A method and system for simulating large-scale social networks based on dynamic topology

Abstract

This invention relates to the field of computer software technology and discloses a method and system for simulating large-scale social networks based on dynamic topology. The method includes: assigning user profiles and memories to corresponding agents to construct a social network for the agents; distinguishing core agents from regular agents based on influence metrics; core agents using a large language model to interact with other core agents through natural language and influence the attitudes of regular agents towards events; regular agents updating their attitudes through trust weights and adjusting the social network using a dynamic link prediction engine; updating the agents' memories, tweet databases, and social networks; and continuously performing a dynamic evolution process at each time step until the event simulation is complete. This invention ensures both detailed characterization of individual behaviors and overall system scalability, significantly improving the decision-making efficiency and simulation accuracy of ultra-large-scale social network simulators.

Description

Technical Field

[0001] This invention relates to the field of computer software technology, and more specifically to a method and system for simulating large-scale social networks based on dynamic topology. Background Technology

[0002] Social network simulation is a research method that explores social phenomena such as information dissemination mechanisms, group attitude evolution, and information cocoons by simulating the interactive behaviors of individuals and groups. Utilizing mathematical models, computer simulation, and data analysis techniques, social network simulation transforms the complex individual interactions and group collaboration rules of real-world social platforms into virtual simulation models. This allows for the reproduction and analysis of the dynamic evolution of social networks in a controlled virtual environment, thereby predicting changes in social trends and revealing patterns in social behavior.

[0003] Social network simulators have significant application value in the following aspects: 1) Sociological research and revelation of public opinion formation mechanisms: Social network simulators can reveal the formation mechanism of public opinion and the laws of event dissemination by simulating and deducing the formation and evolution of various group emotions; 2) Event prediction and early warning: By simulating the evolution of social network events, simulators can predict the trend of public opinion or events, helping relevant personnel to provide early warnings and formulate response strategies; 3) Conducting cognitive intervention experiments: Different public opinion and cognitive intervention guidance strategies can be simulated and verified in a virtual environment, and their effects can be evaluated, thus providing a reference for public opinion guidance in reality.

[0004] Previous agent-based (ABM) social network simulations designed predefined behavioral rules for the simulated objects, enabling them to make decisions based on the environment and the behavior of other agents within the simulated environment. This approach focuses on macroscopic modeling and mechanistic analysis of group interaction patterns, neglecting the microscopic driving effect of individual heterogeneous behavior on the development of event situations, and thus exhibits poor adaptability in complex social environments. In recent years, large language model (LLM)-driven social network simulators have accurately reproduced event evolution patterns and group behavior phenomena by modeling user behavior decisions on real social platforms with fine granularity, opening up new directions for research on the operating mechanisms of social platforms. The scale of agents has become a core issue in research on large language model-driven social network simulators, driving the exponential growth of simulation scale. For example, existing technologies have successfully captured news dissemination patterns by leveraging the human-like capabilities of large language models in perception, reasoning, and behavior, achieving preliminary explorations in small-scale social network simulation. Existing technologies have also constructed environments containing thousands of agents, replicating individual behavioral decisions and the evolution of group attitudes by simulating the interactive behaviors of agents such as posting, forwarding, and commenting. To address the issues of low efficiency and high cost in large-scale user simulation, many studies have focused on improving simulation efficiency. These studies typically employ a hierarchical design involving core agents and regular agents, selectively activating a small subset of core agents at each time step to perform human-like interactions, thereby successfully reproducing larger-scale group behavior.

[0005] While existing technologies have made significant progress in social network simulation, blindly expanding the scale of agents is unwise. Most current research relies on static social networks, neglecting the dynamic evolution of social relationships, and modeling dynamic social networks faces significant challenges. Firstly, during event propagation, user behavior decisions are not merely interactive actions but also involve adaptive adjustments to their social relationships. The factors influencing the evolution of social relationships are multidimensional and complex. For example, users typically tend to connect with users who share similar views, reduce interaction with users who hold opposing views, and even disconnect. Furthermore, user influence, one-way attention, and unequal interaction lead to asymmetric social relationships, while content quality and timeliness determine the attractiveness and reach of information, thus influencing the formation and evolution of social relationships. The interactions between these factors greatly increase the complexity of modeling dynamic social networks. Secondly, to balance simulation efficiency and scale, existing large-scale simulators typically select a fixed set of core agents with human-like interactive capabilities. These core agents, as key nodes in event propagation, drive the evolution of events. However, as events unfold, user roles change. The inherent propagation characteristics of core agents, such as significant influence, make random or static selection strategies unsuitable. Therefore, effectively quantifying these key features and dynamically identifying core agents at different time steps becomes a crucial path to aligning with the propagation of real-world events. In conclusion, modeling the dynamics of social networks is a significant challenge that urgently needs to be addressed in current large-scale social network simulations. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method and system for simulating large-scale social networks based on dynamic topology structures. Specifically, this invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a method for simulating large-scale social networks based on dynamic topology, used to extrapolate attitudes towards events occurring within a target user group, including:

