Rumor simulation analysis method and system based on large language model agent

Abstract

The application provides a rumor simulation analysis method and system based on a large language model agent, comprising determining initialization agent feature information and constructing a certain number of role agents according to a preset propagation network topology structure; based on the unprocessed speeches of the role agents at the current time step, a large language model is used to simulate human thinking activities to obtain the mutant rumors generated by the role agents at the current time step for language propagation; rumor propagation simulation is completed based on the propagation network topology structure and the mutant rumors at each time step; in each time step, the propagation probability between the role agents is determined according to the propagation distance and the credibility of each role agent; after multiple iterations, the role influence of each role agent is determined according to the propagation probability and the view intensity of the role agent. The application achieves the technical effect of high-precision simulation of the rumor propagation process in a complex social network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, more specifically, to a rumor simulation analysis method and system based on a large language model agent. BACKGROUND

[0002] As an important platform for information dissemination, social networks play a crucial role in modern society. However, the proliferation of information poses a serious threat to the security and stability of social networks. The spread of rumors in social networks is complex and variable, and its transmission path is not only affected by individual cognitive differences and emotional tendencies, but also closely related to social relationship networks and information environment. Traditional rumor statistical models have obvious shortcomings in accurately depicting the dynamic evolution process of rumor transmission. At the same time, existing rumor tracing and intervention technologies mostly rely on post-data backtracking analysis, lacking the ability to simulate and predict rumor transmission paths in real time.

[0003] Existing technologies include building rumor transmission simulation models to simulate and predict rumors in real time, such as SIR models and complex network models, but still have the following drawbacks: 1) the behavior modeling is relatively simple, simplifying individuals as homogeneous nodes; however, in actual human language interaction processes, subjective judgment, emotional expression, and dynamic decision-making play an important role, and traditional models have low fidelity in rumor simulation scenarios. 2) The network topology is static, making it difficult to adapt to the influence of social network heterogeneity on transmission paths. 3) The interaction mechanism is not perfect, lacking dynamic simulation of complex behaviors such as information revision, secondary creation, and group discussion.

[0004] Therefore, there is an urgent need for a rumor simulation analysis method that can efficiently simulate and predict. SUMMARY

[0005] In view of the above problems, the purpose of the present application is to provide a rumor simulation analysis method and system based on a large language model agent to solve at least one problem existing in the prior art.

[0006] According to one aspect of the present application, a rumor simulation analysis method based on a large language model agent is provided, applied to an electronic device, comprising:

[0007] According to the preset transmission network topology structure, the initialization agent feature information is determined and a certain number of role agents are constructed; the initialization agent feature information includes the to-be-processed speech and the role assignment information of the agent corresponding to each network node;

[0008] Based on the pending speech of the role agent at the current time step, a variant rumor generated by the role agent at the current time step for language propagation is obtained by simulating human thinking activities through a large language model; rumor propagation simulation is completed based on the propagation network topology and the variant rumor at each time step;

[0009] In each round of time steps, the propagation probability between each role agent is determined according to the propagation distance and the credibility of each role agent; after multiple rounds of iteration, the role influence of each role agent is determined according to the propagation probability and the view intensity of the role agent;

[0010] According to the role influence, the key network node in the propagation network topology is determined.

[0011] In addition, the optional technical solution is that the role includes an ordinary user, an opinion leader and a rumor maker.

[0012] In addition, the optional technical solution is that the propagation network topology includes one or more of a small-world network, a scale-free network and a random network;

[0013] When the number of propagation network topologies is more than two, the propagation network topologies are dynamically switched.

[0014] In addition, the optional technical solution is,

[0015] Based on the pending speech of the role agent at the current time step, a variant rumor generated by the role agent at the current time step for language propagation is obtained by simulating human thinking activities through a large language model

[0016] It is realized by the following formula:

[0017]

[0018] Wherein, the large language model is LLM large language model, is a role agent, is a variant rumor; the basic rumor is the pending speech of the role agent at the current time step, and the context information includes historical propagation records and node interaction behaviors; the information of the role agent includes age, gender and working condition.

[0019] In addition, the optional technical solution is that the propagation probability between each role agent is determined according to the propagation distance and the credibility of each role agent, which is realized by the following formula,

[0020]

[0021] Wherein, is a network a role agent in the network; a neighbor of the node; a trust degree of the role agent; a shortest path distance between nodes, a spatial attenuation coefficient. Further, the trust degree of each role agent is realized by the following formula,

[0022]

[0023]

[0024] wherein, is a role influence coefficient, is an agent in the network.

