A social emergent phenomenon simulation system and method based on heterogeneous memory

Abstract

The application discloses a social emergence phenomenon simulation system and method based on heterogeneous memory, and belongs to the cross field of computational social science and artificial intelligence. The method comprises the following steps: initializing a plurality of individual agents in a social network, and configuring a cognitive core comprising a first memory module, a second memory module and a decision module for each individual agent, the first memory module stores long-term stable beliefs, and the second memory module stores recent context information; when the individual agent receives new information, the first memory module and the second memory module are called by the decision module, the credibility of the new information is calculated based on the relationship between the new information, the long-term belief and the recent information; and the decision module selects to execute an action according to the credibility. The application realizes the simulation from micro complex decision to macro phenomenon emergence by constructing the agent with heterogeneous memory, enhances the reality and interpretability of the simulation, and can effectively capture nonlinear social phenomena such as public opinion reversal.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of computational social science and artificial intelligence, specifically to a simulation system and method for social emergence phenomena based on heterogeneous memory. Background Technology

[0002] Simulation of social phenomena is of great significance in fields such as computational social science. In existing technologies, one approach employs traditional computational models, such as the susceptible-infected-recovery model or cellular automata model. These models typically simplify individuals in society into nodes with a single state, interacting through pre-defined, homogeneous rules. While these models can simulate macroscopic diffusion trends, they neglect the complexity and heterogeneity of individuals, failing to reflect heterogeneous behaviors driven by complex individual beliefs, and thus struggling to explain complex social phenomena such as sudden reversals of public opinion.

[0003] Another approach is the multi-agent system based on large language models, which has emerged in recent years. While agents in these systems possess some language interaction capabilities, their memory mechanisms are typically limited to a finite context window, lacking a structured belief system for maintaining long-term identities and values. This leads to inconsistent behavior or identity drift in long-term simulations, reducing the realism of the simulation.

[0004] Furthermore, existing technologies also employ methods for analysis and prediction using both long-term and short-term data. For example, in the health field, risk is assessed and abrupt changes in risk values ​​can be detected by analyzing an individual's long-term health records and short-term daily behavior data. However, these methods primarily focus on time-series data analysis of a single object, aiming at prediction and early warning, and do not address a complex system composed of a large number of interacting individuals. They lack mechanisms that enable individuals to make complex decisions based on their intrinsic beliefs, and they also lack means to simulate social interactions between individuals. Therefore, they cannot be directly applied to simulate macro-level social phenomena spontaneously formed by a large number of heterogeneous individuals through complex interactions.

[0005] In summary, existing technologies generally suffer from a disconnect between macroscopic phenomena and microscopic behaviors. In other words, there is a lack of an effective mechanism that enables individual intelligent agents in simulation systems to evaluate dynamically changing social information based on long-term, stable internal beliefs, thereby driving the spontaneous emergence of complex, quantifiable, and identifiable social phenomena. Summary of the Invention

[0006] The purpose of this invention is to provide a simulation system and method for social emergent phenomena based on heterogeneous memory, aiming to solve the technical problem that existing social simulation models cannot effectively simulate macro-emergent phenomena driven by micro-complex behaviors because they cannot take into account both the long-term consistency of individual beliefs and the complexity of short-term decisions.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A simulation system for social emergence based on heterogeneous memory includes:

[0009] The individual intelligent agent module is responsible for creating and managing several individual intelligent agents. Each individual intelligent agent is configured with a cognitive core including a first memory module, a second memory module, and a decision module. The first memory module is used to store first data representing the long-term stable beliefs of the individual intelligent agent. The second memory module is used to store second data representing the recent context information of the individual intelligent agent. The decision module is responsible for coordinating the use of information in the long-term memory module and the short-term memory module.

[0010] The dynamic interaction network module allows the individual intelligent agent to receive information from other individual intelligent agents and feed back the action information generated by its decision to the network, thereby influencing other individual intelligent agents.

[0011] The global state monitoring module is used to periodically aggregate the state data of all or some of the individual intelligent agents from the individual intelligent agent module, thereby forming a macro state index, and judging whether a specific social emergence phenomenon has occurred based on the time series changes of the macro state index.

