A multi-agent public opinion evolution simulation method and system for sudden public events

Abstract

The application discloses a kind of multi-agent public opinion evolution simulation method and system for sudden public event, solve the technical problem that the existing public opinion evolution simulation method causes serious computing redundancy and resource idling waste, finally directly lead to simulation efficiency is greatly reduced.Method includes obtaining sudden public event original data and preprocessing, get the event-driven input structure data suitable for continuous time simulation;Subsequently complete public opinion environment construction, user agent initialization, then generate conversation scheduling data using preset activity intensity function, obtain updated user state data after processing by preset perception-memory-reflection update model, finally complete simulation by preset public opinion stage evaluation function, obtain final simulation result data.

Description

Technical Field

[0001] This invention relates to the field of online public opinion evolution technology, and in particular to a multi-agent simulation method and system for public opinion evolution in the face of sudden public events. Background Technology

[0002] As social media has become the core space for information dissemination, emotional expression, and opinion competition in public emergencies, how to perform computable and interpretable simulations of public opinion evolution in the early, middle, and reactivation stages of an event has become an important technical direction for research on public opinion governance and crisis communication. Traditional agent-based models typically control information dissemination and state changes between individuals through a number of predefined rules, and can describe basic processes such as diffusion and aggregation. However, in open platform environments, it is difficult to uniformly model natural language comments, complex emotions, heterogeneous users, and authoritative information intervention.

[0003] In recent years, intelligent agent systems powered by large language models have provided new technological pathways for social simulation. One type of research improves the semantic richness of user behavior generation by embedding perception, memory, and action modules into intelligent agents; another type of research combines time-aware mechanisms, environmental structures, and multi-agent interactions to simulate the spread of hot topics, social movements, or rumors.

[0004] Existing public opinion evolution simulation methods generally adopt the principle of fixed simulation density, maintaining a uniform user participation scale, session sampling frequency, and computing resource allocation strategy throughout the entire simulation cycle, without designing differentiated control logic for the stage characteristics of public opinion development. This fixed operation mode cannot adapt to the dynamic changes of public opinion from outbreak, peak to decay and low activity period. In simulation scenarios with long window or large user scale, even when entering the low activity stage of public opinion, the same amount of computing resources are still invested to maintain high-frequency session sampling and full user participation, resulting in a large number of invalid and duplicate calculations, causing serious computational redundancy and resource waste, and ultimately directly leading to a significant decrease in simulation efficiency. Summary of the Invention

[0005] This invention provides a multi-agent public opinion evolution simulation method and system for sudden public events, which solves the technical problem that existing public opinion evolution simulation methods cause serious computational redundancy and resource waste, ultimately leading to a significant decrease in simulation efficiency.

[0006] The first aspect of this invention provides a multi-agent simulation method for public opinion evolution in the face of sudden public events, comprising:

[0007] Acquire raw data of public emergencies and preprocess the raw data to obtain event-driven input structure data suitable for continuous-time simulation;

[0008] The event-driven input structure data suitable for continuous-time simulation is used to construct a public opinion environment to obtain the current public opinion environment data;

[0009] User agent initialization is performed on the user group configuration data in the event-driven input structure data suitable for continuous-time simulation to obtain hierarchical user agent initial data.

[0010] Using a preset activity intensity function, session scheduling is generated from the event-driven input structure data suitable for continuous-time simulation, the current public opinion environment data, and the initial data of the hierarchical user agent to obtain dynamic session scheduling data.

[0011] A preset perception-memory-reflection update model is used to perform perception and reflection processing on the dynamic session scheduling data, the current public opinion environment data, and the initial data of the hierarchical user intelligent agent to obtain updated user status data.

[0012] A preset public opinion stage evaluation function is used to simulate the updated user status data, the current public opinion environment data, and the event-driven input structure data suitable for continuous-time simulation, so as to obtain the final simulation result data.

[0013] Optionally, the preprocessing of the raw data of the sudden public event to obtain event-driven input structure data suitable for continuous-time simulation includes:

[0014] The raw data of the public emergency is format-checked and invalid information is removed, and the preliminarily cleaned input data is output.

[0015] The input data after initial cleaning is processed to standardize the fields, and the standardized input data is output.

[0016] The input data after the fields are standardized is then normalized to obtain event-driven input structure data suitable for continuous-time simulation.

[0017] Optionally, the step of constructing a public opinion environment from the event-driven input structure data suitable for continuous-time simulation to obtain current public opinion environment data includes:

[0018] Extract event main post data from the event-driven input structure data suitable for continuous-time simulation, initialize the main post object of the event main post data, and output the initialized main post data;

[0019] The initial main post data is used to initialize the comment list, and the initial comment list data is output.

[0020] The initial comment list data is used to initialize the reply list, and the initial reply list data is output.

[0021] Establish a parent-child relationship between the initial comment list data and the initial reply list data, and output the comment reply association data;

[0022] The official posting records for the comment reply association data are initialized, and the initialized official posting record data is output.

[0023] The initial values ​​for aggregation and statistics are set for the official release record data, and the initial aggregated statistical data is output.

[0024] The initialization aggregate statistics are used to initialize the environment log framework, and the initialization environment log data is output.

[0025] By integrating the initial main post data, the comment reply association data, the initial official release record data, the initial aggregated statistics data, and the initial environment log data, the current public opinion environment data is obtained.

[0026] Optionally, the initialization of user agents in the user group configuration data of the event-driven input structure data suitable for continuous-time simulation to obtain hierarchical user agent initial data includes:

[0027] Extract basic user configuration parameters from the user group configuration data;

[0028] The user basic configuration parameters are used to segment user groups into different strata, and user segmentation data is output.

[0029] User profiles are generated for the users in the user hierarchical data, and the user profile data is output.

[0030] The user profile data is initialized for long-term memory, and the user long-term memory data is output.

[0031] The user's long-term memory data is initialized with short-term memory, and the user's short-term memory data is output.

[0032] The user's short-term memory data is assigned a value, and the user's initial state data is output.

[0033] By integrating the user hierarchical segmentation data, the user profile data, the user long-term memory data, the user short-term memory data, and the user initial state data, the initial data of the hierarchical user agent is obtained.

[0034] Optionally, the step of using a preset activity intensity function to generate dynamic session scheduling data by processing the event-driven input structure data suitable for continuous-time simulation, the current public opinion environment data, and the initial data of the hierarchical user agent, includes:

[0035] The preset activity intensity function is used to sample the first session start time of the hierarchical user agent's initial data based on the current public opinion environment data and the simulation control parameters in the event-driven input structure data suitable for continuous time simulation, and the user's first session time data is output.

[0036] Assign a unique identifier to each session in the user's first session time data and output the session unique identifier data;

[0037] The session unique identifier data is used to associate and bind user intelligent agents, and the session user association data is output.

[0038] The session duration is recorded for the associated user data, and basic session information data is output.

[0039] A dynamic priority queue is constructed based on the basic session information data, and the initial priority queue data is output.

[0040] The initial priority queue data is sorted according to the session start timestamp, and the time-series session queue data is output.

[0041] By integrating the time-series session queue data and the session user association data, dynamic session scheduling data is obtained.

[0042] Optionally, the step of employing a preset perception-memory-reflection update model to perform perception and reflection processing on the dynamic session scheduling data, the current public opinion environment data, and the initial data of the hierarchical user agent to obtain updated user state data, including:

[0043] The current timestamp is matched with the current session in the time-series session queue of the dynamic session scheduling data, and the user session data to be executed is output.

[0044] Schedule the user intelligent agent corresponding to the user session data to be executed, and read the visible information of the current public opinion environment data, and output the visible information data of the environment;

[0045] The environmental visible information data is processed by extracting text semantics, sentiment, and publishing subject attributes to output user perception feature data.

[0046] Long-term memory retrieval is performed on the user perception feature data and the initial data of the hierarchical user agent to output the user's historical memory data;

[0047] Using the preset perception-memory-reflection update model, real-time impressions are generated from the user's perceptual feature data and the user's historical memory data, and the user's real-time impression data is output.

[0048] The user's real-time impression data is corrected using the preset perception-memory-reflection update model, and the updated user status data is output.

[0049] Optionally, the step of using a preset public opinion stage evaluation function to simulate the updated user status data, the current public opinion environment data, and the event-driven input structure data suitable for continuous-time simulation, to obtain the final simulation result data, including:

[0050] The current public opinion environment data is used as the simulation benchmark environment to make user behavior decisions based on the updated user status data and output user behavior data to be executed.

[0051] Update the public opinion environment data of the user's pending actions and output the updated public opinion environment data.

[0052] Extract basic data for phase assessment from the updated public opinion environment data;

[0053] Using the preset public opinion stage evaluation function, the public opinion stage is determined based on the stage evaluation basic data, and the public opinion stage determination result data is output.

[0054] Based on the public opinion stage determination result data and the pruning threshold in the event-driven input structure data applicable to continuous time simulation, the scheduling parameters are adjusted, and the pruned scheduling control data is output.

[0055] Based on the post-pruning scheduling control data, perform a cyclic simulation and output complete simulation process data;

[0056] The complete simulation process data is integrated and encapsulated according to the output format parameters in the event-driven input structure data suitable for continuous-time simulation to obtain the final simulation result data.

[0057] The second aspect of this invention provides a multi-agent public opinion evolution simulation system for sudden public events, comprising:

[0058] The acquisition module is used to acquire raw data of public emergencies and preprocess the raw data of public emergencies to obtain event-driven input structure data suitable for continuous-time simulation.

[0059] The construction module is used to construct the public opinion environment from the event-driven input structure data suitable for continuous-time simulation, and obtain the current public opinion environment data.

[0060] An initialization module is used to initialize the user agent in the user group configuration data in the event-driven input structure data suitable for continuous-time simulation, so as to obtain the hierarchical user agent initial data.

[0061] The scheduling module is used to generate dynamic session scheduling data by using a preset activity intensity function to process the event-driven input structure data suitable for continuous-time simulation, the current public opinion environment data, and the initial data of the hierarchical user agent.

[0062] The perception and reflection module is used to perform perception and reflection processing on the dynamic session scheduling data, the current public opinion environment data, and the initial data of the hierarchical user intelligent agent using a preset perception-memory-reflection update model to obtain updated user status data.

