-sequence attention mechanism to fuse interactive features between different behavioral sequences. (Using a sliding interactive view v) s and browsing interactive view v b For example, the cross-view attention mechanism M(v) s v b ) = A s (v b v s v s )⊙A b (v s v b v b in This mechanism can effectively capture the correlation and interaction patterns between asynchronous behavior sequences, and obtain the output of the cross-view attention layer.

[0110] Multi-task optimization: A multi-task learning framework is designed to improve the ability to identify user behavior patterns by simultaneously predicting short-term and long-term reading intentions. Feature embedding vectors from different behavior sequences are fused through concatenation and then input into the multi-task learning network. Loss functions are defined for the short-term and long-term reading intention prediction tasks, respectively.

[0111] The loss function L for the short-term reading intent prediction task short :

[0112]

[0113] Loss function L for long-term reading intent prediction task long:

[0114]

[0115] Where y is the actual reading label, and N is the total number of samples in the dataset. It is the training set.

[0116] Global loss function L global Defined as a weighted sum of the losses from the two tasks, to achieve multi-task learning:

[0117] L global =λL short +(1-λ)L long (12)

[0118] Here, λ is a hyperparameter used to balance the importance of the two tasks.

[0119] In this way, the model can simultaneously learn to predict both the robot's immediate reading behavior and long-term reading intentions. This not only improves the model's understanding of robot behavior but also enhances its adaptability and accuracy in practical applications. This multi-task learning approach enables the model to capture the complexity and diversity of robot behavior, thus playing a crucial role in information consumption and social interaction.

[0120] Agent Preference Feature Generation and Deployment: A dynamic update mechanism for user profiles is formed by combining real-time preference prediction results with persistent intent prediction results. Through a reverse update mechanism, the generated user profile features are continuously applied to agents on the cluster, ultimately cultivating the agents' preference features. Then, agents with these preferences are deployed to the cluster environment. In this environment, the generated agent preference features are used to respond to user interactions in real time, and the preference features are dynamically adjusted based on changes in user behavior data to improve the agents' responsiveness and adaptability.

[0121] In this embodiment, user profiles of intelligent agents can be effectively developed, enabling them to make personalized responses and adjustments based on users' behavioral characteristics and preferences.

[0122] Step 103: Create a task for the agent based on the BS architecture.

[0123] In this embodiment, the process of creating tasks for intelligent agents based on a B / S architecture is managed through an intuitive web interface, enabling flexible task configuration and efficient agent scheduling. Users select the target platform, task time, task behavior, and required parameter settings via the web interface. Users input the task name and select a suitable intelligent agent from a predefined pool to execute the task, achieving precise matching between tasks and intelligent agents.

[0124] In this optional implementation, the task management system fully leverages the server-side processing power, making the allocation and execution of complex tasks efficient and reliable. Users create tasks through a web-based interface, select or customize the agent executing the task, and configure detailed task parameters, including the target platform (e.g., social media platform, e-commerce platform), the scheduled task execution time, a specific task description (e.g., data collection, comment analysis, user behavior tracking), and the expected task results. For tasks with special requirements, the system allows users to select specific types of agents based on the application scenario, such as agents suitable for different regions and language environments, thereby ensuring effective task execution globally.

[0125] This application further allows users to formulate and assign corresponding tasks based on the characteristics of the task, the deployment platform of the agent, and the content domain of interest, enabling the agent to automatically drive execution. Through this task management model based on a B / S architecture, the creation, scheduling, and monitoring processes of tasks are more efficient and transparent. Users can flexibly adjust task execution plans to ensure accurate task completion, greatly improving the system's usability and the execution efficiency of the agent.

[0126] Step 104: Collect information consumption data generated by the activities of the intelligent agent.

[0127] In this embodiment, the system ensures the transparency and controllability of task execution by real-time monitoring and data collection of the agent's operational status. During task execution, the system monitors the agent's behavior and task completion status in real time, including task progress, current interaction status, and the agent's performance metrics. If the system indicates that the task has ended without errors, the user can download all relevant task data with a single click.

[0128] To ensure the accuracy and completeness of data collection, the system provides intuitive task progress bars and status indicators, allowing users to clearly understand the task's progress. Through the real-time monitoring interface, users can clearly view information such as the number of tasks received, completed, and reported errors. Once the agent completes a task or the task progress bar reaches 100%, the system automatically confirms the task's successful completion and organizes all relevant data generated by the task. This data can be downloaded with a single click, allowing users to quickly obtain task results, log files, analysis reports, and more.