[0008] User profiles and memories are assigned to corresponding intelligent agents, and social networks are built for the intelligent agents based on the outgoing and incoming followers of users on social platforms;

[0009] Dynamic evolution process: The influence metric of each agent is calculated based on the number of second-order incoming followers and the variance of their attitudes towards the event. This metric distinguishes core agents from regular agents. Core agents are driven by a large language model and interact with other core agents through natural language, influencing the attitudes of regular agents towards the event. Core agents use user profiling, memory mechanisms, and information flow-based link prediction methods to decide on interaction behaviors with other tweets or agents and adjust their social networks. Regular agents are driven by an agent model and interact with other regular agents. Regular agents update their attitudes through trust weights and adjust their social networks using a dynamic link prediction engine. The agents' memories, tweet databases, and social networks are also updated.

[0010] The dynamic evolution process continues at each time step until the event deduction is completed and the user's attitude towards the event is obtained.

[0011] In one embodiment, the step of calculating the influence metric of each agent by using the number of second-order incoming followers corresponding to each agent and the variance of the attitudes of incoming followers towards the event specifically includes:

[0012] Influence measurement quantifies an agent's dissemination ability by the number of second-order incoming followers, and measures the content diversity of the agent's tweets by the variance of incoming followers' attitudes towards the event.

[0013] ;

[0014] in, This represents the measure of the influence of agent i at time t. This represents the attitude of agent i towards the event at time t. This indicates whether agent i is interested in agent j. This indicates that agent i does not pay attention to agent j. This indicates that agent i is interested in agent j; Represents the list of incoming followers of agent i; This indicates whether agent j is interested in agent k. This indicates that agent j does not pay attention to agent k. This indicates that agent j is interested in agent k.

[0015] In one embodiment, distinguishing between core agents and regular agents based on influence metrics specifically includes:

[0016] At each time step, the top K agents with the highest influence metrics are selected as core agents, and the other agents are selected as regular agents.

[0017] In one embodiment, the core agent uses user profiling, memory mechanisms, and information flow-based link prediction methods to decide on interaction behaviors with other tweets or agents and adjust the social network, specifically including:

[0018] Calculate the recommendation scores of core agent i and candidate tweet j. , It integrates the content representation of the latest tweets from the core intelligent agent i. Content representation of candidate tweets Content matching degree, lifecycle factor of candidate tweets, and content influence of candidate tweet j :

[0019] ;

[0020] in, Represents cosine similarity. It is a natural constant. Indicates the publication time of candidate tweet j. This represents the index at time t. This represents the decay rate of the life cycle factor.

[0021] The recommendation list of core agent i is formed based on the recommendation score: among the agents not followed by core agent i, agents that meet the following conditions are selected: the number of incoming followers exceeds a set threshold or the outgoing followers of core agent i; the tweets published by the selected agents are used as candidate tweets, and the N tweets with the highest recommendation scores are selected from the candidate tweets to form the recommendation list.

[0022] The core agent i makes decisions based on user profiles, memories, messages received from neighboring agents, and personalized information streams based on recommendation lists, and obtains interaction behaviors with other tweets or agents and adjusts its own attention relationships.

[0023] In one embodiment, the core agent i makes decisions based on user profiles, memories, received messages from neighboring agents, and personalized information streams based on recommendation lists to determine interactive behaviors with tweets or other agents, specifically including:

[0024]

[0025] For the decision-making process, User profile for core intelligent agent i This represents the memory of the core agent i at time t-1. The personalized information flow of the core intelligent agent i at time t. The message received by the core agent i from its neighboring agents at time t-1; This represents the set of actions performed at time t. The actions in the set include: uploading a tweet, retweeting a tweet, commenting on a tweet, following other agents, unfollowing other agents, liking tweets uploaded by others, and doing nothing.

[0026] In one embodiment, the conventional agent updates its attitude through trust weights, specifically including:

[0027] Based on trust quantification, the attitude interaction relationship between a regular agent and other agents is quantified. Agents with higher trust or more incoming followers are assigned higher weights.

[0028] ;

[0029] ;

[0030] in, This represents the attitude of a regular agent i towards an event at time t. Subject to neighboring intelligent agents The weight of influence Indicates trust in weight The proportion in This represents the degree of trust that agent i has in agent j. Let i be the set of neighboring agents of a regular agent i at time t. For agent j, there is a list of incoming followers. Re-weight attitudes.