[0063] The simulation module is used to simulate the updated user status data, the current public opinion environment data, and the event-driven input structure data suitable for continuous-time simulation using a preset public opinion stage evaluation function, so as to obtain the final simulation result data.

[0064] The third aspect of the present invention provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the multi-agent public opinion evolution simulation method for sudden public events as described above.

[0065] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed, it implements the multi-agent public opinion evolution simulation method for sudden public events as described above.

[0066] As can be seen from the above technical solutions, the present invention has the following advantages:

[0067] The above-mentioned technical solution of the present invention provides a multi-agent public opinion evolution simulation method for public emergencies. It acquires raw data of the public emergency and preprocesses the raw data to obtain event-driven input structure data suitable for continuous-time simulation. It then constructs a public opinion environment from the event-driven input structure data suitable for continuous-time simulation to obtain current public opinion environment data. Finally, it initializes user agents from the user group configuration data in the event-driven input structure data suitable for continuous-time simulation to obtain hierarchical user agent initial data. Using a preset activity intensity function, it generates dynamic session scheduling data from the event-driven input structure data suitable for continuous-time simulation, the current public opinion environment data, and the hierarchical user agent initial data. Using a preset perception-memory-reflection update model, it performs perception and reflection processing on the dynamic session scheduling data, the current public opinion environment data, and the hierarchical user agent initial data to obtain updated user state data. Finally, it uses a preset public opinion stage evaluation function to evaluate the updated user state data and the current public opinion environment data. Simulations are performed using pre-public opinion environment data and event-driven input structure data suitable for continuous-time simulation to obtain the final simulation results. Based on the above scheme, this invention obtains standardized event-driven input structure data by acquiring and preprocessing the original data of sudden public events. Then, it constructs current public opinion environment data and builds a unified and centralized public opinion interaction space. Initial data of hierarchical user agents is generated through user agent initialization to achieve differentiated modeling of user groups. Then, dynamic session scheduling data is generated by combining event-driven input, public opinion environment, and hierarchical user data through a preset activity intensity function to achieve temporal and differentiated scheduling of user sessions. Subsequently, a preset perception-memory-reflection update model is used to complete user perception and reflection processing and output updated user status data. Finally, simulation is carried out based on a preset public opinion stage evaluation function. The scheduling parameters can be adjusted in reverse according to the public opinion dynamics, and dynamic pruning can be implemented. In the low-activity stage of public opinion, the simulation density can be accurately reduced and non-critical computing paths can be reduced, effectively reducing the amount of invalid computation and improving resource utilization, directly achieving a significant improvement in simulation efficiency. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0069] Figure 1 This is a flowchart illustrating the steps of a multi-agent public opinion evolution simulation method for sudden public events, as provided in Embodiment 1 of the present invention.

[0070] Figure 2This is an overall framework diagram of a multi-agent public opinion evolution simulation method for sudden public events provided in Embodiment 1 of the present invention;

[0071] Figure 3 This is a structural block diagram of a multi-agent public opinion evolution simulation system for sudden public events, provided in Embodiment 2 of the present invention. Detailed Implementation

[0072] This invention provides a multi-agent public opinion evolution simulation method and system for sudden public events, which solves the technical problem that existing public opinion evolution simulation methods cause serious computational redundancy and resource waste, ultimately leading to a significant decrease in simulation efficiency.

[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that in the optional embodiments of the present invention, the object information and other related data involved require the permission or consent of the object when the embodiments of the present invention are applied to specific products or technologies, and the collection, use, and processing of related data must comply with relevant laws, regulations, and standards. That is to say, if the embodiments of the present invention involve data related to the object, it needs to be obtained with the authorization and consent of the object, the authorization and consent of relevant departments, and in compliance with relevant laws, regulations, and standards. If personal information is involved in the embodiments, the acquisition of all personal information requires the consent of the individual. If sensitive information is involved, the separate consent of the information subject is required, and the embodiments also need to be implemented with the authorization and consent of the object.

[0074] Terminology Explanation:

[0075] Multi-agent system: refers to a collaborative simulation system composed of multiple intelligent agents with relatively independent perception, memory, decision-making and action capabilities.

[0076] Public emergencies refer to accidents, disasters, public safety incidents, social conflicts, or service crises that occur within a short period of time and can trigger widespread public attention, discussion, and emotional fluctuations.

[0077] Public opinion environment: refers to the centralized information interaction space built around a main post of a single event, which includes at least the main post, the comment list, the reply structure, official releases, and aggregated statistics.

[0078] Time-aware scheduling: refers to a simulation scheduling method that organizes user entry, browsing, interaction, and exit according to specific timestamps rather than synchronous rounds.

[0079] Official publishers: refers to dedicated nodes that simulate official accounts of government departments, authoritative institutions, enterprises, or mainstream media to inject authoritative information into the public opinion environment at specific times.

[0080] Reflection mechanism: refers to the mechanism by which an intelligent agent updates its current cognitive state, emotional state, and behavioral tendencies after receiving external information or completing an interaction.

[0081] Pruning mechanism: refers to the mechanism that improves simulation efficiency by reducing participation density, reducing repeated sessions, or compressing non-critical computational paths during periods of low public opinion activity.

[0082] Please see Figure 1 , Figure 1 The flowchart illustrates the steps of a multi-agent public opinion evolution simulation method for sudden public events, as provided in Embodiment 1 of the present invention.

[0083] This invention provides a multi-agent simulation method for public opinion evolution in the face of sudden public events, comprising:

[0084] Step 101: Obtain the raw data of the public emergency and preprocess the raw data to obtain event-driven input structure data suitable for continuous-time simulation.

[0085] Raw data of public emergencies refers to various basic data related to public emergencies that are collected initially without processing. Specifically, it includes event main post data, event development timeline data, official release configuration data, user group configuration data, and simulation control parameters, providing raw information sources for subsequent data preprocessing and public opinion simulation.

[0086] It should be noted that, firstly, raw information containing event basics, configurations, and time series-related information is collected. Then, through verification, cleaning, and standardization, the unstructured raw data is converted into structured input data adapted for time series simulation, providing a standardized data foundation for subsequent public opinion simulation.

[0087] Further, step 101 may include the following sub-steps:

[0088] S11. Perform format verification and invalid information removal on the raw data of public emergencies, and output the pre-cleaned input data.

[0089] S12. Perform field standardization processing on the input data after initial cleaning, and output the input data after field standardization.

[0090] S13. Standardize the input data after field standardization to obtain event-driven input structure data suitable for continuous-time simulation.

[0091] It should be noted that this invention receives at least the following data:

[0092] (1) Event main post data: including title, summary, body, event tags and initial posting time, used to constitute the core discussion object in the public opinion environment;

[0093] (2) Event development timeline data: including key progress nodes arranged in chronological order, each node containing at least a timestamp, progress text, node type, and impact level;

[0094] (3) Official release configuration data: including the identity of the official releaser, release time, release content, release priority and its corresponding event nodes;

[0095] (4) User group configuration data: including number of users, user segmentation ratio, user attribute set, participation probability and session length parameters;

[0096] (5) Simulation control parameters: including time window length, activity intensity function, number of comments displayed, number of replies displayed, pruning threshold and output format, etc.

[0097] Preferably, the main event post, the event timeline, and the official release configuration together constitute an "event-driven input structure," so that the system no longer runs around a single static topic text, but simulates a continuous chain of "event subject - progress nodes - authoritative response."

[0098] Specifically, the first step involves format validation and invalid information removal of the raw data for public emergencies, outputting pre-cleaned input data. This step processes the raw data, including event main posts, event timeline data, official release configuration data, user group configuration data, and simulation control parameters: For event main posts, the format of fields such as title, summary, body, and initial release time is validated, removing invalid posts with garbled characters, incorrect timestamps, or duplicates; for the event timeline data, key progress nodes arranged chronologically are validated, removing nodes with conflicting timestamps, unclear node types, or invalid impact levels; for official release configuration data, the publisher's identity, release time, release content, release priority, and corresponding event nodes are validated, removing configuration items with no valid release time, empty content, or incorrect associated nodes; for user group configuration data, parameters such as the number of users, stratification ratio, and participation probability are validated, removing abnormal configurations with values ​​exceeding reasonable ranges; and for simulation control parameters, such as the time window length and pruning threshold, invalid parameters that do not conform to simulation logic are validated, ultimately resulting in pre-cleaned input data with compliant format and no redundant or invalid information. Next, the input data after initial cleaning is processed to standardize the fields, and the input data is output after the fields are standardized. This step standardizes the field formats and definitions of various types of data after initial cleaning: all text fields in the main event post data are standardized to the standard encoding format, and time fields are standardized to the standard time format; the timestamps in the event development timeline data are standardized to the standard format, and fields such as node type and impact level are standardized with encoding rules; the release priority and event node identifier fields in the official release configuration data are standardized with encoding; the user stratification ratio and participation probability fields in the user group configuration data are standardized with numerical formats; at the same time, the field naming, units and value rules of the simulation control parameters are aligned to ensure that all data fields are clearly defined and uniformly formatted, and the input data is output after the fields are standardized. Finally, the input data after field standardization is standardized to obtain event-driven input structure data suitable for continuous-time simulation. This step integrates and standardizes various types of data after field standardization. Among them, the event main post data, event development timeline data, and official release configuration data are integrated into an "event-driven input structure". A standardized data organization form is constructed around the continuous chain of "event body - progress node - authoritative response", so that the data no longer exists in the form of static and isolated topic text, but adapts to the temporal progression logic of continuous-time simulation. At the same time, user group configuration data and simulation control parameters are adapted to this structure and standardized. Finally, a structured input set containing a complete event chain, user configuration, and simulation parameters is formed, resulting in event-driven input structure data suitable for continuous-time simulation.

[0099] Step 102: Construct a public opinion environment from event-driven input structure data suitable for continuous-time simulation to obtain the current public opinion environment data.

[0100] It should be noted that, based on the event-driven input structure data suitable for continuous-time simulation obtained from the aforementioned preprocessing, the event main post data, event development timeline data, and official release configuration data are extracted to build a basic interactive framework that carries the core discussion objects, time sequence progress nodes, and authoritative information, thus completing the initial construction of the public opinion environment and obtaining the current public opinion environment data.