[0025] Further, the viewpoint strength of the role agent is realized by the following formula,

[0026]

[0027] wherein, when the role agent successfully spreads rumors to the neighbor, the viewpoint strength of the role agent is expressed as , and the viewpoint strength ∈[0, 1] represents the subjective acceptance degree of the agent to the rumor; is a trust degree of the role agent, is a role acceptance weight.

[0028] In another aspect, the present application also provides a rumor simulation analysis system based on a large language model agent, which utilizes the rumor simulation analysis method based on a large language model agent as described above to perform rumor simulation analysis; comprising:

[0029] a role agent construction unit configured to determine initialization agent feature information and construct a predetermined number of role agents according to a preset propagation network topology structure; the initialization agent feature information includes to-be-processed speeches and role assignment information of agents corresponding to each network node;

[0030] a propagation simulation unit configured to obtain a mutated rumor generated by language propagation of the role agent at a current time step through a large language model to simulate human thinking activities based on to-be-processed speeches of the role agent at the current time step; and complete rumor propagation simulation based on the propagation network topology structure and the mutated rumor at each time step;

[0031] ​​​​​​The role influence determination unit is configured to determine a propagation probability between each role agent according to a propagation distance and a credibility of each role agent in each time step; and determine a role influence corresponding to each role agent according to the propagation probability and a viewpoint intensity of the role agent after multiple rounds of iteration; and determine a key network node in the propagation network topology according to the role influence.

[0032] The rumor simulation analysis method and system based on the large language model agent of the present application fully consider subjective judgment, emotional expression and dynamic decision-making and other factors in the process of human language interaction in behavior modeling, successfully overcome the limitations brought by simplifying individuals into homogeneous nodes in traditional models, and make the behavior of the agent more close to the performance of real humans in rumor propagation. Different network topologies such as small-world network, scale-free network and random network are supported, and the propagation influence of different users on rumors is quantified through role assignment and credibility calculation. The rumor simulation analysis method and system based on the large language model agent of the present application achieves high-precision simulation of the rumor propagation process in a complex social network, and provides a scientific basis for rumor governance.

[0033] To achieve the above and related purposes, one or more aspects of the present application include the features to be described in detail later. The following description and drawings describe certain exemplary aspects of the present application in detail. However, these aspects indicate only some of the various ways in which the principles of the present application can be used. In addition, the present application is intended to include all these aspects and their equivalents. BRIEF DESCRIPTION OF DRAWINGS

[0034] Other objects and results of the present application will become more apparent and easy to understand by referring to the following description in conjunction with the accompanying drawings. In the drawings:

[0035] Figure 1 A flowchart of a rumor simulation method based on a large language model agent according to an embodiment of the present application;

[0036] Figure 2 A node degree distribution diagram related to network structure of a rumor simulation analysis method based on a large language model agent according to an embodiment of the present application;

[0037] Figure 3 An influence analysis diagram of different roles of a rumor simulation analysis method based on a large language model agent according to an embodiment of the present application;

[0038] Figure 4 A module schematic diagram of a rumor simulation analysis system based on a large language model agent according to an embodiment of the present application;

[0039] Figure 5An internal structure schematic diagram of an electronic device for implementing a rumor simulation analysis method based on a large language model agent is provided for an embodiment of the present application.

[0040] The same reference numbers in all the figures indicate similar or corresponding features or functions. DETAILED DESCRIPTION

[0041] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0042] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings. In the description of the embodiments of the present application, unless otherwise specified, “and / or” in the text only represents a description of the associated relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone.

[0043] Hereinafter, the terms “first” and “second” are only for descriptive purposes, and cannot be understood as implying or suggesting relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with “first” and “second” can explicitly or implicitly include one or more of the features, and in addition, in the description of the embodiments of the present application, “a plurality of” means two or more than two.

[0044] In the present specification, the reference to “one embodiment” or “some embodiments” and the like means that a particular feature, structure or characteristic described in connection with the embodiment is included in one or more embodiments of the present application. Therefore, the statements “in one embodiment”, “in some embodiments”, “in other some embodiments”, “in further some embodiments” and the like appearing in different places in the present specification are not necessarily all referring to the same embodiment, but mean “one or more but not all embodiments”, unless otherwise specifically emphasized. The terms “include”, “contain”, “have” and their variants mean “include but not limited to”, unless otherwise specifically emphasized.