[0031] In one embodiment, adjusting the social network using a dynamic link prediction engine specifically includes:

[0032] Predict scores using missing links between regular agents i and j. And fake link identification score Modeling the creation and discontinuation of social relationships. The highest-ranking agent j is the outgoing follower that the regular agent i newly focuses on. The highest-ranking agent j is the latest outgoing follower who unfollowed the regular agent i.

[0033] In one embodiment, the missing link prediction score And fake link identification score The calculation method is as follows:

[0034] ;

[0035] ;

[0036] in, This represents the attitude of a regular agent i towards an event at time t. This represents the attitude of agent j towards the event at time t. For agent j, there is a list of incoming followers. This represents the set of candidate agents that a regular agent i can follow in a social network. The elements of the candidate agent set include the outgoing followers of regular agent i and agents whose number of incoming followers exceeds a set value. This indicates whether agent j pays attention to the regular agent i. For a regular intelligent agent i, the list of followers is displayed. This represents the degree of trust that agent i has in agent j.

[0037] In one embodiment, the degree of trust that agent i has in agent j is... Established based on agent k, who acts as an intermediary in all trust propagation paths:

[0038] ;

[0039] ;

[0040] in, This represents the trust relationship established between agent i and agent j based on agent k in the trust propagation path.

[0041] In a second aspect, the present invention provides a computer system including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method of any embodiment of the first aspect.

[0042] Compared with the prior art, the beneficial technical effects of the present invention are:

[0043] 1. Significantly Improved Efficiency in Large-Scale Social Network Simulation: By employing a hierarchical design to dynamically distinguish between core agents with human-like decision-making capabilities and regular agents based on predetermined rules, this invention achieves a rational and optimized allocation of computing resources. In this way, core users with key characteristics can play a pivotal role in the evolution of public opinion, while a large proportion of regular users operate efficiently through a low-cost model. This strategy supports the dynamic interaction of ultra-large-scale agents, ensuring both detailed characterization of individual behavior and overall system scalability. It significantly improves the decision-making efficiency and simulation accuracy of ultra-large-scale social network simulators, realistically reflecting the complex and ever-changing process of public opinion dissemination in real society.

[0044] 2. Superior Dynamic Evolution and Simulation Capabilities Compared to Static Networks: Dynamic social topologies can effectively capture the dynamic evolution of public attitudes, demonstrating superior accuracy, stability, and adaptability across different events. Dynamic social networks can better reflect real-world communication, accelerate the polarization of attitudes, and show a clear clustering pattern in newly added attention relationships. Furthermore, dynamic social networks can predict the growth of incoming followers during event dissemination, providing empirical support for the cultivation of opinion leaders.

[0045] In summary, this invention provides a high-fidelity experimental platform for simulating the dynamics of event propagation and reproducing group behavior phenomena. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the method flow in an embodiment of the present invention.

[0047] Figure 2 This is a schematic diagram of the frame structure in an embodiment of the present invention. Detailed Implementation

[0048] A preferred embodiment of the present invention will now be described in detail with reference to the accompanying drawings.

[0049] like Figure 1 As shown, the present invention provides a large-scale social network simulation method based on dynamic topology, used to extrapolate events occurring within a target user group, comprising the following steps:

[0050] S1 assigns user profiles and memories to corresponding intelligent agents, and builds social networks for the intelligent agents based on the user's outgoing and incoming followers on the social platform.

[0051] S2, Dynamic Evolution Process: The influence metric of each agent is calculated based on the number of second-order incoming followers and the variance of their attitudes towards the event. This metric distinguishes core agents from regular agents. Core agents are driven by a large language model and interact with other core agents through natural language, influencing the attitudes of regular agents towards the event. Core agents use user profiling, memory, and information flow-based link prediction methods to decide on interaction behaviors with tweets or other agents and adjust their social networks. Regular agents are driven by an agent model and interact with other regular agents. Regular agents update their attitudes through trust weights and adjust their social networks using a dynamic link prediction engine. The agents' memories, tweet databases, and social networks are also updated.

[0052] S3, the dynamic evolution process continues at each time step until the iterative simulation time step reaches the set number, the event deduction is completed, and finally the user's attitude towards the event is obtained.

[0053] Outgoing followers are other users that the current user follows; incoming followers are other users who follow the current user, which are commonly referred to as "fans" on social media platforms.

[0054] To overcome the limitations of existing social simulation methods, this invention divides users into core agents based on large language models (LLM) and regular agents driven by agent models (ABMs), corresponding to opinion leaders and ordinary users in the real world, respectively.