[0012] As a further aspect of the present invention: the first memory module is a knowledge graph.

[0013] As a further aspect of the present invention: the first memory module is a structured JSON file.

[0014] As a further aspect of the present invention: the second memory module is a vector database.

[0015] As a further aspect of the present invention: the decision-making module is based on a belief-desire-intention model.

[0016] As a further aspect of the present invention, the decision-making module is a rule-based decision engine.

[0017] This invention also provides a method for simulating social emergence phenomena based on heterogeneous memory, comprising the following steps:

[0018] Initialization steps: Initialize several individual intelligent agents in the social network, and independently configure a cognitive core for each individual intelligent agent. The cognitive core includes:

[0019] The first memory module is used to store first data representing the long-term stable beliefs of the individual intelligent agent;

[0020] A second memory module is used to store second data representing the recent context information of the individual intelligent agent; and a decision module;

[0021] Evaluation steps: When the individual intelligent agent receives new information about a certain issue, the decision module collaboratively calls the first memory module and the second memory module to calculate the credibility of the new information based on the relationship between the new information and the first and second data.

[0022] Action steps: The decision-making module selects one of the preset action types to execute based on the credibility level;

[0023] Monitoring steps: The global state monitoring module periodically aggregates the contents of the first data and / or second data of multiple individual intelligent agents to form a macro state indicator, and determines whether a specific social emergence phenomenon has occurred based on whether the change of the macro state indicator within a preset time window meets a preset threshold.

[0024] As a further aspect of the present invention: in the evaluation step:

[0025] The relationship between the new information and the first data includes calculating a degree of conflict between the new information and the first data;

[0026] The relationship between the new information and the second data includes calculating an instantaneous correlation between the new information and the second data;

[0027] Furthermore, the credibility is calculated by weighted summation of the conflict degree and the instantaneous relevance.

[0028] As a further aspect of the present invention: the specific social emergence phenomenon is a reversal of public opinion; the macro-state indicator is the ratio of the number of agents who hold a supportive attitude and an opposing attitude towards a certain issue.

[0029] As a further aspect of the present invention: the individual intelligent agent is a node in a social network, the node having neighboring nodes, social attributes, and a list of followers; and, in the action step, when choosing to perform a share or forward action, the individual intelligent agent selects a target subset from its list of followers for targeted information dissemination based on its current internal state composed of its first data and second data, and the social attributes of its neighboring nodes.

[0030] Compared with the prior art, the present invention has at least the following beneficial effects:

[0031] By configuring each individual agent with an independent first memory module containing long-term stable beliefs, the consistency of the agent's behavior in long-term simulation is ensured, the deep heterogeneity of individual agents is realized, and the simulation results are closer to real society.

[0032] By introducing a decision-making module and combining it with heterogeneous memory to compute information credibility, a micro-decision-making closed loop from belief assessment to action planning is realized, so that macro phenomena are driven by the complex and interpretable internal logic of each individual intelligent agent.

[0033] Through complex, nonlinear interaction models among individual intelligent agents, typical nonlinear emergent phenomena such as self-organization, mutation, and reversal of public opinion can be effectively simulated and quantified, providing a powerful tool for social dynamics analysis. Since the action decisions of each individual intelligent agent are based on beliefs in its long-term memory and environmental information in its short-term memory, the evolution process of the entire macroscopic phenomenon can be decomposed and traced back to specific micro-decision chains, greatly enhancing the credibility and interpretability of the simulation results. Attached Figure Description

[0034] The invention will now be further described with reference to the accompanying drawings.

[0035] Figure 1 This is a structural block diagram of a social emergence phenomenon simulation system based on heterogeneous memory provided in an embodiment of the present invention;

[0036] Figure 2 This is a flowchart illustrating a method for simulating social emergence based on heterogeneous memory, provided in an embodiment of the present invention.