[0129] The collected task data includes, but is not limited to: project source, project category tags, browsing and collection time, number of likes and comments, keyword identification or classification results of project content, actual project price, and number of payments. Upon task completion, the system stores this data in a structured format. Users can download the complete project information, including the original HTML code, author ID, number of project favorites, number of views, number of shares, and project publication time. Furthermore, the system supports further analysis and statistics of the task data collected by the agent. Users can perform secondary processing on the data according to their needs to obtain deeper analytical conclusions.

[0130] This embodiment provides a secure, efficient, and user-friendly intelligent agent management platform through group feature profiling, intelligent agent profile development, real-time task monitoring, and one-click download functionality, enhancing user experience and system usability. The above describes the methods provided by one or more embodiments of this application. Based on the same concept, this application also provides a corresponding social intelligent agent device, see reference... Figure 2 The diagram illustrates different module information, including: An agent module, a web automation robot based on Auto.js or Jxbrowser, distributed across various public and private cloud hosts or mobile phones. This is an automated program capable of autonomously browsing, reading, following, commenting, liking, and engaging in other social behaviors on designated social media platforms based on assigned task information, while also possessing some text classification, content monitoring, and image recognition capabilities. A task allocation module assigns tasks to one or more specific social agent programs based on execution time and related attributes. A task collection module collects relevant task feedback and data. It is an automated program that handles errors related to task execution. A message broker module includes a task queue unit and a task feedback unit, decoupling the relationship between the agent and the task allocator. It distributes different types of tasks through a single queue or batch-distributed topics, and provides social agent feedback through a queue. A data storage module stores social agent task information, feedback information, browsing behavior information, social agent state information, agent profile information, etc.

[0131] After a user publishes a task, the task information is stored in the task information table of the data storage module. The task allocation module reads the task information and sends the task to the task queue unit in the message broker module. The agent module retrieves the task from the task queue unit and executes it. After execution, it sends the task result to the feedback queue unit in the message broker module. The task collection module reads the task feedback information from the feedback queue unit and stores the feedback information, browsing behavior information, profile information, social agent information, and other data in the data storage module. Furthermore, the agent module performs a separate backup of all information in the data storage module.

[0132] Example 2

[0133] like Figure 2 This embodiment relates to a social intelligent agent cluster control device, including: an intelligent agent module, a web page automation robot based on Autojs or Jxbrowser, distributed on various public and private cloud hosts or mobile phones; an automatic program that can autonomously perform information consumption and social behaviors such as browsing, reading, following, commenting, and liking in designated social media according to the assigned task information, and also has the ability of text classification, content monitoring and image recognition;

[0134] The task assignment module assigns tasks to one or a group of social AI agents based on execution time and related attributes.

[0135] The task collection module collects relevant task feedback and data; and an automated program handles errors during task execution.

[0136] The message broker module includes a task queue unit and a task feedback unit, decoupling the relationship between the agent and the task allocator; it distributes different types of tasks through a single queue or a batch-distributed topic, and provides feedback to the social agent through a queue.

[0137] The data storage module stores social agent task information, feedback information, browsing behavior information, social agent status information, and agent profile information;

[0138] After a user publishes a task, the task information is stored in the task information table in the data storage module; the task allocation module reads the task information and sends the task to the task queue unit in the message broker module; the agent module retrieves the task from the task queue unit and executes it, and after execution, sends the task result to the feedback queue unit in the message broker module; the task collection module reads the task feedback information in the feedback queue unit and stores the feedback information, browsing behavior information, profile information, social agent information, and other data in the data storage module; in addition, the agent module will perform separate backup storage of all information in the data storage module.