[0055] This invention innovatively introduces a dynamic hierarchy, selecting core agents by quantifying the propagation potential and content diversity at each time step. This enables core agents to adaptively switch based on key features, thus balancing efficiency and accuracy in large-scale social simulations.

[0056] Furthermore, this invention designs different dynamic social relationship modeling strategies for different types of agents. The core agent employs an information flow-based link prediction method, which evaluates attitude similarity, content timeliness, and tweet influence to recommend potential like-minded non-neighbor agents. This allows the core agent to autonomously decide whether to follow or unfollow other agents, thereby simulating real-world homogeneous connection behavior and dynamic relationship evolution. The regular agent constructs a behavioral decision-making process oriented towards unequal interactions. By introducing the concept of trust to quantify unequal interactions between agents, and employing a dynamic link prediction engine to simulate relationship evolution driven by multiple factors, this reflects how the core agent and local neighbors influence the passive behavior of most agents in the real world.

[0057] See Figure 2 The technical solution of the present invention will be described in detail in several parts below.

[0058] 1. Task request and environment server initialization.

[0059] Upon receiving an event situation simulation task request, the environment server first initializes the user profiles and social network topology of the target user group. The main function of the environment server is to maintain the state and data of the social media platform, such as user information, historical tweets, and user relationships. The environment server consists of four main modules: the user module, the tweet module, the relationship module, and the recommendation list. The user module stores the user profile and memory of each user; the tweet module contains all tweets posted by all users and records detailed information such as the number of comments, likes, and creation time for each tweet; the relationship module stores the structure of the social network, including each user's outgoing and incoming follower lists; and the recommendation list module uses a link prediction method based on information flow to personalize the recommendation list for each user based on their memory and tweet information.

[0060] 2. Dynamic hierarchical module identifies core intelligent agents.

[0061] To ensure high accuracy and efficiency in large-scale user simulations, the dynamic layering module constructs a propagation potential and content richness evaluation system, identifies core agents that play a crucial role in event propagation, and adjusts the information interaction methods between agents based on agent type. The dynamic layering module comprises two parts: core agent selection and dynamic interaction. Regarding core agent selection, core agents typically possess higher propagation potential and content diversity; the former represents the breadth and depth of information dissemination, while the latter reflects the agent's knowledge level, depth of thought, and willingness to express opinions. To effectively identify core agents at different propagation stages, this invention designs an influence measurement method. And at each time step, select the one with the highest The top K agents are selected as core agents. Core agents with key characteristics are dynamically and adaptively switched at different time steps, allowing computational resources to focus on the most influential core users, thus balancing efficiency and accuracy in large-scale social simulations. Regarding dynamic agent interaction, the interaction method depends on the agent type. Core agents interact through natural language dialogue, while regular agents use Agent Models (ABMs) to transmit information. Furthermore, content generated by core agents is converted into attitude scores using a Large Language Model (LLM), which then influences the attitude updates of regular agents through the Agent Models (ABMs), while the attitudes of core agents remain unaffected by regular agents. The dynamic hierarchical module balances computational efficiency and simulation detail, accurately reproducing real-world propagation dynamics while achieving dynamic, hierarchical simulation of large-scale social networks.

[0062] 3. Core agent behavior decision-making.

[0063] The core agent driven by the Large Language Model (LLM) corresponds to opinion leaders in real-world social networks. The core agent is equipped with user profiling and a memory mechanism. The user profile includes name, gender, age, occupation, interest set, and personality traits, ensuring high personalization and behavioral diversity while maintaining the rationality of the user profile. The memory mechanism considers the agent's personal experience and environmental interaction memory. Personal experience memory represents the user's historical behavioral records, while environmental interaction memory captures the attitudes and behavioral responses of neighboring agents to specific events. The memory mechanism includes retrieval, updating, and reflection operations, enabling the core agent to make appropriate behavioral decisions based on the most relevant and urgent memories. Furthermore, to facilitate the core agent's reflection on the evolution of social relationships, this invention designs a link prediction method based on information flow. By quantifying attitude similarity, content timeliness, and tweet influence, it calculates the recommendation score between core agent i and candidate tweet j. This system recommends potential like-minded non-neighbor agents, enabling them to autonomously decide on interactive behaviors and adjust social relationships, realistically simulating homogeneous connection behavior and dynamic relationship evolution. The core agent's behavioral decisions reflect its attitude towards specific events. Specific behaviors include uploading tweets, retweeting tweets, commenting on tweets, following other agents, unfollowing other agents, liking tweets uploaded by others, and doing nothing. Each behavior is closely related to the user's social interaction and information dissemination.

[0064] 4. Conventional agent behavior decision-making.