[0101] Furthermore, step 102 may include the following sub-steps:

[0102] S21. Extract event main post data from event-driven input structure data suitable for continuous-time simulation, initialize the main post object of the event main post data, and output the initialized main post data.

[0103] S22. Initialize the comment list based on the initial main post data, and output the initial comment list data;

[0104] S23. Initialize the reply list based on the initial comment list data, and output the initial reply list data;

[0105] S24. Establish a parent-child relationship between the initial comment list data and the initial reply list data, and output the comment-reply association data;

[0106] S25. Initialize the official posting records for the comment reply related data, and output the initialized official posting record data;

[0107] S26. Set initial values ​​for aggregated statistics of the initial official release record data, and output the initial aggregated statistics data;

[0108] S27. Initialize the environment log framework for the initial aggregated statistical data and output the initial environment log data;

[0109] S28. Integrate and initialize main post data, comment reply related data, official release record data, aggregated statistical data, and environment log data to obtain the current public opinion environment data.

[0110] The current public opinion environment data, in the simulation startup phase, uses the initialized public opinion environment data as the initial baseline form. This integrates the core discussion objects, event timeline progression nodes, and authoritative release-related basic information required in the early stages of the simulation, providing basic environmental support for user interaction activities and the orderly progress of the simulation process in the initial stage. After entering the formal simulation operation phase, throughout the entire process of user agent information perception and reflection, user behavior decision-making and execution, and the full-process simulation cycle, this type of data will continuously evolve dynamically beyond its initial fixed state. On the one hand, it updates the content structure and interaction statistics in real time following user comments, replies, likes, and reposts; on the other hand, it combines preset timeline nodes to complete the input and update of official release information. Through this dual update method, the environmental data content is continuously iterated and optimized to accurately adapt to the constantly changing real-time public opinion state during the simulation process. This type of data specifically includes initialized main post data, comment and reply related data, initialized official release record data, initialized aggregated statistical data, and initialized environment log data. Among them, the initialized main post data is the core discussion object data constructed from the corresponding event main post data, including the main post title, summary, body, event tags, initial release time, and unique main post identifier; comments... The reply-related data is a collection of initialized comment list data, initialized reply list data, and the parent-child relationships between them. The initialized comment list data is the comment storage framework for the main post in the initial simulation state, and the initialized reply list data is the reply storage framework for each comment. The parent-child relationships define the hierarchical structure of comments and replies. The initialized official release record data is the initial record of authoritative information releases constructed according to the official release configuration data, including the preset official releaser identity, release time, release content, release priority, and corresponding event node identifier. The initialized aggregated statistical data is a set of public opinion statistics items in the initial simulation state, including initial values ​​for statistical indicators such as discussion volume, popularity value, user participation, and the proportion of positive and negative sentiment. The initialized environment log data is the initial configuration of the log framework for recording changes in public opinion status during the simulation, including the initial definitions of log storage rules, recording dimensions, and timestamp format.

[0111] Initializing the main post data refers to the structured data obtained after the main post object is initialized. It stores the title, summary, body, event tags, initial posting time, and unique identifier of the main post in the form of an object, providing anchor points for subsequent comments, replies, and other interactions.

[0112] Initializing the comment list data refers to the comment storage framework data obtained after the comment list has been initialized. It reserves space for fields such as unique identifier, posting time, poster identifier, and interaction count for each comment to support the storage of subsequent comment content.

[0113] Initialized reply list data refers to the reply storage framework data obtained after the reply list initialization process. It reserves storage fields for the reply content under each comment to support the recording of subsequent reply information.

[0114] Comment reply related data refers to the structured data obtained after the parent-child relationship is established. It includes the initial comment list data, the initial reply list data, and the parent-child relationship between the two, forming a hierarchical storage structure of "main post-comment-reply".

[0115] Initializing official release record data refers to the authoritative information storage framework data obtained after the official release record initialization process, which reserves storage fields for information such as injection time, content, related event nodes and release priority of official releases.

[0116] Initialized aggregated statistical data refers to the statistical object data obtained after the initial value setting process of aggregated statistics. It includes the initial values ​​of statistical indicators such as total comments, total replies, interaction volume, average sentiment, and stage status, which provide the basis for subsequent public opinion data statistics in the simulation process.

[0117] Initialized environment log data refers to the log framework data obtained after initialization processing by the environment log framework. It reserves recording fields for events such as changes in the state of the public opinion environment and information interaction during subsequent simulation processes, supporting the backtracking and management of the simulation process.

[0118] It should be noted that this invention establishes a centralized information environment around a single public emergency. Unlike a decentralized dissemination environment based on user relationship networks, this invention preferably adopts an environment organization method of "single event center + comment / reply hierarchical structure". The environment maintains at least the following objects: event main post object, comment list object, reply list object, official release record object, aggregate statistics object, and environment log object.

[0119] The event main post object stores the title, body, summary, and metadata; the comment list object records comments posted on the main post, with each comment having a unique identifier, posting time, poster identifier, and interaction count; the reply list object records replies to comments and establishes a relationship with the comment object through parent node relationships; the official posting record object records the time, content, and associated nodes of authoritative information injection; and the aggregate statistics object stores the total number of comments, total number of replies, interaction volume, average sentiment, and stage status in the current environment.

[0120] Once the environment is built, all subsequent participants will conduct information observation and behavioral decisions around this unified environment, thereby ensuring that the discussion space during the simulation process is focused, clearly structured, and easy to manage at each stage.

[0121] Specifically, the process involves extracting main event post data from event-driven input structure data suitable for continuous-time simulation, initializing the main event post data into post objects, and outputting the initialized main event post data. This step focuses on building a core anchor point for a centralized information environment around a single sudden public event. It encapsulates information such as the title, summary, body text, event tags, and initial publication time from the main event post data into main event post objects, saving the corresponding text content and metadata, providing a core discussion object for the entire public opinion environment. The next step is to initialize the comment list based on the initialized main event post data, outputting the initialized comment list data. This step creates comment list objects for the event post objects corresponding to the initialized main event post data, building a comment storage framework for the main event post, and storing comments for each post. The system reserves space for fields such as unique identifier, posting time, poster identifier, and interaction count for comments. It initializes the reply list by initializing the comment list data and outputs the initial reply list data. This step creates a corresponding reply list object for each comment in the initial comment list data, establishing a reply storage framework under the comment and reserving fields for storing subsequent reply content. It establishes a parent-child relationship between the initial comment list data and the initial reply list data, outputting comment-reply association data. This step configures parent node relationships for each reply list object in the initial reply list data, binding them to the corresponding comment list object and constructing a hierarchical structure of "main post - comment - reply". The system then performs official processing on the comment-reply association data. The first step initializes the official release record, outputting initial official release record data. This step creates an official release record object based on the official release configuration data in the event-driven input structure data, establishing a storage framework for authoritative information and reserving storage fields for information such as the injection time, content, related event nodes, and priority of official releases. The second step sets initial values ​​for aggregated statistics on the initialized official release record data, outputting initial aggregated statistical data. This step creates an aggregated statistical object, setting initial values ​​for statistical indicators such as total comments, total replies, interaction volume, average sentiment, and stage status, providing a foundation for dynamic statistics of public opinion data during subsequent simulations. The third step initializes the environment log framework for the initial aggregated statistical data, outputting the initial environment log. The first step involves creating an environment log object to establish a log recording framework for changes in the public opinion environment. This reserves recording fields for events such as state changes and information interactions of various objects during subsequent simulations. The second step integrates and initializes main post data, comment reply related data, official release record data, aggregated statistical data, and environment log data to obtain the current public opinion environment data. This step consolidates the aforementioned initialized objects into a centralized information environment, adopting a "single event center + comment / reply hierarchical structure" environment organization method. All user interactions, authoritative information injections, and public opinion status updates in subsequent simulations revolve around this unified environment, ensuring a centralized discussion space, clear structure, and easy phase management during the simulation process.

[0122] Step 103: Initialize the user agent in the user group configuration data in the event-driven input structure data suitable for continuous-time simulation to obtain the initial data of the hierarchical user agent.

[0123] The initial data of hierarchical user agents refers to the structured data set obtained after the user agents have been initialized. It includes user agent instances of different influence levels and their initial attributes. Each agent instance is configured with user identifier, level, participation probability, session length range and initial state parameters.

[0124] It should be noted that, based on the event-driven input structure data suitable for continuous-time simulation obtained above, user group configuration data is extracted from it. According to the parameters of user number, user stratification ratio, user attribute set, participation probability and session length, user agent instances of different levels are constructed and initial attributes are configured to obtain the initial data of stratified user agents.

[0125] Furthermore, step 103 may include the following sub-steps:

[0126] S31. Extract basic user configuration parameters from user group configuration data;

[0127] S32. Perform user group segmentation based on user basic configuration parameters and output user segmentation data;

[0128] S33. Generate user profiles for users in the user stratification data and output the user profile data;

[0129] S34. Initialize the user profile data for long-term memory and output the user's long-term memory data.

[0130] S35. Initialize the user's long-term memory data for short-term memory and output the user's short-term memory data.

[0131] S36. Assign values ​​to the user's short-term memory data and output the user's initial state data;

[0132] S37. Integrate user hierarchical segmentation data, user profile data, user long-term memory data, user short-term memory data, and user initial state data to obtain hierarchical user agent initial data.

[0133] User basic configuration parameters refer to the set of basic parameters extracted from user group configuration data and used for creating user intelligent agents. These parameters include the number of users, user stratification ratio, user attribute set, participation probability, and session length.

[0134] User segmentation data refers to the segmentation results obtained after user group segmentation processing, which includes user segmentation identifiers and corresponding configuration parameters at different influence levels.

[0135] User profile data refers to structured data obtained after user profile generation and processing, including basic attribute information such as demographic attributes, social media usage attributes, account influence attributes, and user type identifiers in the user profile sub-module.

[0136] User long-term memory data refers to the initial data of the user's long-term memory portion obtained after long-term memory initialization processing, which is used to store the user's stable preferences, past experiences, and general stance.

[0137] User short-term memory data refers to the initial data of the user's short-term memory portion obtained after short-term memory initialization processing. It is used to record changes in the user's recent interaction state and to provide a basis for generating current impression memories.