[0139] As an optional implementation in this embodiment, the social intelligent agent module can have 11 behaviors: recommended browsing, AI-recommended browsing, homepage browsing, AI-based homepage browsing, category browsing, search browsing, following, commenting, liking, replying, and posting content. Recommended browsing: Browsing the social platform's recommendation page with preferences, where preferences are keywords given by the user; AI-recommended browsing refers to browsing the social platform's recommendation page with preferences, where preferences are topic categories given by the user, and the results of classification by a classifier model are compared. For example, if the AI ​​agent's preference is positive news, a positive energy detection classifier is used to determine whether the news on the recommended page is positive news before deciding whether to click to read it. Homepage browsing: Browsing the social platform user's homepage with preferences, where preferences are keywords given by the user; AI-based homepage browsing refers to browsing the social platform user's homepage with preferences, where preferences are topic categories given by the user, and the results of classification by a classifier model are compared. For example, if the AI ​​agent's preference is positive news, a positive energy detection classifier is used to determine whether the news on the recommended page is positive news before deciding whether to click to read it. Search Browsing: Randomly browse keyword search pages on social media platforms, with keywords provided manually; Category Browsing: Randomly browse category topic pages on social media platforms, with categories and topics provided manually; Following: Follow users on social media platforms; Commenting: Comment on articles or videos on social media platforms; Liking: Like articles or videos on social media platforms; Replying: Replying to a comment on an article or video on social media platforms; Posting Content: Post specified content on social media platforms, which can be AI-generated or manually specified text and image content or a video.

[0140] As an optional implementation of this embodiment, refer to Figure 3 The task allocation module is responsible for querying the task table and broadcasting due tasks to the message broker module. After a task is published, the task allocation module enters a brief dormant state. Simultaneously, the task allocation module records the task status. Task completion requires the following condition: number of task profiles received = number of profiles completed + number of profiles that failed.

[0141] As an optional implementation of this embodiment, refer to Figure 4The task collection module receives status feedback or task feedback from the social agent. When the feedback is status feedback, the task collection module updates the social agent's status information and stores the updated status information in the data storage module. When the task feedback includes task failure information and the task still has remaining attempts, the task collection module changes the task status to "Pending Retry" and requests the social agent module to retry. If the task has no remaining attempts, the task status changes to "Task Failed." If the task is successful, and this task is a browsing behavior, the relevant browsing data (such as task browsing time, reading content, etc.) will be recorded and stored in the data storage module. If the task is not a browsing behavior, the relevant data of this behavior will be recorded and stored. If the task still has loop attempts, it will enter a loop and repeat the task. If the task has no remaining loop attempts, the task status will change to "Task Completed."

[0142] As an optional implementation in this embodiment, the message broker module includes a task queue unit and a task feedback unit. It is responsible for decoupling the relationship between the social agent and the task distributor, distributing different types of tasks through a single-point queue or a batch-distributed topic, and providing feedback to the social agent through a feedback queue. Specifically, the message broker module acts as a bridge between the social agent and the task distributor and task collector, and is responsible for indirectly storing relevant data into the data storage module.

[0143] As an optional implementation in this embodiment, the data storage module serves as the data center for the entire social agent cluster, primarily storing and protecting social agent task information, feedback information, browsing behavior information, social agent state information, and agent profile information.

[0144] Example 3

[0145] This application also provides a computer-readable medium storing a computer program that can be used to perform the above-described actions. Figure 1 and the methods provided Figure 2 The provided device.

[0146] This embodiment provides a non-transitory computer-readable storage medium that stores computer instructions, which cause the computer to execute the methods provided in the above-described method embodiments.

[0147] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various storage media that can store program code, such as ROM, RAM, magnetic disk, or optical disk.

[0148] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0149] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of each embodiment or some parts of the embodiments.

[0150] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention are within the scope of protection claimed by the present invention.

Claims

1. A method for controlling a social intelligent agent cluster, characterized in that, include: S1. Collect statistics on different user groups on the internet platform and construct corresponding user profiles; S2. Based on the user profile of Internet users, user profiles are cultivated for the agents deployed on the cluster to obtain agents with preferences; the real-time preference prediction results are combined with the long-term preference prediction results to form a dynamic update mechanism for user profiles; the dynamic update mechanism for user profiles includes three aspects: real-time preference prediction, long-term preference prediction, and global user profile update; through the reverse update mechanism, the generated user profile features are continuously applied to the agents on the cluster, and finally the cultivation of the agent's preference features is completed. Then, the agents with preferences are deployed to the cluster environment. In the cluster environment, the generated agent preference features are used to respond to the user's interaction behavior in real time, and the preference features are dynamically adjusted according to changes in user behavior data to improve the agent's responsiveness and adaptability. S3. Create information consumption activity tasks for the intelligent agent based on the BS architecture; S4. Collect information consumption data generated by the activities of intelligent agents.