[0065] To balance computational efficiency and scalability, conventional intelligent agents employ Agent Models (ABMs) to describe the interaction process between users, which can be uniformly expressed using selection functions, attitude update functions, and information transfer functions. However, traditional Agent Models (ABMs) cannot adapt to the problems of unequal interactions and dynamic social networks in real-world social networks. Therefore, this invention constructs a behavioral decision-making model for conventional intelligent agents oriented towards unequal social relationships. It introduces the concept of trust to quantify unequal attitude interactions between users and utilizes a dynamic link prediction engine to capture social relationship adjustment patterns driven by multi-dimensional factors such as attitude similarity and one-way attention. This better adapts to unequal interactions and social relationship adjustments in real-world social networks, simulating the behavioral patterns of most nodes in real-world social networks that are passively influenced by core nodes and local neighbors.

[0066] 5. Environment server updates and iterative cycles.

[0067] After each simulation round, the environment server updates the agent's memory, tweet library, and social network. The environment server provides a dynamic and scalable infrastructure for the entire simulation framework, ensuring data flow and state updates during the simulation process, which will influence the agent's decisions in the next round. Through continuous iteration at preset time steps, the social network simulator can perform large-scale event situational simulations and group behavior simulations.

[0068] In one embodiment, the present invention designs an influence metric for dynamic hierarchical modules. To distinguish between core agents and regular agents, and to select the agent with the highest [status] at each time step. The top K agents are selected as the core agents. Influence measurement. The propagation capability of an agent is quantified by the number of second-order incoming followers, and content diversity is measured by the variance of the attitudes of incoming followers. The calculation method can be expressed as follows:

[0069] ;

[0070] in, The score represents the attitude of agent i at time t. This represents the list of incoming followers of agent i. This indicates whether agent i is interested in agent j. This indicates that agent i does not pay attention to agent j. This indicates that agent i is interested in agent j.

[0071] In one embodiment, during the core agent behavior decision-making process, a link prediction method based on information flow calculates the recommendation score between agent i and candidate tweet j. This score integrates the content representation of the agent's latest tweets. Content representation of candidate tweets Content matching degree The lifecycle factors of candidate tweets and their content influence The recommendation score is given by the following formula:

[0072] ;

[0073] in, Indicates the publication time of candidate tweet j. Content influence is quantified by weighted numbers of likes, shares, comments, and followers, with a lifecycle factor following... The rate exhibits an exponential decay law. To ensure... Not zero, in terms of influence of the original content Add 1 to the base.

[0074] The recommendation list for core agent i is formed based on the recommendation score. The candidate tweets in the recommendation list are selected from tweets published by agents that have never been followed by core agent i. The publishers of the candidate tweets are agents whose number of inbound followers exceeds a set threshold (the threshold for the number of inbound followers in a scale of 100,000 agents can be set to 1,000) or outbound followers of core agent i's outbound followers (i.e., friends of friends). The recommendation list is composed of the N tweets with the highest recommendation scores published from the candidate tweets.

[0075] Based on user profiles Historical memory Messages received from neighboring intelligent agents and personalized information feed The decision-making process of agent i It can be modeled as:

[0076] ;

[0077] in, This represents the set of actions performed at time t. The core agent, through behavioral interactions, integrates new information and personalized information streams from its neighbors, thereby driving dynamic updates of attitudes and evolution of social relationships within the social network.

[0078] In one embodiment, the present invention introduces trust when a conventional intelligent agent makes behavioral decisions. The concept of agent attitude is used to quantify unequal attitudinal interactions between users, assigning higher weights to agents with high trust levels or a large number of inbound followers, thereby defining agent attitude. Set of neighboring intelligent agents Influence methods:

[0079] ;

[0080] .

[0081] in, This represents the attitude of a regular agent i towards an event at time t. Subject to neighboring intelligent agents The weight of influence This indicates the proportion of trust in the weighting. It is the weight of historical attitudes in the process of attitude updating.

[0082] In one embodiment, the present invention employs a dynamic link prediction engine to model how conventional agents adjust social relationships. The adjustment of social relationships is influenced by multiple dimensions, including user attitude similarity, one-way attention, user influence, and trust level. Based on these considerations, the present invention defines the following missing link prediction score. And fake link identification score Used to model the creation and discontinuation of social relationships. The highest-ranking agent j is a new follower of agent i. Similarly.

[0083] ;

[0084] .

[0085] in, This represents the list of outgoing followers of a regular intelligent agent i on a social network. Let represent the set of candidate agents that agent i can follow in the social network, including only friends of friends and users with a large number of incoming followers. Furthermore, when a user adds a new follower, agents i and j need to establish a trust relationship, which can be established through a middleman k. Combining all potential trust propagation paths, the trust relationships between agents are... This can be expressed as,

[0086] ;

[0087] .