[0138] User initial state data refers to the initial state parameters of the user agent obtained after assigning values ​​to the user's short-term memory data. These parameters include the user's initial participation probability, the range of session length, and initial emotions and viewpoints, providing an initial state configuration for the user to enter the simulation.

[0139] It should be noted that this invention uses a user intelligent agent module to generate and drive different types of participants. Each user intelligent agent includes at least a user profile submodule, a perception submodule, a memory submodule, a reflection submodule, and an action submodule.

[0140] The user profile submodule provides relatively stable basic attributes, including at least demographic attributes, social media usage attributes, account influence attributes, and user type identifiers. Preferably, users are categorized by influence into high-influence users, medium-influence users, and ordinary users, with different parameters configured for initial sampling ratios, participation probabilities, behavioral activity levels, and text styles for different user categories.

[0141] The Perception submodule reads the currently visible content after a user enters the environment. Visible content includes at least the main event post, the top-ranked comments, replies related to the target comment, and official posts injected after the user's most recent session. The Perception submodule then passes the raw observations to the Memory and Reflection submodules for subsequent impression generation and decision-making.

[0142] The memory submodule preferably includes long-term memory, short-term memory, and current impression memory. Long-term memory is used to represent the user's stable preferences, past experiences, and general stance; short-term memory is used to record emotional fluctuations, changes in viewpoints, and social confidence in recent interactions; and current impression memory is used to store the immediate understanding formed after entering the environment in the current round.

[0143] The action submodule outputs specific behaviors, including at least posting a comment, replying to a comment, liking, sharing, viewing more comments, and exiting. Different users can generate different action sequences based on their current status and visible information. High-influence users are preferred to have a higher probability of original comments, medium-influence users are preferred to have a higher probability of commenting and replying, and ordinary users are preferred to have a higher probability of viewing and liking.

[0144] Specifically, the process involves extracting basic user configuration parameters from user group configuration data. These parameters include user quantity, user segmentation ratio, user attribute set, participation probability, and session length, providing a configuration basis for creating user agents. Next, user groups are segmented based on these basic configuration parameters, outputting user segmentation data. This step categorizes users into high-influence, medium-influence, and ordinary users based on their influence, configuring different parameters for initial sampling ratio, participation probability, behavioral activity, and text style for each user level, resulting in a segmented user group. User profiles are generated from the segmented user data, outputting user profile data. This step creates a user profile submodule for each user level, configuring basic attributes such as demographic attributes, social media usage attributes, account influence attributes, and user type identifiers, forming stable basic user profile information. Finally, long-term memory initialization is performed on the user profile data, outputting user long-term memory data. This step initializes the long-term memory portion of the memory submodule for each user agent, used for table... The system displays stable user preferences, past experiences, and general stances, providing a long-term, stable cognitive foundation for the user agent. It initializes short-term memory for the user's long-term memory data, outputting short-term memory data. This step initializes the short-term memory portion of the memory submodule for each user agent, recording emotional fluctuations, changes in viewpoints, and social confidence during recent interactions, while reserving storage space for subsequent generation of current impression memories. The system assigns values ​​to the user's short-term memory data, outputting initial user state data. This step configures the user agent's initial participation probability, conversation length range, and initial emotional and viewpoint states, forming the initial state parameters when the user enters the simulation. Finally, it integrates user hierarchical segmentation data, user profile data, user long-term memory data, user short-term memory data, and user initial state data to obtain hierarchical user agent initial data. This data contains user agent instances at different influence levels. Each agent instance is configured with a user profile, long-term memory, short-term memory, and initial state parameters, providing a complete initial agent configuration for subsequent perception, reflection, and action.

[0145] Step 104: Using a preset activity intensity function, session scheduling is generated from event-driven input structure data suitable for continuous-time simulation, current public opinion environment data, and initial data of hierarchical user agents to obtain dynamic session scheduling data.

[0146] The preset activity intensity function refers to a preset rule function used to calculate the triggering timing and participation probability of user agent sessions. To adapt to the time-series progression logic of continuous-time simulation, it is essentially a composite non-homogeneous intensity function combining event timing, public opinion intensity, and user attributes. The function input includes three types of variables: the first type is the event timing information in the event-driven input structure data, i.e., the key node timestamps and stage attributes of the event development timeline; the second type is the discussion intensity indicators in the current public opinion environment data, i.e., the real-time public opinion status such as the total number of comments, interaction volume, and average sentiment in the aggregated statistical data; the third type is the user attributes in the initial data of the hierarchical user agent. The parameters include the user's level, basic participation probability, behavioral activity level, and text style configuration. The function consists of three parts: an event timing weight term, a public opinion heat response term, and a user attribute correction term. The event timing weight term adjusts the basic intensity according to the time progression of the event node, the public opinion heat response term amplifies or reduces the session trigger intensity according to the real-time public opinion heat, and the user attribute correction term adjusts the individual intensity benchmark according to the user level and behavioral parameters. Finally, it outputs the session trigger intensity of each user agent at different simulation times, thereby generating a session scheduling basis adapted to the simulation process and providing quantitative support for the timing triggering and participation decisions of subsequent user sessions.

[0147] Dynamic session scheduling data refers to structured data obtained after session scheduling generation and processing. It includes user session sequences arranged in chronological order, as well as corresponding triggering times and participation probability information, providing a scheduling basis for subsequent perception and reflection processing of user agents.

[0148] It should be noted that, based on the event-driven input structure data suitable for continuous-time simulation, the current public opinion environment data, and the initial data of hierarchical user agents obtained above, the preset activity intensity function is called to calculate the session triggering timing and participation probability of different user agents, and user session sequences are generated in chronological order to complete session scheduling and obtain dynamic session scheduling data.

[0149] Furthermore, step 104 may include the following sub-steps:

[0150] S41. Using a preset activity intensity function, based on the current public opinion environment data and the simulation control parameters in the event-driven input structure data suitable for continuous-time simulation, the initial data of the hierarchical user agent is sampled for the first session start time, and the user's first session time data is output.

[0151] S42. Assign a unique identifier to the session in the user's first session time data and output the session unique identifier data;

[0152] S43. Associate and bind user intelligent agents with the session unique identifier data, and output the session user association data;

[0153] S44. Record the session duration for the associated user data and output the basic session information data.

[0154] S45. Build a dynamic priority queue for the basic session information data and output the initial priority queue data.

[0155] S46. Sort the initial priority queue data according to the session start timestamp and output the time-series session queue data;

[0156] S47. Integrate the time-series session queue data and session user association data to obtain dynamic session scheduling data.

[0157] The user's first session time data refers to the set of first session start times for each user agent obtained by sampling through a preset activity intensity function, providing an initial time basis for user participation in the simulation.

[0158] Session unique identifier data refers to the independent set of identifiers assigned to each session, used for session tracking, association and management, to ensure that each session can be uniquely identified during the scheduling process.

[0159] Session user association data refers to the data set that establishes the binding relationship between a session and the corresponding user agent, clarifies the execution subject of each session, and provides the subject basis for driving the subsequent actions of the user agent.

[0160] Session duration refers to the duration Δt of a single user's participation in a single session, providing time boundary constraints for the execution of continuous actions such as entering, browsing, and commenting within a user session.

[0161] Basic session information data refers to a structured data set containing a unique session identifier, associated user agent, session start time, and session duration, providing basic information support for the construction of dynamic priority queues.

[0162] The initial priority queue data refers to the initial structure of the priority queue of pending sessions built based on the basic session information data, which is used to store all user sessions to be scheduled.

[0163] The time-series session queue data refers to the priority queue data sorted by session start timestamp, which maintains all sessions to be executed in chronological order and supports the session scheduling logic for asynchronous simulation.

[0164] It should be noted that this invention injects authoritative information into the public opinion environment at preset times through an official publisher module. This module is not an ordinary user node, but a dedicated information source directly linked to the timeline of event development. Its basic operation is as follows: when the system time reaches the preset official release timestamp, the scheduling control layer triggers the official publisher module to write the corresponding official text into the environment. The writing method can be to publish a new main post supplementary explanation, generate authoritative comments under the main post, write clarification replies to existing comment chains, or a combination of multiple methods.

[0165] The key role of the official publisher module is to simulate the process of "authoritative information structurally entering the discussion space" during sudden public events. Since official announcements generally correspond to important event nodes, their entry into the environment not only changes the set of information visible to subsequent users, but also affects the activity intensity function of the environment, causing the user participation density to increase over a period of time, leading to further discussion, clarification, or renewed controversy.

[0166] In a preferred embodiment, the system allows for the designation of multiple official publishers, corresponding to different identities such as government agencies, involved entities, official media, and platform announcements. These official publishers can have different priorities to control the injection order within the same time window.

[0167] Furthermore, this invention replaces the synchronous round-robin approach with a time-aware scheduling module to achieve asynchronous simulation that more closely resembles real-world social media usage scenarios. This module defines a single user's participation process as a user session, with each session having a start time T and a duration Δt. Within a session, a user can sequentially perform multiple actions such as entering, browsing, viewing more comments, liking, commenting, replying, and exiting.

[0168] Specifically, the system first samples the first entry time for each user based on the activity intensity function; then, it maintains all pending sessions in chronological order using a dynamic priority queue. Whenever a session is popped from the priority queue, the system drives the user to complete a series of consecutive actions based on the current environment. After the session ends, the system can continue to sample the user's next entry time and rewrite it into the priority queue.

[0169] The activity intensity function is preferably correlated with the event lifecycle, showing a rapid increase during the event's outbreak phase, a deceleration around the peak, and a gradual decrease during the decay phase. When key event nodes occur or official content is released, the activity intensity function is locally boosted to simulate a phased reactivation. Thus, the system can more realistically represent the public opinion rhythm of "first outbreak, then decay, and then being boosted again by developments or official information."