2. The social intelligent agent cluster control method according to claim 1, characterized in that, Step S2 also includes: S201 Data Acquisition and Preprocessing: Using user profiles, time-series data is extracted from internet users' swiping, touching, and browsing interactions. Simultaneously, historical user data and user configuration file data are acquired, serving as input sources for user feature profiling and providing comprehensive data support for subsequent steps. The collected behavioral sequence data is then cleaned and standardized, outliers are removed, and the data is arranged chronologically to form a complete time-series feature dataset. Missing values ​​are imputed using interpolation or other appropriate methods to ensure the continuity and usability of the time-series data. S202. Behavioral Sequence Modeling: Time-series feature models are constructed for swipe interaction, touch interaction, and browsing interaction behaviors respectively. Specifically, a recurrent neural network (RNN) structure is used to model the above behavioral sequences to capture the temporal characteristics and long-term dependencies of user behavior data. To alleviate the gradient vanishing problem that may occur during RNN network training, a gated recurrent unit (GRU) is further adopted. The calculation formula of GRU is as follows: Update Gate: r t =σ(W er e t +W hr h t-1 +b r (1) Reset Gate: z t =σ(W ez e t +W hz h t-1 +b z (2) Candidate hidden state: Where e t is the embedding vector of the t-th action, σ is the sigmoid function, and ⊙ is the element-wise multiplication operator; To support dynamic updates, a dynamic weight adjustment mechanism is introduced, which dynamically updates the hidden state through a time decay function and an update increment. To reduce the impact of outdated behaviors on the current user profile, the following time decay formula is designed: w t =a·e -β·(T-t) (4) Where T is the current timestamp, t is the behavior timestamp, and α and β are adjustable parameters used to control the decay rate of the time weight. The time decay weight is used to reduce the impact of historical behavior on the current decision and ensure that the model can prioritize recent behavior. Dynamic hidden state update: Where h t h is currently hidden. t-1 This is the previous hidden state; The candidate hidden state for the current behavior is calculated using the core formula of GRU; w t The time decay weight is used to control the balance between historical behavior and current behavior; The current hidden state is jointly determined by the historical hidden state and the current behavioral state, but the contribution of the historical state decays over time; thus, the model can dynamically adjust the hidden state according to the time weight to balance the influence of short-term and long-term interests. S203. Dynamic update mechanism for user profiles: The core of this mechanism is to form a dynamic adjustment capability for user profiles by integrating real-time preference prediction and long-term preference prediction; compared with traditional static profiles, this mechanism can adaptively update user profiles and enhance the agent's ability to respond to changes in user interests. Real-time preference prediction: P real-time =softmax(W r ·h T +b r ) (6) Where P real-time W represents the real-time preference vector, which in turn represents the current distribution of user interests. r The weight matrix represents the real-time preference prediction, used to convert hidden states into preference probabilities; h T The hidden state of the latest time step reflects the user's current behavioral characteristics; b r This is a bias term used to adjust the baseline value for preference prediction; At this point, the intelligent agent can predict the distribution of user preferences for various interest tags in real time based on the latest behavioral data; Long-term preference prediction: This is the updated long-term preference vector. Let P be the long-term preference vector from the previous time step. real-time γ represents the current real-time preference vector, and γ is the long-term preference update weight, ranging from 0 to 1, used to balance real-time preferences with historical trends. Long-term preference predictions are gradually adjusted based on current prediction results to ensure that the profile reflects users' immediate interests while maintaining long-term stability. Global user profile update: P global =δ·P real-time +(1-δ)·P long-term (8) P global A global user profile is created, integrating short-term interests and long-term trends, which serves as the final basis for the agent's response; δ is the fusion weight, ranging from 0 to 1, which adjusts the ratio of short-term preferences to long-term preferences; By creating a global profile, the intelligent agent can integrate short-term and long-term interests to provide users with more accurate response services. S204. Hierarchical feature fusion: It is divided into two categories: one is intra-sequence feature extraction, and the other is cross-sequence feature fusion. Intra-sequence feature extraction: An intra-sequence attention mechanism is applied individually to each behavior sequence to calculate the importance weight of each action within the sequence, thereby identifying key actions in the user's behavior sequence. Specifically, the attention score is calculated using the following formula: in This is the final output state of the feedforward GRU model. This is the t-th output state of the bidirectional GRU model. The attention score a is calculated using this formula. t Able to reflect action h t With the current action The relationship between the two-way GRU and the corresponding attention score vector is used to determine the important actions; then the output of the bidirectional GRU is multiplied element-wise with the corresponding attention score vector to obtain the output of the attention layer within the sequence. Cross-sequence feature fusion: Utilizing a cross-sequence attention mechanism to fuse interactive features between different behavioral sequences; using a sliding interactive view v s and browsing interactive view v b For example, the cross-view attention mechanism IA(v s ,v b ) = A s (v b ,v s ,v s )⊙A b (v s ,v b ,v b in This mechanism can effectively capture the correlation and interaction patterns between asynchronous behavior sequences, and obtain the output of the cross-view attention layer. S205. Multi-task optimization: A multi-task learning framework is designed to improve the ability to identify user behavior patterns by simultaneously predicting real-time and long-term preferences; the feature embedding vectors of different behavior sequences are fused by concatenation and then input into the multi-task learning network; loss functions for real-time preference prediction and long-term preference prediction tasks are defined respectively. Loss function L for real-time preference prediction task short : Loss function L for long-term preference prediction task long : Where y is the actual reading label, and N is the total number of samples in the dataset. It is the training set; Global loss function L global Defined as a weighted sum of the losses from the two tasks, to achieve multi-task learning: THE global =λL short +(1-λ)L long (12) Here, λ is a hyperparameter used to balance the importance of the two tasks.