[0037] Figure 3 This is a schematic diagram of the decision-making process of a single intelligent agent in an embodiment of the present invention;

[0038] Figure 4 This is a sequence diagram of individual intelligent agent interactions with attention costs introduced in this embodiment of the invention.

[0039] In the diagram: 100, simulation system; 10, individual intelligent agent; 11, long-term memory; 12, short-term memory; 13, behavior planner; 20, dynamic interactive network module; 30, global state monitoring module. Detailed Implementation

[0040] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0041] Please see Figure 1 This invention provides a simulation system for social emergence phenomena based on heterogeneous memory. This system can be applied to a computer-executed virtual social environment to simulate, analyze, and predict macro-level social phenomena spontaneously formed by a large number of heterogeneous individuals through complex interactions, such as the formation, evolution, and reversal of public opinion. The core of this technical solution lies in constructing a cognitive core for each simulated individual (i.e., the agent) that includes long-term stable beliefs and short-term dynamic situations, and driving complex evaluations and decisions based on this core. By reproducing highly realistic individual behavior at the micro-level, complex group phenomena that can be quantified and identified emerge at the macro-level.

[0042] The simulation system 100 can be deployed on computing devices such as servers, computer clusters, or cloud platforms. Logically, the simulation system 100 mainly includes an individual intelligent agent module, a dynamic interactive network module 20, and a global state monitoring module 30.

[0043] The individual agent module, as the core unit of the simulation system, is responsible for creating and managing a large number of individual agents 10. Each agent 10 possesses an independent cognitive core, which simulates the human cognitive structure, specifically including long-term memory 11, short-term memory 12, and a behavior planner 13. Long-term memory 11 can be implemented as a first memory module, storing primary data representing the long-term stable beliefs of the individual agent 10, such as its worldview, values, and core principles. It should be noted that this data is typically stable or changes slowly during the simulation, forming the cornerstone of the individual agent 10's "personality." Short-term memory 12 can be implemented as a second memory module, storing secondary data representing the recent contextual information of the individual agent 10, such as recent conversation history, newly received information, and the current environmental state; this data is dynamically changing. The behavior planner 13 can be implemented as a decision-making module, acting as the brain of the individual agent 10, responsible for coordinating the use of information from long-term memory 11 and short-term memory 12, evaluating external inputs, and planning and deciding on the next action based on the evaluation results and its own internal motivation.

[0044] The dynamic interaction network module 20 constructs a social environment for information exchange among individual intelligent agents 10, defining the connection relationships (such as following and friendship relationships) and information propagation rules among individual intelligent agents 10. Individual intelligent agents 10 in the individual intelligent agent module receive information from other individual intelligent agents 10 through the dynamic interaction network module 20, and feed back their decisions and actions (such as publishing new information) into the network, thereby influencing other individual intelligent agents 10.

[0045] The global state monitoring module 30 acts as a "social observer." It does not directly participate in micro-level interactions, but rather periodically aggregates state data from all or some individual agents 10 from the individual agent modules (e.g., retrieving their fundamental stance on a specific issue from long-term memory 11, or their current viewpoint from short-term memory 12), thereby forming macro-level state indicators. Based on the time-series changes of these macro-level state indicators, the global state monitoring module 30 determines whether specific social emergent phenomena have occurred.

[0046] Please see Figure 2 This application also provides a method for simulating social emergence phenomena based on heterogeneous memory. The execution process of this method may include the following steps:

[0047] Step S201: Initialize the social network and individual agents 10. In this step, the system first constructs a social network topology, such as a small-world network or a scale-free network, to simulate the structure of real social relationships. Subsequently, a specified number of individual agents 10 are created and initialized on the network nodes. For all individual agents 10, the system independently configures a cognitive core for them, that is, generates or loads unique long-term memories 11 and empty short-term memories 12, and configures a behavior planner 13.

[0048] Step S202: Inject initial events or information. In order to start the simulation, the system injects initial information into one or more specific individual agents 10 in the network. This can simulate the outbreak of news events, the emergence of new topics, or the initial spread of rumors in the real world.