[0170] Specifically, a preset activity intensity function is used to sample the first session start time of the hierarchical user agent initial data based on the current public opinion environment data and simulation control parameters in the event-driven input structure data suitable for continuous-time simulation. This results in the output of the user's first session start time data. In this step, the preset activity intensity function is related to the event lifecycle, rising rapidly during the event outbreak phase, decelerating around the peak, and gradually decreasing during the decay phase. Furthermore, when key event nodes occur or official content is released, the activity intensity function is locally boosted to simulate the phased reactivation of public opinion. The sampling obtains the first session start time for each user agent. A unique identifier is assigned to each session in the user's first session time data, outputting unique session identifier data. Each session is assigned an independent identifier for subsequent session tracking and management. Finally, user agent association data is generated based on the unique session identifier data, establishing the binding relationship between sessions and corresponding user agents, thus clarifying the executing entity of each session. The system records the duration of sessions associated with users and outputs basic session information. A single user's participation is defined as a user session, and the duration Δt of each session is recorded to provide a time boundary for subsequent user actions. A dynamic priority queue is constructed from the basic session information data, outputting initial priority queue data and establishing a priority queue storage structure for sessions to be executed. The initial priority queue data is sorted by session start timestamp, outputting a time-series session queue. The dynamic priority queue maintains all sessions to be executed in chronological order, replacing the synchronous round-robin approach and achieving asynchronous simulation that more closely resembles real social media usage scenarios. The time-series session queue data and user-associated session data are integrated to obtain dynamic session scheduling data. Subsequently, the system will use this data to drive the corresponding user agent to complete a series of continuous actions based on the current environment whenever a session pops from the priority queue. After the session ends, the system can continue to sample the user's next entry time and rewrite it into the priority queue.

[0171] It is worth mentioning that, in terms of the official release method, it can be written as supplementary information to a new main post, or as a comment, reply, pinned announcement, or system notification; as long as authoritative information enters the environment within a preset time and affects subsequent user perception, it falls within the scope of protection of this invention and should be considered as an alternative form.

[0172] Step 105: Using a preset perception-memory-reflection update model, the dynamic session scheduling data, current public opinion environment data, and initial data of hierarchical user agents are processed by perception and reflection to obtain updated user status data.

[0173] The pre-defined perception-memory-reflection update model is a closed-loop processing model that drives the dynamic update of user status based on the user profile sub-module, perception sub-module, memory sub-module, and reflection sub-module of the user intelligent agent. Its core is a progressive process of "perception triggering based on profile attributes - hierarchical memory retrieval - reflective impression generation and status update". It takes dynamic session scheduling data, current public opinion environment data, and hierarchical user intelligent agent initial data as input to realize a single session update of user status. During model operation, the user profile submodule provides basic user attributes and hierarchical identifiers, providing a basis for the perception submodule to determine the scope of visible information. After a user session is triggered, the perception submodule, based on the attributes of the user profile submodule, reads the corresponding set of visible information from the current public opinion environment data, including the content of the main event post, the currently ranked comments, related replies, and subsequent official releases, and transmits the original observation information to the memory and reflection submodules. The memory submodule contains three levels: long-term memory, short-term memory, and current impression memory. Long-term memory stores the user's stable preferences, past experiences, and general stance; short-term memory records the user's emotions, opinions, and confidence status in recent interactions; and current impression memory reserves space for immediate understanding of the current session. The reflection submodule combines the observation information from the perception submodule and the hierarchical memory data from the memory submodule to comprehensively infer changes in the user's opinions and emotions, generate the current impression of the current session, and write it into the current impression memory. At the same time, it updates the user's short-term memory and state parameters such as participation probability and emotional tendency, completing the closed loop of user state update.

[0174] Updated user state data refers to the user agent state data obtained after perceptual reflection processing, which includes updated long-term memory, short-term memory, current impression, user participation probability, emotional tendency and other state parameters, providing a basis for subsequent user behavior decisions.

[0175] It should be noted that user sessions are triggered based on dynamic session scheduling data. The corresponding hierarchical user agent is scheduled to read the visible information in the current public opinion environment data, call the preset perception-memory-reflection update model, and process the perceived information in combination with the user agent's memory data to generate an instant impression and update the user status. The perception-reflection processing is completed to obtain the updated user status data.

[0176] Furthermore, step 105 may include the following sub-steps:

[0177] S51. Match the current timestamp of the time-series session queue of dynamic session scheduling data and pop the session to be executed, and output the user session data to be executed.

[0178] S52. Schedule the user intelligent agent corresponding to the user session data to be executed, and read the visible information of the current public opinion environment data, and output the visible information data of the environment;

[0179] S53. Extract text semantics, sentiment tendency, and publishing subject attributes from the visible environmental information data, and output user perception feature data.

[0180] S54. Perform long-term memory retrieval on user perception feature data and hierarchical user agent initial data, and output user historical memory data.

[0181] S55. Using a preset perception-memory-reflection update model, real-time impressions are generated from user perception feature data and user historical memory data, and the user's real-time impression data is output.

[0182] S56. Using a preset perception-memory-reflection update model, the user's real-time impression data is corrected, and the updated user status data is output.

[0183] It should be noted that the present invention updates the internal state after the user comes into contact with external information through the reflection update module. The process can be described as follows: enter the environment → read visible information → form an immediate impression → trigger reflection → update state variables → generate behavior.

[0184] In a preferred embodiment, the reflection update module maintains at least three types of dynamic state variables: emotional intensity e, social confidence c, and emotional polarity s. Emotional intensity e reflects the user's emotional fluctuations during the current period; social confidence c reflects the user's subjective trust in the system, fairness, or event handling; and emotional polarity s reflects the direction and intensity of the user's currently expressed emotions.

[0185] When users receive new official information, event updates, or other user comments, the system triggers a reflection and update module. Based on current observations, existing short-term memory, and long-term preferences, it generates updated state variables. This module is not limited to simple numerical addition and subtraction but allows for a comprehensive judgment based on textual semantics regarding whether the current stance has been revised, whether emotions have been strengthened, and whether hesitation or a shift has occurred. This allows for a more realistic simulation of the phenomenon in reality where "different users develop different subsequent attitudes after seeing the same official announcement."

[0186] Specifically, the process involves: matching the current timestamp of the time-series session queue in the dynamic session scheduling data and popping the session to be executed, outputting the user session data to be executed. This step is based on asynchronous simulation scheduling logic, popping the matching session from the time-series session queue according to the current system timestamp, triggering the user agent's participation process; scheduling the user agent corresponding to the user session data to be executed, and reading the visible information of the current public opinion environment data, outputting the visible information data of the environment. In this step, after the user agent enters the environment, it reads the visible information in the initial public opinion environment according to its hierarchical attributes, including the content of the main post of the event, the currently ranked comments, replies related to the target comments, and the official release content injected after the user's most recent session; extracting text semantics, sentiment tendency, and publishing subject attributes from the visible information data of the environment, outputting user perception feature data. This step performs structured feature extraction on the read environmental information, providing a quantitative basis for subsequent impression generation and reflection updates; and retrieving the user perception feature data and the initial data of the hierarchical user agent from long-term memory, outputting the user's historical memory data. This step retrieves the stable preferences, past experiences, and general stance stored in the user agent's long-term memory, as well as the data from short-term memory. Recorded recent interaction data on changes in emotion and opinion serve as a historical reference for user cognition. A pre-defined perception-memory-reflection update model is used to generate immediate impressions from user perceptual feature data and historical memory data, outputting immediate impression data. This step, based on user perceptual features and historical memory, forms an immediate understanding of current environmental information. The pre-defined perception-memory-reflection update model then corrects the immediate impression data, outputting updated user state data. In this step, the reflection update module is triggered, comprehensively judging the immediate impression based on the user's current observations, existing short-term memory, and long-term preferences. This judgment goes beyond simple numerical addition and subtraction; it comprehensively assesses whether the user's current stance has been corrected, whether emotions have been strengthened, and whether hesitation or a shift has occurred based on textual semantics. Simultaneously, it updates three dynamic state variables: emotional intensity, social confidence, and emotional polarity. Emotional intensity reflects the user's emotional fluctuations in the current period; social confidence reflects the user's subjective trust in the system, fairness, or event handling; and emotional polarity reflects the direction and strength of the user's currently expressed emotions. This completes the closed loop of user state updates, realistically simulating the phenomenon of different users forming different subsequent attitudes when faced with the same information.

[0187] Among them, the user session data to be executed refers to the set of user sessions to be scheduled obtained after being popped out by matching the timestamps of the time-series session queue. It includes session identifiers, associated user agents and basic session information, providing a basis for triggering user participation in the process.

[0188] Visible information data in the environment refers to the set of visible information that a user agent reads from the current public opinion environment data based on its hierarchical attributes after entering the environment. This includes the main post content, the top-ranked comments, related replies, and subsequent official releases.

[0189] User-perceived feature data refers to structured feature data obtained by extracting and processing textual semantics, emotional tendencies, and attributes of the publishing subject from visible environmental information data, providing quantitative support for subsequent real-time impression generation and reflective updates.

[0190] User historical memory data refers to the set of user memory data retrieved from the initial data of hierarchical user agents. It includes stable preferences, past experiences, and general stances in long-term memory, as well as recent interactive emotions and changes in opinions in short-term memory, providing historical reference for user cognition.

[0191] Real-time user impression data refers to the user's immediate understanding of the current environment, generated based on user perception feature data and user historical memory data, providing a basis for subsequent state variable updates.

[0192] Emotion intensity refers to the degree of emotional fluctuation of the user's intelligent agent in the current period, and is one of the dynamic state variables maintained by the reflection and update module.

[0193] Social confidence refers to the degree of subjective trust that a user agent has in a system, fairness, or the handling of events, and is one of the dynamic state variables maintained by the reflection and update module.

[0194] Emotional polarity refers to the direction and intensity of the emotion currently expressed by the user's intelligent agent, and is one of the dynamic state variables maintained by the reflection and update module.

[0195] It is worth mentioning that, in terms of reflection and updating, explicit language model reflection prompts, rule-based state correction, vectorized state updates, or a combination of the above methods can be adopted; as long as the internal state is computably updated after the user comes into contact with external information, it is an alternative solution of the present invention.

[0196] Step 106: Using a preset public opinion stage evaluation function, simulate the updated user status data, current public opinion environment data, and event-driven input structure data suitable for continuous-time simulation to obtain the final simulation result data.

[0197] The preset public opinion stage evaluation function is a multi-index composite judgment function designed for continuous-time multi-agent public opinion simulation. It is used to dynamically divide the development stages of public opinion and control the simulation process. The core is a closed-loop processing logic of "multi-dimensional state input - stage feature matching - process control output". It takes updated user state data, current public opinion environment data and event-driven input structure data suitable for continuous-time simulation as three core inputs to realize the dynamic judgment of public opinion stages and the regulation of the simulation process.