[0045] The application provides a multi-agent simulation framework based on a large language model (LLM), which gives each agent the ability to make integrated decisions in terms of cognition, emotion and behavior through the LLM, enabling it to generate rumor content (such as emotional expressions and logical flaws) that conforms to human language logic, and simulate the dynamic behaviors of individuals in the propagation process, such as questioning, forwarding and correcting, effectively improving the realism of the simulation scene. The system can dynamically switch between different structures such as small-world networks, scale-free networks and random networks, and quantitatively analyze the influence mechanism of network heterogeneity on rumor propagation efficiency, coverage and life cycle. At the same time, the application supports dynamic interaction and evolution analysis, traces the entire process of rumor generation, mutation and extinction through real-time interaction between agents, extracts key propagation nodes (such as "opinion leaders"), information evolution paths and group polarization phenomena, and provides data support for developing accurate rumor-busting strategies. Compared with traditional methods, the application combines LLM and agents to realize a paradigm shift from "static statistical prediction" to "dynamic behavior simulation", providing multi-dimensional technical support for public opinion monitoring, network governance policy design, social science research and other fields.

[0046] To describe the rumor simulation analysis method and system based on the large language model agent of the application in detail, the specific embodiments of the application will be described in detail below with reference to the drawings. Embodiment 1

[0047] Figure 1 A flowchart of the rumor simulation analysis method based on the large language model agent according to the embodiment of the application is shown.

[0048] As Figure 1 shown, the rumor simulation analysis method based on the large language model agent of the embodiment of the application includes steps S110-S130.

[0049] S110: According to the preset propagation network topology structure, determine the initialization agent feature information and construct a certain number of role agents; wherein the initialization agent feature information includes the to-be-processed speech and the role allocation information of the agent corresponding to each network node.

[0050] When encountering a speech propagation scene, the rumor simulation analysis method based on the large language model agent can simulate and analyze the speech (rumor) propagation process of the event through multi-agent large language model, which can realize the risk warning of rumor propagation in social networks. First, the topology network is initialized and the characteristics of each agent in the multi-agent are set. The topology network is used to describe the social relationship between agents, where each node represents an agent, and each agent is used to simulate the behavior and characteristics of a propagation role. For the role allocation information of the agent, in addition to setting the role of the agent to be a normal user, an opinion leader or a rumor maker, the role allocation information of the agent can also include the age, gender and working status of the role.

[0051] In a specific embodiment, the propagation network topology structure comprises one or more of a small-world network, a scale-free network and a random network; when the number of propagation network topology structures is two or more, the propagation network topology structure is dynamically switched. In terms of the heterogeneity of social networks, the application discards the traditional static network topology structure and adopts a dynamically adaptive network topology model, which can accurately capture the influence of social network heterogeneity on rumor propagation paths, thereby more accurately reflecting the propagation rules of rumors in different social network environments. In the specific implementation process, through a toolkit such as networkx, different network structures such as small-world networks, scale-free networks, etc. are generated, and then evolution is performed on the network, and the nodes in the network are numbered, corresponding to the agents of the large language model. Therefore, the type of evolved network structure can be set. That is, the agents are connected to each other through the preset network topology structure, and the number of neighbors is controlled by the scale-free network, and the connection probability is controlled by the random network. k and reconnection probability β The small-world network controls the local clustering and global connectivity, generates a "hub-node" structure through a "preferential connection" mechanism, simulates the heterogeneous scale-free network of a social platform, and randomly establishes connections between nodes. For example, in the case of a small-world network, a node is set to spread rumors, which spread to neighbor nodes according to certain rules. The time of diffusion is fixed, such as 20 times of diffusion, which is 20 time steps, and the number of people infected by rumors on the network is counted each time.

[0052] Each agent is assigned a specific role, including "ordinary user", "opinion leader", "rumor maker", etc. Among them, part of the roles are randomly assigned by probability distribution (such as ordinary users accounting for 80%), and high-connectivity nodes in the network location (such as hub nodes in the scale-free network) are preferentially assigned "opinion leader" roles to simulate the characteristics of influential nodes in social networks. In addition, when building role agents, credibility modeling information is added, combining the social network properties of nodes with role weights to calculate credibility. Specifically, credibility is a quantitative indicator of the influence of rumor propagation, which is determined by the network properties of nodes and role weights. The network properties of nodes are the degree of the node, representing the influence of the node in the current environment, and the role is the role setting of the node, for example, in the traffic accident rumor propagation scenario, the credibility of the police role is 0.8.

[0053] The credibility of each role agent is realized by the following formula,

[0054]

[0055] wherein, is the role influence coefficient (such as an opinion leader value of 2 and an ordinary user value of 1), For agents in the network Ensure that nodes with high connectivity or high role weight have higher credibility, u For agents in the network V .