[0049] Step S203: Individual agent 10 executes the "receive-evaluate-act" loop. This is the core driving loop of the simulation. Within each simulation time step, all activated individual agents 10 execute a complete internal decision-making process.

[0050] Step S204: The global state monitor collects macroscopic state statistics. At each time step or every few time steps, the global state monitoring module 30 traverses all individual agents 10, collects their key state data, and aggregates it into macroscopic statistical indicators.

[0051] Step S205: Determine whether an emergent event has occurred. The global state monitoring module 30 uses a preset algorithm model to determine whether a specific social emergent phenomenon (such as a reversal of public opinion) has occurred based on the time series of macro-state indicators obtained in step S204.

[0052] Step S206: Determine if the simulation termination condition has been met. The system checks whether the preset termination condition is met, such as reaching the maximum simulation time step, network state convergence, or successful detection of the target phenomenon. If the termination condition is not met, the process returns to step S203 to continue the simulation for the next time step; if it is met, the simulation ends.

[0053] Example 1

[0054] In initialization step S201, the simulation system 100 constructs a scale-free network containing 1000 nodes as the underlying structure of the dynamic interaction network module 20 to simulate the "opinion leader" phenomenon prevalent in social networks. Next, the system instantiates an individual agent 10 on each network node, forming an individual agent module. For each agent 10, the system configures an independent cognitive core. Specifically, its long-term memory 11 is implemented using a knowledge graph structure. For example, for individual agent A, its knowledge graph may contain triples such as (agent A, core belief, believing in science) and (agent A, trusted entity, official news agency), which constitute its long-term stable first data. Its short-term memory 12 is implemented using a vector database, initially empty, to store subsequently received information and its vectorized representation, serving as its dynamically changing second data. Its behavior planner 13 is implemented using a belief-desire-intention model, which enables the agent to form and execute action intentions based on its beliefs (obtained from long-term memory 11 and short-term memory 12) and intrinsic desires (such as pre-set motivations such as "seeking truth" and "expanding influence").

[0055] In the initial event injection step S202, the system selects the five agents with the highest connectivity in the network (i.e., "opinion leaders") and injects a rumor about a certain issue into their short-term memory 12, such as: "City X has suffered serious water pollution due to the construction of a PX project."

[0056] Subsequently, the simulation enters the core loop phase. In loop step S203, the agent 10, having received the rumor information, begins making a decision. (See also...) Figure 3 This diagram details the decision-making process within a single agent 10. Taking agent B, which receives a rumor, as an example, its decision-making process is as follows:

[0057] Step S301: Receive information. Individual agent B receives the aforementioned rumor information from its neighboring nodes.

[0058] Step S302: Calculate the conflict degree. The behavior planner 13 of individual agent B initiates the evaluation process, first calling its long-term memory 11 (knowledge graph). Assume that the knowledge graph of individual agent B contains a core belief (Agent B, Core Belief, Believe in Official Releases). Through graph reasoning, the behavior planner 13 discovers a potential conflict between the rumor that "the water source is severely polluted" and the fact that "the official information has not yet been released," and quantifies this conflict as a high conflict degree score, such as 0.8. This process reflects an evaluation based on the relationship between new information and primary data.

[0059] Step S303: Calculate relevance. Simultaneously, the behavior planner 13 invokes its short-term memory 12 (vector database) to vectorize the rumor information and performs a similarity search in the vector database to determine if the topic has been discussed recently. Assuming this is the first time the topic has appeared, and the search results are empty or have very low relevance, the behavior planner 13 calculates a low immediate relevance score, such as 0.3. This process embodies an evaluation based on the relationship between new information and secondary data.

[0060] Step S304: Calculate credibility. The behavior planner 13 calculates the final credibility of the rumor based on a preset function, combining conflict level and immediate relevance. In this embodiment, credibility can be calculated using a weighted summation method, for example,

[0061] Credibility = 0.3 * (1 - Conflict) + 0.7 * Immediate Relevance.

[0062] Substitute the values,

[0063] Credibility score = 0.3*(1-0.8)+0.7*0.3=0.06+0.21=0.27, which is a low credibility score.