[0198] The input metrics for this function are divided into three categories: The first category is updated user status data, corresponding to dynamic user status metrics, including the average emotional intensity of the user group, the mean social confidence, the distribution of emotional polarity, and the level of participation and activity; the second category is current public opinion environment data, corresponding to aggregated statistical metrics on the environment side, including the total number of comments, the total number of replies, the amount of interaction, the mean emotional sentiment, and the current stage status marker; the third category is event-driven input structure data suitable for continuous-time simulation, corresponding to event-side time series and control metrics, including key node timestamps of the event development timeline, the simulation time window length, and pruning threshold parameters.

[0199] The function internally constructs a phase judgment rule corresponding to the development cycle of public opinion, and realizes the phase division through multi-indicator threshold matching: the fermentation phase is judged as the phase in which the growth rate of interaction exceeds the preset threshold and the average sentiment rises rapidly; the outbreak phase is judged as the phase in which the interaction reaches its peak, user participation is at its highest, and the sentiment intensity is at a high level; the decay phase is judged as the phase in which the interaction continues to decline, user participation decreases, and the average sentiment gradually falls back; the calm phase is judged as the phase in which the interaction is below the pruning threshold, the probability of user participation approaches 0, and the sentiment intensity tends to be stable. At the same time, the phase judgment results are corrected by combining key nodes in the event development timeline (such as the completion of event handling and the node of authoritative information release). When a key node is triggered, the function is allowed to temporarily modify the decay phase to the re-discussion phase to simulate the secondary fluctuation of public opinion caused by the release of authoritative information.

[0200] The function outputs two types of process advancement criteria based on the stage determination results: first, stage marking results, which are used to mark the current stage of public opinion, providing a basis for subsequent adjustments to user participation probability, official release module triggering, and activity intensity function correction; second, simulation process control instructions, which output simulation termination instructions when the stage is determined to be a calm period and the simulation time window end conditions are met, and output instructions to continue session scheduling when the stage is in other stages, and output instructions to reduce session sampling frequency if the interaction volume is lower than the pruning threshold, thereby achieving matching and control between the simulation process and the actual development rhythm of public opinion.

[0201] It should be noted that, based on the updated user status data, the current public opinion environment data, and the event-driven input structure data suitable for continuous-time simulation, the preset public opinion stage evaluation function is called to dynamically evaluate the public opinion development stage during the simulation process, and advance the simulation process until the termination condition is met, so as to obtain the final simulation result data containing the full-cycle public opinion status, user interaction records, and stage evaluation results.

[0202] Furthermore, step 106 may include the following sub-steps:

[0203] S61. Call the current public opinion environment data as the simulation benchmark environment, make user behavior decisions on the updated user status data, and output user behavior data to be executed.

[0204] S62. Update the public opinion environment data of user pending actions and output the updated public opinion environment data.

[0205] S63. Extract basic data for phase assessment from the updated public opinion environment data;

[0206] S64. Use a preset public opinion stage assessment function to determine the public opinion stage based on the stage assessment data and output the public opinion stage determination result data.

[0207] S65. Based on the public opinion stage judgment result data and the pruning threshold in the event-driven input structure data applicable to continuous time simulation, adjust the scheduling parameters and output the pruned scheduling control data.

[0208] S66. Perform loop simulation based on the scheduling control data after pruning, and output complete simulation process data;

[0209] S67. Integrate and encapsulate the complete simulation process data according to the output format parameters in the event-driven input structure data suitable for continuous-time simulation to obtain the final simulation result data.

[0210] User pending action data refers to the set of user intelligent agent actions to be executed after user behavior decisions are made. It includes action type, target object and action-related parameters, providing a basis for updating the public opinion environment.

[0211] Updated public opinion environment data refers to the public opinion environment data updated based on user behavior data, including the updated comment list, reply list, interaction count, and comment sorting results, providing environmental input for subsequent stage evaluation and user perception.

[0212] The basic data for phase assessment refers to a multi-dimensional data set extracted from the updated public opinion environment data and used to determine the phase of public opinion. It includes changes in the amount of discussion within the time window, changes in the average sentiment, interaction speed, user participation activity, and information on the progress of events.

[0213] The public opinion stage determination result data refers to the current stage marker data of public opinion obtained after being determined by the preset public opinion stage evaluation function, including the determination results of the outbreak period, peak period, decay period or reactivation period, which provides a basis for adjusting scheduling parameters.

[0214] Post-pruning scheduling control data refers to the session scheduling parameters adjusted based on the public opinion stage judgment results and pruning thresholds. It includes session sampling density, user participation intervals, and log writing rules for different stages, and is used to optimize simulation efficiency.

[0215] Complete simulation process data refers to the complete set of time-series data recorded during the simulation execution process, including complete process information on user behavior, environmental status, and stage determination, providing raw data for result integration.

[0216] It should be noted that after the state variables are updated, this invention outputs specific actions based on the current state and environmental conditions through the action submodule. If the output is a comment, the system generates a new comment object and writes it to the comment list; if the output is a reply, the system generates a reply object under the target comment; if the output is a like or a share, the system updates the interaction count of the corresponding object. After each action is completed, the environmental statistics object is updated synchronously, thereby changing the popular comments, latest replies, and stage statistics results visible to subsequent users.

[0217] In a preferred embodiment, the system can also set a comment sorting strategy, such as prioritizing display based on overall popularity, publication time, or authoritative content. In this way, preceding user actions not only generate information content but also further influence the cognitive input of subsequent users by changing the order in which the environment is displayed.

[0218] Furthermore, to improve simulation efficiency over longer time windows or with a larger user base, this invention includes a public opinion phase assessment and pruning module. This module identifies the current period as either the outbreak phase, peak phase, decline phase, or reactivation phase based on changes in discussion volume, average sentiment, interaction speed, and whether new developments have occurred within the time window.

[0219] When the system determines that it has entered a low-activity decay phase, the pruning module performs at least one of the following operations: reduces the sampling density of new sessions, extends the interval between user re-entry, reduces the number of repeated visits by low-influence users, merges consecutive low-value log writes, or retains only key user participation. Conversely, when the system detects the arrival of a new progress node or an official release node, the pruning module reduces the constraint strength and restores a higher participation density.

[0220] Through the phase evaluation and pruning modules, this invention can reduce computational redundancy during low-activity periods without destroying the key evolutionary structure, thereby balancing simulation accuracy and operational efficiency.

[0221] Furthermore, this invention outputs multi-level results of the public opinion evolution process through a result output module. The output content includes at least:

[0222] (1) Item-by-item behavior log: Records user ID, timestamp, action type, target object, and action text;

[0223] (2) Environment status log: Records the number of comments, replies, interactions, average sentiment, and stage determination within each time window;

[0224] (3) User status log: Records the emotional intensity, social confidence and emotional polarity of each user after each interaction;

[0225] (4) Aggregate time series: Output the overall public opinion evolution trajectory by hour or other preset granularity for subsequent analysis, comparison or contingency plan simulation.

[0226] Specifically, the system uses the current public opinion environment data as the simulation baseline environment, makes user behavior decisions based on the updated user status data, and outputs user action data to be executed. In this step, the user agent's action submodule outputs specific actions based on the current state and environmental conditions. Optional actions include posting a comment, replying to a comment, liking, forwarding, viewing more comments, and exiting. Different users generate different action sequences based on their hierarchical attributes: high-influence users have a higher probability of original comments, medium-influence users have a higher probability of commenting and replying, and ordinary users have a higher probability of browsing and liking. The system then updates the public opinion environment based on the user action data to be executed, outputting the updated public opinion environment data. This step performs corresponding environment update operations based on the type of user action to be executed. If the action is a comment, a new comment object is generated and written to the comment list; if it is a reply, a reply object is generated under the target comment; if it is a like or forward, the interaction count of the corresponding object is updated. Simultaneously, the system can set comment sorting strategies, such as prioritizing display based on overall popularity, publication time, or authoritative content. The updated environment statistics are updated synchronously, changing the popular comments and most viewed comments visible to subsequent users. New responses and phase statistics; extracting basic phase assessment data from the updated public opinion environment data, including changes in discussion volume, average sentiment, interaction speed, user participation activity, and whether new development nodes have appeared within the time window, providing input basis for public opinion phase determination; using a preset public opinion phase assessment function to determine the public opinion phase based on the basic phase assessment data, outputting the public opinion phase determination result data. This step identifies the current public opinion in the outbreak, peak, decay, or reactivation phase, providing a basis for subsequent scheduling parameter adjustments; adjusting scheduling parameters based on the public opinion phase determination result data and the pruning threshold in the event-driven input structure data suitable for continuous time simulation, outputting pruned scheduling control data. When the low-activity decay phase is determined, the pruning module performs operations such as reducing the sampling density of new sessions, extending the interval between user re-entry, reducing the number of repeated browsing by low-influence users, merging continuous low-value log writing, or retaining only key user participation; when a new development node or official release node is detected, the constraint strength is reduced, and a higher participation density is restored, reducing computational redundancy without destroying the key evolution structure.Based on the pruned scheduling control data, a closed-loop simulation process is initiated, continuously performing operations such as matching and popping time-series conversation queues, processing user agent perception and reflection, generating user behavior decisions, dynamically updating the real-time public opinion environment, and injecting periodically released official information. After each round of the process, a preset public opinion stage evaluation function is called to determine the public opinion stage. Based on the determination result, relevant scheduling parameters are dynamically adjusted to continuously iterate and advance the overall simulation process. The entire loop process runs uninterrupted until the preset simulation duration target is met, and the public opinion enters a stable and calm state, among other predetermined simulation termination conditions. After the simulation process terminates, the entire simulation output is summarized and retained. The simulation process generates complete simulation data by processing all time-series information, including user interaction behavior, public opinion environment status, user status, and public opinion stage determination. Then, it retrieves pre-defined output format parameters from the event-driven input structure data suitable for continuous-time simulation. Following strict adherence to the parameter definition rules, classification standards, and storage formats, the complete simulation data is uniformly sorted, classified, collected, and standardized for integration and encapsulation. Finally, it yields structurally sound and comprehensive simulation result data, which can be presented in a predetermined format as individual behavior logs, environment status logs, user status logs, and aggregated time-series data.