[0088] The dynamic layering module of this invention designs an influence metric to quantify the propagation potential and content diversity of agents. It identifies core agents with key characteristics at each time step and constructs a dynamic layered design of core agents and regular agents. The dynamic layering module balances computational efficiency and simulation detail, accurately reproducing real-world propagation dynamics while achieving dynamic, hierarchical simulation of large-scale social networks.

[0089] This invention designs a link prediction method based on information flow for core intelligent agents. By quantifying attitude similarity, content timeliness, and tweet influence, it recommends potential like-minded non-neighbor intelligent agents, enabling core intelligent agents to autonomously make interactive decisions and adjust social relationships. It realistically simulates homogeneous connection behavior and dynamic relationship evolution, thus accurately aligning with the attitude dynamics and relationship evolution in the real world.

[0090] This invention constructs a behavioral decision-making module for unequal interactions in conventional intelligent agents. It introduces the concept of trust to quantify unequal attitude interactions between users and uses a dynamic link prediction engine to capture social relationship adjustment patterns driven by multi-dimensional factors such as attitude similarity and one-way attention, thereby better adapting to unequal interactions and social relationship adjustments in real social networks.

[0091] The technical solution of the present invention will be described in more detail below with reference to an embodiment.

[0092] This embodiment simulates the attitude evolution and behavioral decisions of 100,000 agents in a social network over 12 days regarding a specific event, including 2,000 core agents and the rest being regular agents.

[0093] Upon receiving an event situation simulation task request, the environment server first initializes the user profiles and social networks of the target user group. Basic user profile attributes are assigned to each agent, including name, age, gender, and occupation. Age is calculated using a truncated normal distribution to accurately reflect population structure, while occupation and gender are allocated to each core agent according to preset proportions. Based on the agents' basic user profile attributes, the large model calculates and assigns each agent 3 to 5 potential interest areas, enhancing their heterogeneity and behavioral diversity. Personality traits are encoded based on the Big Five personality model, encompassing five dimensions: openness, conscientiousness, extraversion, agreeableness, and emotional stability. These dimensions will become the intrinsic driving factors influencing attitude updates and behavioral performance. Each agent has an initial attitude score for each event request to be evolved, ranging from -1 to 1, where -1 represents skepticism and +1 represents support. An attitude closer to 0 indicates a more neutral attitude, and the initial attitude distribution reflects the diversity of the real population. The agent is equipped with a personal experience memory bank to store past behavior records and interaction content. Simultaneously, environmental interaction memory records neighbor behavior and external recommendation information, enabling the agent to make appropriate behavioral decisions based on the most relevant and urgent memories. This multi-layered initialization provides a foundation for subsequent complex interactions and rumor propagation among agents, while ensuring that cognitive differences and behavioral diversity are reflected in the simulation. Furthermore, the environment server initializes the social network topology using a directed graph. This represents the attention relationships between agents. At different time steps, agent i uses the social network... Receive from incoming follower list and outgoing follower list The message allows social networks (for brevity, (The representation of time t is omitted) undergoes adaptive changes over time.

[0094] To ensure high accuracy and efficiency in large-scale user simulations, a dynamic hierarchical module constructs a propagation potential and content richness evaluation system, identifies core agents that play a crucial role in event propagation, and adjusts the information interaction methods between agents based on agent type. The module selects core agents with higher propagation potential and content diversity from regular agents. These core agents are driven by a Large Language Model (LLM), supporting complex natural language interaction and reasoning capabilities, endowing them with human-like behavioral decision-making abilities. Regular agents are driven by Traditional Agent Models (ABMs), enabling rapid attitude value updates. The module dynamically and adaptively switches core agents with key event propagation characteristics at different time steps, allowing computational resources to focus on the most influential core users, thus balancing efficiency and accuracy in large-scale social simulations. Core agents interact through natural language dialogue, while regular agents utilize Agent Models (ABMs) for information transmission. Furthermore, the content generated by core agents is converted into attitude scores using the Large Language Model (LLM) and influences the attitude updates of regular agents through Agent Models (ABMs), while the attitudes of core agents remain unaffected by regular agents. The dynamic hierarchical module balances computational efficiency and simulation detail, enabling dynamic and hierarchical simulation of large-scale social networks while accurately reproducing the propagation dynamics in reality.