[0064] Step S305: Select Action. The Behavior Planner 13's Belief-Desire-Intention Model makes a decision based on the calculated low confidence level (0.27) and the individual agent B's current intrinsic desire (assumed to be "truth-seeking"). The low confidence level triggers the intention to "question".

[0065] Step S306: Execute the action. Accordingly, individual agent B ultimately chooses to execute the "comment / question" action and publishes a new message to the dynamic interactive network module 20: "Information source unknown, awaiting official notification."

[0066] In contrast, for another agent C in long memory 11, lacking relevant beliefs or holding "anti-establishment" beliefs, its calculated conflict level might be low, leading to higher credibility. If its desire is to "expand its influence," it might choose the "share / forward" action. In performing this action, agent C would select a target subset from its follower list based on the social attributes of its neighboring nodes (e.g., choosing neighbors with similar beliefs), and conduct targeted information dissemination, rather than broadcasting to all neighbors.

[0067] At the macro level, as the simulation progresses, in the global state monitoring step S204, at the end of each time step, the global state monitoring module 30 counts the number of individual agents in the network who support (forward or like rumors) and oppose (question or refute rumors) the issue of "PX project causing pollution", and these are respectively denoted as the number of supporters N-Pro and the number of opponents N-Con.

[0068] At the 50th time step of the simulation, the system simulates an "official clarification" event, injecting a highly authoritative message into the network: "Official announcement: The PX project in City X meets environmental standards, and the water source is not polluted." A large number of agents like agent B, who hold the belief that "official announcements are true," receive this message, calculate its extremely high credibility, and actively forward it or refute existing rumors. This leads to a sharp increase in the number of opponents (N-Con) and a corresponding decrease in the number of supporters (N-Pro).

[0069] At this point, in step S205 of determining the emergent event, the global state monitoring module 30 initiates its determination algorithm. In this embodiment, the parameters of the monitoring algorithm can be set as follows: the preset time window Tau is 5 simulation time steps, and the preset reversal threshold Theta is 20%. The global state monitoring module 30 detects that within the time window Tau, from simulation time t to 55, the majority party in the network changes from supporters (N-Pro > N-Con) to opponents (N-Con > N-Pro), and the absolute value of the change in the majority party's proportion in the total population exceeds the reversal threshold Theta. Therefore, the system determines that a social emergent phenomenon of "opinion reversal" has occurred at this moment. Subsequently, after reaching the preset total simulation time, the system can determine in step S206 that the termination condition is met and end the simulation.

[0070] This embodiment successfully simulates the macro-level public opinion reversal phenomenon driven by complex, belief-based micro-decision-making by individuals through the detailed steps described above, and achieves quantitative identification of this phenomenon, thereby verifying the effectiveness of the proposed solution.

[0071] Example 2

[0072] Based on Example 1, this embodiment introduces a more refined interaction mechanism, namely "attention cost," which aims to simulate the filtering behavior of individuals when faced with information overload in the real world, thereby enhancing the realism of the simulation.

[0073] The system architecture and overall process of this embodiment are basically the same as those of Embodiment 1. The core improvement lies in the internal logic of the agent executing the "receive-evaluate-act" loop step S203, especially after the information receiving step S301, an additional "attention filtering" step is added.

[0074] Please see Figure 4 This demonstrates the process by which information is transmitted from a sending individual intelligent agent i to a receiving individual intelligent agent j via a dynamic interactive network. Unlike in Example 1, where individual intelligent agent j unconditionally processes all received information, in this example, individual intelligent agent j first performs an internal "attention cost calculation" operation before initiating the complete belief evaluation process (i.e., steps S302 to S306 in Example 1).

[0075] Specifically, when individual agent j receives information from individual agent i, its behavior planner 13 calculates the attention cost score according to a preset formula.

[0076] The formula can be designed as follows:

[0077] ;

[0078] in, It is the influence score of the individual intelligent agent i that is the source of the information (e.g., its connectivity in the network or the number of times it has been forwarded in history). It is the distance between the core beliefs of individual agents i and j (e.g., calculated by comparing the vector embedding representations of their knowledge graphs in long-term memory 11), while and It is an adjustable weighting coefficient used to balance the importance of the influence of the information source and the homogeneity of beliefs.