[0227] Among them, the individual behavior log refers to log data that records the behavior information of the user's intelligent agent, including user identifier, timestamp, action type, target object and action text, which is used to track user interaction behavior.

[0228] The environmental status log refers to the log data that records the status of the public opinion environment within each time window, including the number of comments, replies, interactions, average sentiment, and stage judgment results, which is used to reflect the dynamic changes in the environment.

[0229] User state logs refer to log data that records the state parameters of each user agent after each interaction, including emotional intensity, social confidence, and emotional polarity, and are used to track the evolution of user state.

[0230] Aggregated time series refers to the overall public opinion evolution trajectory data integrated according to a preset time granularity, which is used for subsequent analysis, comparison or contingency planning.

[0231] It is worth mentioning that, in terms of pruning, methods such as reducing session sampling density, reducing the participation of ordinary users, shortening low-value browsing paths, merging logs, or suspending the calculation of non-critical indicators can be adopted. As long as the goal is to dynamically reduce redundant calculations based on the stage of public opinion, these can all be used as alternative solutions.

[0232] For comparison of technical effects, existing technologies can be used as a reference, which can be roughly divided into four categories:

[0233] (1) Public opinion results display solutions: These solutions focus on the collection, classification, statistics and graphical representation of public opinion, such as counting the number of public opinion events by sentiment category and time period and performing visual coding. These solutions can help users observe the state of public opinion, but do not involve multi-agent dynamic interaction, official release injection and public opinion evolution simulation.

[0234] (2) Hot Topic Simulation Solution: Centered on a centralized topic page, this solution simulates the short-term life cycle of hot topics through time-aware interaction, comment / reply structure, and user perception-memory-action chain. This type of solution can well describe asynchronous participation in hot topic pages, but it is usually designed for topic or attack scenarios and lacks an official information intervention mechanism for sudden public events.

[0235] (3) Rumor propagation and large-scale social network simulation solutions: These solutions improve the efficiency and scalability of large-scale propagation simulation by using a hybrid approach of large language model agents and agent-based rule agents, while also introducing dynamic grouping, hierarchical networks, or intervention strategy modules. These solutions focus more on the propagation of rumors, misinformation, and complex networks, and are not specifically designed around the event timeline and authoritative release structure of public emergencies.

[0236] (4) Social Movement and Cognitive Subject Simulation Scheme: By modeling core users and ordinary users in a hierarchical manner, the evolution of opinions and interactive behaviors are simulated in a timeline environment similar to a social platform. This type of scheme is inspiring in terms of role heterogeneity and behavior generation, but its application background is mostly social movements or cognitive confrontations, and it has not formed a complete public opinion evolution and operation system centered on sudden public events and revolving around official releases and phased reactivation.

[0237] Furthermore, the above method has the following problems:

[0238] 1) Existing simulation schemes for hot topics or rumors lack an input structure for the timeline of event development that is compatible with public emergencies, making it difficult to simulate the chain process of "event progress - official statement - public re-reaction".

[0239] 2) Existing multi-agent systems mostly process official release nodes at the level of ordinary information sources, lacking a structured mechanism that can inject authoritative information at a preset time and reshape the rhythm of the environment.

[0240] 3) Existing solutions often use synchronous rounds or short lifecycle designs in terms of time progression, which do not provide sufficient support for asynchronous entry, continuous sessions and multi-stage discussions in real platforms.

[0241] 4) Existing solutions are prone to redundant calculations and low-activity phases during long-term simulations, resulting in low efficiency and making them unsuitable for deployment under conditions with more events and larger scale.

[0242] Therefore, the technical problem to be solved by this invention can be divided into five parts:

[0243] 1) Existing rule-based public opinion simulation methods mostly rely on fixed state transition and numerical update rules, which makes it difficult to express natural language interaction, emotional fluctuations and cognitive correction processes in the context of social media, resulting in an unrealistic portrayal of the evolution of public opinion in sudden public events.

[0244] 2) Existing multi-agent social simulation research focuses on risky scenarios such as rumor spread, hotspot manipulation, and information poisoning, and lacks systematic modeling of the rhythm reshaping effect after official information intervention and authoritative releases enter the discussion space in public emergencies.

[0245] 3) Existing simulation systems generally adopt synchronous round-robin progression, which makes it difficult to reflect the operational characteristics of users in real social platforms, such as asynchronous entry, continuous browsing, phased re-participation, and event reactivation at different times.

[0246] 4) Existing solutions often only focus on the dissemination of information itself, and do not adequately model the cognitive reflection and status update of users after they come into contact with official releases, other people's comments and the progress of events, making it difficult to reflect the behavioral consequences of user differentiation under the exposure of similar information.

[0247] 5) Under long-term windows or large user scales, existing models are prone to computational redundancy, resource waste during low-activity phases, and decreased simulation efficiency. There is a lack of effective technical mechanisms to dynamically adjust simulation density according to the public opinion phase.

[0248] To address the aforementioned issues, this invention provides a multi-agent simulation method for public opinion evolution in the context of public emergencies. By constructing a centralized public opinion environment, an official publisher injection mechanism, a time-aware scheduling mechanism, a user agent perception-memory-reflection-action closed loop, and a public opinion stage evaluation and pruning mechanism, it achieves unified simulation of information dissemination, emotional changes, stage reactivation, and the role of authoritative intervention in public emergencies.

[0249] Specifically, such as Figure 2As shown, this invention includes at least an event input module, a public opinion environment construction module, an official publisher module, a time-aware scheduling module, a user intelligent agent module, a reflection and update module, a public opinion stage evaluation and pruning module, and a result output module. The architecture, from left to right, includes: an event input layer, an environment construction layer, a scheduling and control layer, an intelligent agent interaction layer, and a result output layer. The event input layer is used to input the main event post, the event development timeline, the official release plan, user group configuration, and simulation parameters; the environment construction layer is used to initialize the main post, comment section, reply section, and statistical status; the scheduling and control layer is used to schedule user sessions and official release nodes according to timestamps; the intelligent agent interaction layer is used to execute user perception, memory, reflection, and behavior generation; and the result output layer is used to output comment sequences, user status sequences, time aggregation indicators, and log files for subsequent analysis.

[0250] During system operation, the event input layer first writes basic event information to the environment construction layer; the scheduling control layer then schedules users and official publishers to enter the environment according to time-aware rules; after users generate comments, replies, likes, and reposts in the environment, the environment state is continuously updated and has a reverse effect on subsequent users entering the system; the stage evaluation and pruning module dynamically adjusts the scheduling density based on the intensity of discussion and changes in sentiment. This forms a closed-loop operation chain of "input—scheduling—interaction—feedback—output". The overall operation flow of this method framework is as follows.

[0251] Step S1: Input the main event post, event timeline, official release configuration, user group configuration, and simulation parameters;

[0252] Step S2: Construct a centralized public opinion environment centered around a single event, and initialize the main post, comment section, reply section, and statistical objects;

[0253] Step S3: Initialize the user agent set and generate high-influence users, medium-influence users, and ordinary users according to the user stratification configuration;

[0254] Step S4: Sample the user session start time based on the activity intensity function and priority queue mechanism, and establish a queue of sessions to be executed;

[0255] Step S5: When the system time reaches a certain user session timestamp, drive the user to enter the environment and read the currently visible information;

[0256] Step S6: The user agent performs perception, memory retrieval, impression formation, and reflection update to obtain the updated internal state variables;

[0257] Step S7: Output the behavior based on the updated state variables, and write the comments, replies, likes or reposts back to the public opinion environment;

[0258] Step S8: When the system time reaches the official release timestamp, the official publisher module is triggered to write authoritative information to the environment;

[0259] Step S9: Perform phase evaluation according to the preset window, and prune the session sampling density and non-critical operation paths if necessary;

[0260] Step S10: Continue iterating until the simulation ends, and output the log file and aggregated time series results.

[0261] In summary, the key points of this invention are:

[0262] (1) A technical solution to build a centralized public opinion environment centered on public emergencies and maintain the discussion space together with the main post of the event, the comment list, the reply structure and the official release record.

[0263] (2) A technical solution that uses the event timeline and official release configuration as input structures and injects authoritative information into the environment at preset timestamps.

[0264] (3) A time-aware scheduling mechanism is adopted to achieve asynchronous participation, continuous browsing and phased reactivation through user sessions and priority queues.

[0265] (4) Dynamically assess the intensity of the discussion, emotional changes and event stages, and implement pruning and participation density adjustment techniques accordingly.

[0266] Compared with traditional agent-based models, this invention not only describes the user's numerical state, but also incorporates user profiles, short-term memory, immediate impressions, page observation, and comment reply chains into the same technical solution. Therefore, it can more completely reproduce the interaction logic and information influence path in social platforms.

[0267] Compared with schemes that rely solely on large language models to generate content independently for each subject, this invention improves engineering controllability and overall operational efficiency through asynchronous scheduling, page-based environment modeling, publisher configuration, and structured state output, making it more suitable for continuous, multi-scenario simulations of sudden public events.

[0268] Compared to general simulations of trending topics or rumor propagation, this invention is better suited for scenarios involving sudden public events. Firstly, it reflects asynchronous participation and continuous conversations on a real platform through a time-aware scheduling mechanism. Secondly, through a closed loop of perception-memory-reflection-action, it enables users to make cognitive corrections after encountering official releases, event updates, and comments from others, rather than simply executing fixed rule updates. Thirdly, it reduces redundant computation during low-activity periods through stage evaluation and pruning mechanisms, thus balancing the expressive power of public opinion structure with operational efficiency. Therefore, this invention is more suitable for applications such as emergency public opinion analysis, contingency plan simulations, policy information release rhythm design, and crisis communication training.