3. The social intelligent agent cluster control method according to claim 1, characterized in that, Step S1, which involves statistically analyzing different user groups on the internet platform and constructing corresponding user feature profiles, includes: After creating the intelligent agent, user profile development is performed based on the set user profile to obtain an intelligent agent with preferences. This includes: acquiring users' historical behavior data, access trajectories, content browsing records, and social interaction information for multi-dimensional data analysis; the analysis dimensions may include, but are not limited to: users' interests, occupational information, social circles, purchasing preferences, and device usage habits; then, based on users' geographical location, age group, gender, and income level, combined with significant features in their historical behavior, segmented user profiles are formed; by classifying and aggregating users according to multiple feature dimensions, high-value user groups and potential interest groups can be identified; finally, user profile features for different user groups are constructed.

4. The social intelligent agent cluster control method according to claim 1, characterized in that, Step S3, which involves creating information consumption activity tasks for the intelligent agent based on a B / S architecture, includes: creating and managing tasks in a user-friendly web interface, and utilizing the powerful processing capabilities of the server to allocate and execute tasks; the user creates new intelligent agent tasks in an intuitive task management interface and selects a suitable intelligent agent from a predefined pool of intelligent agents to perform the task; during task creation, the user sets task parameters, including the target platform, the scheduled execution time, the specific description of the task, and the expected results; and the user selects a specific type of intelligent agent to execute the task according to its specific requirements.

5. The social intelligent agent cluster control method according to claim 1, characterized in that... Step S4, which involves collecting information consumption data generated by the agent's activities, includes: recording information data received by the agent during access to an internet media platform, including content type, source, publication time, and interaction volume; for content that is viewed in detail, collecting comment-related data, including comment content, author, time, and interaction volume; and the agent controller receiving and storing this data, and performing subsequent analysis and querying based on the collected behavior record metadata.

6. A social intelligent agent cluster control device, characterized in that, include: The intelligent agent module is a web automation robot based on Autojs or Jxbrowser, distributed on various public and private cloud hosts or mobile phones; it is an automated program that can autonomously consume information and engage in social behavior such as browsing, reading, following, commenting, and liking on designated social media according to the assigned task information, and also has the ability of text classification, content monitoring and image recognition. The task assignment module assigns tasks to one or a group of social AI agents based on execution time and related attributes. The task collection module collects relevant task feedback and data; and an automated program handles errors during task execution. The message broker module includes a task queue unit and a task feedback unit, decoupling the relationship between the agent and the task allocator; it distributes different types of tasks through a single queue or a batch-distributed topic, and provides feedback to the social agent through a queue. The data storage module stores social agent task information, feedback information, browsing behavior information, social agent status information, and agent profile information; After a user publishes a task, the task information will be stored in the task information table in the data storage module; The task allocation module reads task information and sends the task to the task queue unit in the message broker module; the agent module retrieves the task from the task queue unit and executes it, and after execution, sends the task result to the feedback queue unit in the message broker module; the task collection module reads the task feedback information from the feedback queue unit and stores the feedback information, browsing behavior information, profile information, and social agent information in the data storage module; in addition, the agent module performs separate backup storage of all information in the data storage module.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the social intelligent agent cluster control method according to any one of claims 1 to 4.

Images