[0079] Individual agent j has a preset attention threshold, for example, 0.6. The calculated attention cost score will be compared with this threshold. Figure 4As shown, there is a choice between two options: if the calculated cost is below a threshold (e.g., the information comes from a highly influential agent i with similar views), agent j considers the information "worthy of attention" and continues with a series of cognitively demanding operations such as assessing credibility and selecting and executing actions. Conversely, if the cost is not below the threshold (e.g., the information comes from a completely unfamiliar agent with very low influence and vastly different views), agent j will directly "ignore the information," and the processing of this information will end there, without proceeding to the subsequent deep evaluation stage.

[0080] By introducing this attention cost mechanism, information dissemination in this embodiment exhibits more complex dynamic characteristics. Compared to the relatively uniform diffusion of information across the network in Embodiment 1, information dissemination in this embodiment will no longer indiscriminately trigger deep thinking in every receiver. Information sources with low influence or whose beliefs differ significantly from those of the receivers are likely to have their information filtered out in large quantities in the early stages of dissemination, making it difficult to generate a large-scale impact. This makes the formation and evolution of public opinion more dependent on the driving force of high-influence nodes and rapid dissemination within "opinion circles" or "information cocoons" with similar beliefs. Understandably, the simulation results of this embodiment can better reproduce the non-uniform information dissemination pattern caused by the limited attention and cognitive resources of individuals in real society, improving the realism and accuracy of the simulation.

[0081] Example 3

[0082] Based on Example 1, this embodiment further introduces a dynamic network evolution and feedback intervention mechanism, aiming to simulate the dynamic changes in social relations due to the interaction of viewpoints, and to explore the possibility of external forces guiding the direction of public opinion.

[0083] The overall process in this embodiment still follows... Figure 2 The framework shown is used, but two functional modules are added to the simulation system 100: one is a network update module inside the dynamic interactive network module 20, and the other is an intervention execution module that is linked with the global status monitoring module 30.

[0084] As an optional implementation, the dynamic network evolution mechanism is implemented through a network update module. This module is activated at the end of each time step and updates the connection weights between individual agents 10 based on their interactions within that time step. Specifically, in the dynamic interaction network module 20, each connection edge, in addition to representing the connection relationship, is assigned a "closeness" weight. For example, if agent A forwards or likes information from agent B within the current time step, the network update module will increase the "closeness" weight of the connection between agent A and agent B; conversely, if agent B repeatedly questions or refutes information posted by agent B, the weight will decrease accordingly. This dynamically changing "closeness" weight will play an important role in subsequent simulations. When agent B performs a "share / forward" action again and filters a subset of targets based on its desires and neighbor attributes (as described in Example 1), the "closeness" weight will be an important consideration; that is, the higher the closeness of a neighboring node, the greater the probability that it will be selected as the target for information dissemination. This mechanism makes social network structures no longer static, but capable of reflecting changes in the closeness of relationships brought about by the interaction of viewpoints, simulating the social phenomenon of "birds of a feather flock together".

[0085] Correspondingly, the feedback intervention mechanism is implemented by the intervention execution module, which is tightly coupled with the global state monitoring module 30. During the simulation, the global state monitoring module 30 continuously performs its monitoring and judgment functions (steps S204 and S205). Once the global state monitoring module 30 determines that a specific emergent event, such as a "reversal of public opinion," has occurred, it not only records the event but also activates the intervention execution module.

[0086] Taking the scenario in Example 1 as an example, suppose that at time step 55, the system detects a "public opinion reversal" event. The intervention execution module immediately starts and executes a preset intervention strategy, such as to consolidate the effect of debunking rumors and prevent public opinion from wavering again. The intervention execution module generates a high-weighted feedback message, such as "In-depth report by authoritative media: Fully revealing the truth and scientific basis behind the PX project rumors," and directly injects it into the short-term memory 12 of a specific individual agent in the network. It should be noted that the injection target is not random, but strategically selected. For example, those with high connectivity in the network and whose views have recently shifted from support to opposition are selected as "opinion wavers." This precise information delivery simulates the hot news push or public opinion guidance operations targeting specific groups in the real world.