[0269] In this embodiment of the invention, a multi-agent public opinion evolution simulation method for public emergencies is provided. This method constructs a centralized public opinion interaction environment around the public emergency, receiving the main event post, the event development timeline, official release node configurations, and user group configurations. It organizes asynchronous user sessions based on a time-aware scheduling mechanism. User agents sequentially execute information perception, memory retrieval, impression formation, reflection and updating, and behavioral decision-making. The environment state is continuously updated through actions such as commenting, replying, liking, and forwarding. The system further sets up an official publisher injection mechanism to write authoritative information into the public opinion environment at a preset timestamp; and a public opinion stage evaluation and pruning mechanism to dynamically adjust participation density and computational resource allocation based on discussion intensity, emotional changes, and event stages. This solution can uniformly model information diffusion, emotional fluctuations, staged reactivation, and the role of official intervention in public emergencies, providing interpretable and reproducible technical support for emergency public opinion analysis, contingency plan simulation, and intervention strategy design.

[0270] Please see Figure 3 , Figure 3 This is a structural block diagram of a multi-agent public opinion evolution simulation system for sudden public events, provided in Embodiment 2 of the present invention.

[0271] This invention provides a multi-agent public opinion evolution simulation system for sudden public events, comprising:

[0272] The acquisition module 301 is used to acquire raw data of public emergencies and preprocess the raw data of public emergencies to obtain event-driven input structure data suitable for continuous-time simulation.

[0273] Module 302 is used to construct the public opinion environment from event-driven input structure data suitable for continuous-time simulation, and obtain the current public opinion environment data.

[0274] The initialization module 303 is used to initialize the user agent in the user group configuration data in the event-driven input structure data suitable for continuous-time simulation, so as to obtain the initial data of the hierarchical user agent.

[0275] The scheduling module 304 is used to generate dynamic session scheduling data by using a preset activity intensity function to generate session scheduling data for event-driven input structure data, current public opinion environment data and initial data of hierarchical user intelligent agents suitable for continuous time simulation.

[0276] The perception and reflection module 305 is used to perform perception and reflection processing on dynamic session scheduling data, current public opinion environment data and initial data of hierarchical user intelligent agents using a preset perception-memory-reflection update model to obtain updated user status data.

[0277] The simulation module 306 is used to simulate the updated user status data, the current public opinion environment data, and the event-driven input structure data suitable for continuous-time simulation using a preset public opinion stage evaluation function, and to obtain the final simulation result data.

[0278] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0279] This invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program; when the computer program is executed by the processor, the processor performs the steps of the multi-agent public opinion evolution simulation method for sudden public events as described in the above embodiments.

[0280] This invention also provides a computer-readable storage medium storing a computer program / instruction thereon, which, when executed by a processor, implements the steps of the multi-agent public opinion evolution simulation method for sudden public events as described in the above embodiments.

[0281] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0282] The units described as separate components may or may not be physically separate. 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0283] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0284] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0285] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-agent simulation method for public opinion evolution in the face of public emergencies, characterized in that, include: Acquire raw data of public emergencies and preprocess the raw data to obtain event-driven input structure data suitable for continuous-time simulation; The event-driven input structure data suitable for continuous-time simulation is used to construct a public opinion environment to obtain the current public opinion environment data; User agent initialization is performed on the user group configuration data in the event-driven input structure data suitable for continuous-time simulation to obtain hierarchical user agent initial data. Using a preset activity intensity function, session scheduling is performed on the event-driven input structure data suitable for continuous-time simulation, the current public opinion environment data, and the initial data of the hierarchical user agent to generate dynamic session scheduling data. A preset perception-memory-reflection update model is used to perform perception and reflection processing on the dynamic session scheduling data, the current public opinion environment data, and the initial data of the hierarchical user intelligent agent to obtain updated user status data. A preset public opinion stage evaluation function is used to simulate the updated user status data, the current public opinion environment data, and the event-driven input structure data suitable for continuous-time simulation, so as to obtain the final simulation result data.

2. The multi-agent public opinion evolution simulation method for sudden public events according to claim 1, characterized in that, The preprocessing of the raw data of the sudden public event yields event-driven input structure data suitable for continuous-time simulation, including: The raw data of the public emergency is format-checked and invalid information is removed, and the preliminarily cleaned input data is output. The input data after initial cleaning is processed to standardize the fields, and the standardized input data is output. The input data after the fields are standardized is then normalized to obtain event-driven input structure data suitable for continuous-time simulation.

3. The multi-agent public opinion evolution simulation method for sudden public events according to claim 1, characterized in that, The process of constructing a public opinion environment from the event-driven input structure data suitable for continuous-time simulation to obtain current public opinion environment data includes: Extract event main post data from the event-driven input structure data suitable for continuous-time simulation, initialize the main post object of the event main post data, and output the initialized main post data; The initial main post data is used to initialize the comment list, and the initial comment list data is output. The initial comment list data is used to initialize the reply list, and the initial reply list data is output. Establish a parent-child relationship between the initial comment list data and the initial reply list data, and output the comment reply association data; The official posting records for the comment reply association data are initialized, and the initialized official posting record data is output. The initial values ​​for aggregation and statistics are set for the official release record data, and the initial aggregated statistical data is output. The initialization aggregate statistics are used to initialize the environment log framework, and the initialization environment log data is output. By integrating the initial main post data, the comment reply association data, the initial official release record data, the initial aggregated statistics data, and the initial environment log data, the current public opinion environment data is obtained.

4. The multi-agent public opinion evolution simulation method for sudden public events according to claim 1, characterized in that, The initialization of user agents in the user group configuration data of the event-driven input structure data suitable for continuous-time simulation, to obtain hierarchical user agent initial data, includes: Extract basic user configuration parameters from the user group configuration data; The user basic configuration parameters are used to segment user groups into different strata, and user segmentation data is output. User profiles are generated for the users in the user hierarchical data, and the user profile data is output. The user profile data is initialized for long-term memory, and the user long-term memory data is output. The user's long-term memory data is initialized with short-term memory, and the user's short-term memory data is output. The user's short-term memory data is assigned a value, and the user's initial state data is output. By integrating the user hierarchical segmentation data, the user profile data, the user long-term memory data, the user short-term memory data, and the user initial state data, the initial data of the hierarchical user agent is obtained.

5. The multi-agent public opinion evolution simulation method for sudden public events according to claim 1, characterized in that, The process employs a preset activity intensity function to generate dynamic session scheduling data by processing the event-driven input structure data suitable for continuous-time simulation, the current public opinion environment data, and the initial data of the hierarchical user agent. This dynamic session scheduling data includes: The preset activity intensity function is used to sample the first session start time of the hierarchical user agent's initial data based on the current public opinion environment data and the simulation control parameters in the event-driven input structure data suitable for continuous time simulation, and the user's first session time data is output. Assign a unique identifier to each session in the user's first session time data and output the session unique identifier data; The session unique identifier data is used to associate and bind user intelligent agents, and the session user association data is output. The session duration is recorded for the associated user data, and basic session information data is output. A dynamic priority queue is constructed based on the basic session information data, and the initial priority queue data is output. The initial priority queue data is sorted according to the session start timestamp, and the time-series session queue data is output. By integrating the time-series session queue data and the session user association data, dynamic session scheduling data is obtained.

6. The multi-agent public opinion evolution simulation method for sudden public events according to claim 1, characterized in that, The method employs a preset perception-memory-reflection update model to perform perception and reflection processing on the dynamic session scheduling data, the current public opinion environment data, and the initial data of the hierarchical user agent, resulting in updated user state data, including: The current timestamp is matched with the current session in the time-series session queue of the dynamic session scheduling data, and the user session data to be executed is output. Schedule the user intelligent agent corresponding to the user session data to be executed, and read the visible information of the current public opinion environment data, and output the visible information data of the environment; The environmental visible information data is processed by extracting text semantics, sentiment, and publishing subject attributes to output user perception feature data. Long-term memory retrieval is performed on the user perception feature data and the initial data of the hierarchical user agent to output the user's historical memory data; Using the preset perception-memory-reflection update model, real-time impressions are generated from the user's perceptual feature data and the user's historical memory data, and the user's real-time impression data is output. The user's real-time impression data is corrected using the preset perception-memory-reflection update model, and the updated user status data is output.

7. The multi-agent public opinion evolution simulation method for sudden public events according to claim 1, characterized in that, The method employs a preset public opinion stage evaluation function to simulate the updated user status data, the current public opinion environment data, and the event-driven input structure data suitable for continuous-time simulation, obtaining the final simulation result data, including: The current public opinion environment data is used as the simulation benchmark environment to make user behavior decisions based on the updated user status data and output user behavior data to be executed. Update the public opinion environment data of the user's pending actions and output the updated public opinion environment data. Extract basic data for phase assessment from the updated public opinion environment data; Using the preset public opinion stage evaluation function, the public opinion stage is determined based on the stage evaluation basic data, and the public opinion stage determination result data is output. Based on the public opinion stage determination result data and the pruning threshold in the event-driven input structure data applicable to continuous time simulation, the scheduling parameters are adjusted, and the pruned scheduling control data is output. Based on the post-pruning scheduling control data, perform a cyclic simulation and output complete simulation process data; The complete simulation process data is integrated and encapsulated according to the output format parameters in the event-driven input structure data suitable for continuous-time simulation to obtain the final simulation result data.

8. A multi-agent public opinion evolution simulation system for sudden public events, characterized in that, include: The acquisition module is used to acquire raw data of public emergencies and preprocess the raw data of public emergencies to obtain event-driven input structure data suitable for continuous-time simulation. The construction module is used to construct the public opinion environment from the event-driven input structure data suitable for continuous-time simulation, and obtain the current public opinion environment data. An initialization module is used to initialize the user agent in the user group configuration data in the event-driven input structure data suitable for continuous-time simulation, so as to obtain the hierarchical user agent initial data. The scheduling module is used to generate dynamic session scheduling data by using a preset activity intensity function to process the event-driven input structure data suitable for continuous-time simulation, the current public opinion environment data, and the initial data of the hierarchical user agent. The perception and reflection module is used to perform perception and reflection processing on the dynamic session scheduling data, the current public opinion environment data, and the initial data of the hierarchical user agent using a preset perception-memory-reflection update model to obtain updated user status data. The simulation module is used to simulate the updated user status data, the current public opinion environment data, and the event-driven input structure data suitable for continuous-time simulation using a preset public opinion stage evaluation function, so as to obtain the final simulation result data.

9. An electronic device, characterized in that, The system includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the multi-agent public opinion evolution simulation method for sudden public events as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the multi-agent public opinion evolution simulation method for sudden public events as described in any one of claims 1-7.

Images