[0056] In general, the present application is to construct a heterogeneous multi-agent system based on large language model, through agent role allocation, credibility modeling, opinion intensity evolution and network connection mechanism, to realize high realistic simulation of rumor propagation process. In the improvement of the interaction mechanism, the present application innovatively introduces the dynamic simulation of complex behaviors such as information correction, secondary creation and group discussion, greatly enriches the interaction details of rumor propagation simulation, so that the simulation process can more comprehensively and realistically reflect various complex phenomena in rumor propagation. The present application combines the semantic generation ability of large language model with the cooperative simulation advantages of multi-agent system, breaks through the limitations of traditional technology in behavior simulation, network adaptation and dynamic interaction, and provides innovative technical means for social network rumor propagation research and governance.

[0057] S120: Based on the pending speech of the role agent at the current time step, the mutated rumor generated by the role agent at the current time step for language propagation is obtained by simulating human thinking activities through a large language model; rumor propagation simulation is completed based on the propagation network topology structure and the mutated rumor at each time step.

[0058] Based on the pending speech of the role agent at the current time step, the mutated rumor generated by the role agent at the current time step for language propagation is obtained by simulating human thinking activities through a large language model, which is realized by the following formula:

[0059]

[0060] Wherein, the large language model is LLM large language model, is the role agent, is the mutated rumor; the basic rumor is the pending speech of the role agent at the current time step; the context information includes historical propagation records and node interaction behaviors; the information of the role agent includes age, gender and working condition.

[0061] Note that a Large Language Model (LLM) is a deep learning model based on artificial neural networks, mainly used for processing Natural Language Processing (NLP) tasks. LLM is usually based on the Transformer architecture, the core of which is the self-attention mechanism, allowing the model to dynamically focus on the relationship between different positions in the sequence when processing sequence data, thus capturing long-distance dependencies. LLM usually contains multiple layers of Transformer encoders or decoders, each layer having multiple attention heads that can parallelly focus on different feature dimensions. The size of the model can be expanded by increasing the number of layers, the number of attention heads, and the number of neurons in each layer. LLM is trained through unsupervised learning on large-scale text corpora. The training objective is usually language modeling, i.e., predicting the next word or character in a text sequence given the previous words or characters. In this way, the model can learn the statistical regularities, semantic information, and grammatical structure of language.

[0062] In this invention, a Large Language Model (LLM) is used to simulate human thought processes based on the speech of a role-playing agent, thereby generating variant rumors through language propagation selection. To fine-tune the LLM for the application scenario of this invention, making it better suited to the specific needs of rumor propagation simulation, the generation of realistic rumor content and the simulation of agent behavior can be performed according to the following steps: First, data collection and preprocessing are conducted. Specifically, public opinion data is collected from platforms such as social media and news websites, including users' basic information, participation in events, comments, forwarding, and liking behaviors, and the data is cleaned and preprocessed. Next, training data is prepared, and training samples are constructed. The input consists of users' personal information, participation in events, previous comments, and behaviors, while the output consists of users' actual comments, forwarding content, or liking behaviors for specific events, and the samples are labeled. Then, the fine-tuning objective is defined. Specifically, this could be to generate rumor content that conforms to human language logic or to simulate agent propagation behavior. A pre-trained LLM is selected as the base model, such as GPT-3, GPT-4, or BERT, and the model parameters are adjusted according to the task requirements. After initializing the model, it is fine-tuned using prepared training data. The model parameters are updated using optimization algorithms (such as LoRA, PPO, etc.) to minimize the loss function. Appropriate evaluation metrics, such as perplexity and BLEU score, are selected, and the model is validated using a validation set. Finally, the fine-tuned model is deployed to the simulation system, and the model is updated periodically based on feedback and new data. This fine-tuned LLM can more accurately simulate the behavior and speech generation process of an agent in rumor propagation, improving the reliability and realism of the simulation system.

[0063] S130: In each time step, the propagation probability between each agent is determined based on the propagation distance and the credibility of each agent. After multiple iterations, the role influence of each agent is determined based on the propagation probability and the viewpoint strength of each agent.

[0064] The implementation process of the opinion strength evolution mechanism first involves setting an initial value for the opinion strength, which is determined by the propagator (e.g., an initial propagator value of 0.8). Then, due to propagation attenuation in subsequent propagations, the opinion strength of each agent is determined by the following formula:

[0065]

[0066] Among them, when the role intelligent agent To the neighbors When a rumor is successfully spread, The strength of the viewpoint is expressed as Strength of viewpoints E [0, 1] represents the subjective acceptance degree of the agent to the rumor; The credibility of the role agent , The role acceptance weight . For example, the role acceptance weight of the opinion leader value is 0.9, and the role acceptance weight of the ordinary user value is 0.4, so as to reflect the sensitivity difference of different roles to the rumor.