[0087] Through this embodiment, the simulation system not only simulates the spontaneous evolution of public opinion and the dynamic reshaping of social relations, but more importantly, it provides a powerful "virtual social experiment platform." Researchers or policymakers can design and test different information intervention strategies on this platform (e.g., changing the wording of intervention information, selecting different delivery targets, adjusting the timing of delivery, etc.), and evaluate the effectiveness of different strategies in controlling public opinion, quelling online rumors, or enhancing social consensus by observing the evolution of subsequent macro-level indicators, thereby providing scientific reference for real-world decision-making.

[0088] Example 4

[0089] This embodiment aims to illustrate that the core technical solution proposed in this application has good universality and scalability. Its innovative value lies in its unique cognitive architecture and methodology, rather than its dependence on a specific implementation technology. To this end, this embodiment uses a relatively simplified set of techniques, different from those in Embodiment 1, to implement heterogeneous memory and behavioral planning, but fully retains the core logic of "long-term beliefs plus short-term context-driven decision-making".

[0090] In this embodiment, the internal components of the individual intelligent agent module are implemented as follows:

[0091] The implementation of long-term memory 11: Instead of using a complex knowledge graph, a structured JSON file is used as the primary memory module. This file clearly stores the core beliefs of each individual agent 10 in key-value pairs. Understandably, this approach can also solidify the core beliefs of the individual agents as their long-term stable primary data.

[0092] The implementation of Behavior Planner 13: Instead of using the complete belief-desire-intention model, it adopts a simplified, rule-based decision engine as the decision-making module. This rule-based decision engine contains a series of "if-then" rules, such as:

[0093] If the credibility is less than 0.4 and the desire is 'truth-seeking', then the action is 'questioning'.

[0094] If the credibility is greater than 0.7 and the desire is 'to expand influence', then the action is 'to share'.

[0095] Simplification of Credibility Calculation: In the evaluation step, the method for calculating the conflict degree (step S302) is simplified accordingly. For example, the behavior planner 13 parses new information, extracts the source entity, and then searches in long-term memory in JSON format whether the entity exists in the "trusted entity" or "untrusted entity" list to determine conflict and assign a conflict degree score. The function for finally calculating credibility (step S304) is also simplified to a linear weighted summation formula: ;

[0096] in, It is the final credibility. It is the calculated degree of conflict. The immediate correlation is calculated from short-term memory 12 (which can still be implemented using a vector database). and These are the weighting coefficients.

[0097] Although the underlying technology has changed, the overall workflow of this embodiment is completely consistent with that of Embodiment 1. After receiving information, the individual intelligent agent's rule-based decision engine still collaboratively calls the long-term memory 11 in JSON format and the short-term memory 12 implemented by the vector database to calculate the conflict degree and relevance in a simplified way, and obtains the credibility score using a linear formula. Finally, it matches the rule base according to the score to determine and execute the final action. At the macro level, the operation mechanism of the global state monitoring module 30 includes using the time window Tau and the reversal magnitude threshold Theta to detect public opinion reversals (such as...). Figure 4 As shown in the figure, it is exactly the same as in Example 1.

[0098] Running this embodiment to conduct the same rumor propagation simulation experiment as in Embodiment 1, the expected results show that, despite the simplification of the agent's decision-making logic and memory structure, the system can still successfully reproduce the core phenomenon of micro-belief driving macro-level public opinion evolution, including the initial spread of rumors, the change of viewpoints after the intervention of official information, and can successfully identify public opinion reversal events through the macro-monitoring mechanism.