[0095] The core agent makes behavioral decisions based on user profiling, a memory mechanism, and information flow link prediction. First, user profiling provides the core agent with personalized background information, helping it understand its own personality and consider personalized needs and preferences in decision-making, thus making its behavior more diverse and rational. Second, the memory mechanism enables the agent to make decisions based on past behavioral records and interactions with neighbors. The agent possesses personal experience memory, allowing it to review its historical behavior, while also capturing the attitudes and behavioral responses of neighboring agents from environmental interaction memory. These memories can be retrieved, updated, and reflected upon, helping the agent make more targeted and timely decisions. Finally, the information flow link prediction method helps the agent analyze the evolution of social relationships. By quantifying attitude similarity, content timeliness, and tweet influence, this process helps the agent identify potential social connections and interaction opportunities, recommending appropriate tweets or interaction targets. By integrating the latest information flow, personalized needs, and historical memory, the core intelligent agent can flexibly make seven types of behavioral decisions based on the current context, such as uploading tweets, forwarding tweets, commenting on tweets, following other intelligent agents, unfollowing other intelligent agents, liking tweets uploaded by others, and doing nothing. Ultimately, this drives interaction and information dissemination in social networks, while also prompting dynamic changes in social relationships and attitudes.

[0096] The behavioral decision-making module for conventional agents in unequal social relationships primarily addresses the issues of unequal interactions and dynamic social relationship adjustments in real-world social networks by introducing the concept of trust and a dynamic link prediction engine. First, to quantify unequal attitude interactions between users, this invention introduces the concept of trust. Trust is used to measure the degree of trust between conventional agent i and its neighboring agent j. Agents with higher trust levels or more inbound followers have a greater impact on the decision-making process. During the attitude update process, the attitude of conventional agent i... Based on the set of neighboring intelligent agents The influence of each neighbor is adjusted based on a weight. The measurement is weighted by considering trust level and the number of inbound followers in the social network. Secondly, the dynamic link prediction engine simulates the dynamic adjustment of social relationships through multi-dimensional factors such as attitude similarity, one-way following, user influence, and trust level. Based on these factors, the dynamic link prediction engine predicts possible new or broken follower relationships between agents. This is achieved by calculating a missing link prediction score. And fake link identification score The dynamic link prediction engine can assess the likelihood of interactions between agents and adjust links in the social network accordingly. Specifically, when new social relationships are established, agents need to build new trust relationships through trust propagation paths. During the simulation, regular agents... Start the time wheel, press By periodically performing link prediction, the probability of adding new follower relationships is... The probability of unfollowing is In summary, the behavioral decision-making module for unequal social relationships enables conventional intelligent agents to better adapt to unequal interactions in the real world, while accurately capturing the evolution of social relationships and precisely aligning with the propagation dynamics of the real world.

[0097] After each simulation round, the environment server updates the agent's memory, tweet library, and social network. Based on the core agent's interactive behavior, it updates detailed information such as the number of comments, likes, and creation time for each tweet, and sends the tweet content to incoming followers to update the agent's memory. Based on the agent's following and unfollowing behavior, it updates the social network topology. The environment server provides a dynamic and scalable infrastructure for the entire simulation framework, ensuring data flow and state updates during the simulation process; these updates will influence the agent's decisions in the next round. Through continuous iteration at preset time steps, the social network simulator can perform large-scale event situational simulations and group behavior simulations.

[0098] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the flowcharts of the accompanying drawings may include multiple steps or stages, which are not necessarily completed at the same time, but may be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but may be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0099] In one embodiment, the present invention provides a computer system, which may be a server. The computer system includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores data used in the methods described above. The network interface communicates with external terminals via a network connection. The computer program is executed by the processor to implement the methods described above.

[0100] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0101] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.

[0102] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for simulating large-scale social networks based on dynamic topology, characterized in that, Used to extrapolate attitudes towards events occurring within a target user group, including: User profiles and memories are assigned to corresponding intelligent agents, and social networks are built for the intelligent agents based on the outgoing and incoming followers of users on social platforms; Dynamic evolution process: The influence metric of each agent is calculated based on the number of second-order incoming followers and the variance of their attitudes towards the event. This metric distinguishes core agents from regular agents. Core agents are driven by a large language model and interact with other core agents through natural language, influencing the attitudes of regular agents towards the event. Core agents use user profiling, memory mechanisms, and information flow-based link prediction methods to decide on interaction behaviors with other tweets or agents and adjust their social networks. Regular agents are driven by an agent model and interact with other regular agents. Regular agents update their attitudes through trust weights and adjust their social networks using a dynamic link prediction engine. The agent's memory, tweet library, and social network are also updated. The dynamic evolution process continues at each time step until the event deduction is completed and the user's attitude towards the event is obtained.

2. The method for simulating large-scale social networks based on dynamic topology according to claim 1, characterized in that, The method of calculating the influence metric of each agent by using the number of second-order incoming followers and the variance of the followers' attitudes toward the event specifically includes: Influence measurement quantifies an agent's dissemination ability by the number of second-order incoming followers, and measures the content diversity of the agent's tweets by the variance of incoming followers' attitudes towards the event. ； in, This represents the measure of the influence of agent i at time t. This represents the attitude of agent i towards the event at time t. This indicates whether agent i is interested in agent j. This indicates that agent i does not pay attention to agent j. This indicates that agent i is interested in agent j; Represents the list of incoming followers of agent i; This indicates whether agent j is interested in agent k. This indicates that agent j does not pay attention to agent k. This indicates that agent j is interested in agent k.

3. The method for simulating large-scale social networks based on dynamic topology according to claim 2, characterized in that, The distinction between core agents and regular agents based on influence metrics specifically includes: At each time step, select the top K agents with the highest influence metrics as core agents, and the other agents as regular agents.

4. The method for simulating large-scale social networks based on dynamic topology according to claim 1, characterized in that, The core intelligent agent, through user profiling, memory mechanisms, and information flow-based link prediction methods, decides on interaction behaviors with other tweets or intelligent agents and adjusts the social network, specifically including: Calculate the recommendation scores of core agent i and candidate tweet j. , It integrates the content representation of the latest tweets from the core intelligent agent i. Content representation of candidate tweets Content matching degree, lifecycle factor of candidate tweets, and content influence of candidate tweet j : ； in, Represents cosine similarity. It is a natural constant. Indicates the publication time of candidate tweet j. This represents the index at time t. This represents the decay rate of the life cycle factor. The recommendation list of core agent i is formed based on the recommendation score: among the agents not followed by core agent i, agents that meet the following conditions are selected: the number of incoming followers exceeds a set threshold or the outgoing followers of core agent i; the tweets published by the selected agents are used as candidate tweets, and the N tweets with the highest recommendation scores are selected from the candidate tweets to form the recommendation list. The core agent i makes decisions based on user profiles, memories, messages received from neighboring agents, and personalized information streams based on recommendation lists, and obtains interaction behaviors with other tweets or agents and adjusts its own attention relationships.

5. The method for simulating large-scale social networks based on dynamic topology according to claim 4, characterized in that, The core intelligent agent i makes decisions based on user profiles, memories, received messages from neighboring intelligent agents, and personalized information streams based on recommendation lists, resulting in interactive behaviors with tweets or other intelligent agents, specifically including: For the decision-making process, User profile for core intelligent agent i This represents the memory of the core agent i at time t-1. The personalized information flow of the core intelligent agent i at time t. The message received by the core agent i from its neighboring agents at time t-1; This represents the set of actions performed at time t. The actions in the set include: uploading a tweet, retweeting a tweet, commenting on a tweet, following other agents, unfollowing other agents, liking tweets uploaded by others, and doing nothing.

6. The method for simulating large-scale social networks based on dynamic topology according to claim 1, characterized in that, The conventional agent updates its attitude through trust weights, specifically including: Based on trust quantification, the attitude interaction relationship between a regular agent and other agents is quantified. Agents with higher trust or more incoming followers are assigned higher weights. ； ； in, This represents the attitude of a regular agent i towards an event at time t. Subject to neighboring intelligent agents The weight of influence Indicates trust in weight The proportion in This represents the degree of trust that agent i has in agent j. Let i be the set of neighboring agents of a regular agent i at time t. For agent j, there is a list of incoming followers. Re-weight attitudes.

7. The method for simulating large-scale social networks based on dynamic topology according to claim 1, characterized in that, The adjustment of social networks using a dynamic link prediction engine specifically includes: Predict scores using missing links between regular agents i and j. And fake link identification score Modeling the creation and discontinuation of social relationships. The highest-ranking agent j is the outgoing follower that the regular agent i newly focuses on. The highest-ranking agent j is the latest outgoing follower who unfollowed the regular agent i.

8. The method for simulating large-scale social networks based on dynamic topology according to claim 7, characterized in that, Missing link prediction score And fake link identification score The calculation method is as follows: ； ； in, This represents the attitude of a regular intelligent agent i towards an event at time t. This represents the attitude of agent j towards the event at time t. For agent j, there is a list of incoming followers. This represents the set of candidate agents that a regular agent i can follow in a social network. The elements of the candidate agent set include the outgoing followers of regular agent i and agents whose number of incoming followers exceeds a set value. This indicates whether agent j pays attention to the regular agent i. This is the list of outgoing followers for a regular intelligent agent i. This represents the degree of trust that agent i has in agent j.

9. A method for simulating large-scale social networks based on dynamic topology according to claim 8, characterized in that, The degree of trust that agent i has in agent j Established based on agent k, who acts as an intermediary in all trust propagation paths: ； ； in, This represents the trust relationship established between agent i and agent j based on agent k in the trust propagation path.

10. A computer system comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 9.

Images