[0099] The success of this embodiment demonstrates the innovation and beneficial effects of the proposed solution, rooted in its comprehensive methodological framework of "heterogeneous memory (long-term / short-term) + belief assessment + behavior planning + macro-monitoring." Whether employing a complex implementation such as knowledge graphs and belief-desire-intention models, or a simplified implementation such as JSON files and rule engines, as long as this core framework is followed, the expected technical effects can be achieved. This fully demonstrates the robustness and versatility of the proposed solution, providing flexibility for deployment in scenarios with varying technical capabilities and application requirements, and supporting the corresponding scope of protection.

[0100] The preferred embodiments of the present invention have been described in detail above and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A simulation system for social emergence phenomena based on heterogeneous memory, characterized in that, include: The individual intelligent agent module is responsible for creating and managing several individual intelligent agents. Each individual intelligent agent is configured with a cognitive core including a first memory module, a second memory module, and a decision module. The first memory module is used to store first data representing the long-term stable beliefs of the individual intelligent agent. The second memory module is used to store second data representing the recent context information of the individual intelligent agent. The decision module is responsible for coordinating the use of information in the long-term memory module and the short-term memory module. The dynamic interaction network module allows the individual intelligent agent to receive information from other individual intelligent agents and feed back the action information generated by its decision to the network, thereby influencing other individual intelligent agents. The global state monitoring module is used to periodically aggregate the state data of all or some of the individual intelligent agents from the individual intelligent agent module, thereby forming a macro state index, and judging whether a specific social emergence phenomenon has occurred based on the time series changes of the macro state index.

2. The social emergence phenomenon simulation system based on heterogeneous memory according to claim 1, characterized in that, The first memory module is a knowledge graph.

3. The social emergence phenomenon simulation system based on heterogeneous memory according to claim 1, characterized in that, The first memory module is a structured JSON file.

4. The social emergence phenomenon simulation system based on heterogeneous memory according to claim 1, characterized in that, The second memory module is a vector database.

5. A simulation system for social emergence based on heterogeneous memory according to claim 1, characterized in that, The decision-making module is based on a belief-desire-intention model.

6. The social emergence phenomenon simulation system based on heterogeneous memory according to claim 1, characterized in that, The decision-making module is a rule-based decision engine.

7. A method for simulating social emergence phenomena based on heterogeneous memory, characterized in that, Includes the following steps: Initialization steps: Initialize several individual intelligent agents in the social network, and independently configure a cognitive core for each individual intelligent agent. The cognitive core includes: The first memory module is used to store first data representing the long-term stable beliefs of the individual intelligent agent; A second memory module is used to store second data representing the recent context information of the individual intelligent agent; and a decision module; Evaluation steps: When the individual intelligent agent receives new information about a certain issue, the decision module collaboratively calls the first memory module and the second memory module to calculate the credibility of the new information based on the relationship between the new information and the first and second data. Action steps: The decision-making module selects one of the preset action types to execute based on the credibility level; Monitoring steps: The global state monitoring module periodically aggregates the contents of the first data and / or second data of multiple individual intelligent agents to form a macro state indicator, and determines whether a specific social emergence phenomenon has occurred based on whether the change of the macro state indicator within a preset time window meets a preset threshold.

8. The method for simulating social emergence based on heterogeneous memory according to claim 7, characterized in that, In the evaluation steps: The relationship between the new information and the first data includes calculating a degree of conflict between the new information and the first data; The relationship between the new information and the second data includes calculating an instantaneous correlation between the new information and the second data; Furthermore, the credibility is calculated by weighted summation of the conflict degree and the instantaneous relevance.

9. A method for simulating social emergence phenomena based on heterogeneous memory according to claim 7 or 8, characterized in that, The specific social emergence phenomenon is a reversal of public opinion; the macro-state indicator is the ratio of agents who support and oppose a certain issue.

10. A method for simulating social emergence phenomena based on heterogeneous memory according to claim 7, characterized in that, The individual intelligent agent is a node in a social network, and the node has neighboring nodes, social attributes, and a list of followers; and in the action step, when choosing to perform the share or forward action, the individual intelligent agent selects a target subset from its list of followers for targeted information dissemination based on its current internal state composed of its first data and second data, as well as the social attributes of its neighboring nodes.

Images