[0067] The present application sets an agent propagation triggering mechanism for rumor propagation. That is, in each time step, the infected node tries to spread the rumor to the neighbor . The agent propagation triggering mechanism for rumor propagation embodies the "near neighbor first" feature, that is, the farther the distance or the lower the credibility of the propagator, the smaller the propagation probability. According to such propagation probability, the propagation strength is combined and a threshold is set to control whether the node is mutated.

[0068] The propagation probability between the role agents is determined according to the propagation distance and the credibility of each role agent, which is realized by the following formula,

[0069]

[0070] Wherein, is the role agent in the network ; is the neighbor of the node ; is the credibility of the role agent ; is the shortest path distance between nodes, is the spatial attenuation coefficient.

[0071] The pseudo code of the rumor simulation analysis method based on the large language model agent of the present application is shown in Table 1:

[0072] Table 1 Flow pseudo code

[0073]

[0074]

[0075] ​The rumor simulation analysis method based on the large language model intelligent agent of the application, for constructing a heterogeneous multi-agent system based on a large language model, realizes high-fidelity simulation of the rumor propagation process through agent role allocation, credibility modeling, opinion intensity evolution and network connection mechanism. The application breaks through the limitations of traditional technology in behavior fidelity, network adaptation and dynamic interaction by combining the semantic generation capability of the large language model and the cooperative simulation advantages of the multi-agent system, and provides an innovative technical means for social network rumor propagation research and governance. Example 2

[0076] In this example, a topic is set as a fixed initial rumor: "a serious car accident occurred at a certain intersection", which is the starting point of all rumor propagation, and subsequent propagators will generate variants with role characteristics based on it. First, the agent role allocation is performed, 10 types of roles are defined, covering different social identities, and the credibility of each role is defined, with higher values representing greater propagation influence. The specific roles and influence are as follows: "passerby": 0.3, "eyewitness": 0.8, "police officer": 0.9, "medical personnel": 0.9, "accident vehicle driver": 0.6, "nearby business owner": 0.4, "journalist": 0.7, "nearby resident": 0.4, "rescue personnel": 0.8, "taxi driver": 0.5. Police officers, medical personnel, etc. are high-credibility roles, while passersby, etc. are low-credibility roles. It should be noted that in the specific implementation process, relevant roles are set, and each node in the network corresponds to a role. The specific values of the role distribution can be set by oneself. For example, set 2 police officers, 20 passersby, etc.

[0077] Secondly, the simulation parameters are set. Specifically, the simulation uses the language model API provided by the Zhi Spectrum Company, selects the GLM-4-FLASH language model, sets the generation temperature to 0.9, and uses a small-world network with an edge reconnection probability of 0.1, a scale-free network with each new node connected to 2 edges, and a random network with an edge existence probability of 0.1. Set the agent format to 100, the time step to 10 for simulation.

[0078] Then, the network topology is dynamically adapted. In this embodiment, the propagation network topology is dynamically switched among small-world networks, scale-free networks and random networks. The node degree distribution of the three propagation network topologies is as shown in Figure 2

[0079] By observing Figure 2 ​It can be seen that the small-world network presents a "inverted U-shaped" distribution, the probability tends to 0 in the low degree area (k<5), reaches the peak (P≈0.12) at k≈15, and then rapidly decays; the scale-free network shows a power-law distribution, presents linear decay in the interval of k=10^0 to 10^1; the random network is a perfect clock curve, and the peak k is 20. Various network settings can reflect the robustness of the simulation model. In terms of the heterogeneity of the social network, the application discards the traditional static network topology structure, adopts a dynamically adaptive network topology model, can accurately capture the influence of the heterogeneity of the social network on the rumor propagation path, and thus more accurately reflects the propagation law of the rumor in different social network environments.

[0080] Finally, in the rumor propagation scenario, the influence of the role is analyzed. It should be noted that the higher the average opinion intensity, the more likely the role is to become an "information amplifier" (such as the initial intensity of the traffic police is high and the credibility is high). The propagated rumor is more likely to be accepted by the subsequent nodes (such as the intensity of the medical staff decays slowly). After 10 rounds of simulation in the embodiment, the influence distribution of different roles represented by each agent is as shown in the following table. Figure 3

[0081] By observing Figure 3 It can be seen that the influence of the reporter and the nearby residents and the nearby shop owners is higher. The reason is that these three groups have a greater opportunity to go to the accident site, and the proportion in the network is high, so the driving effect on public opinion is large. The influence of the passerby and the taxi driver is relatively low, because the initial credibility of these two groups is low, and although the initial credibility of the traffic police, medical staff and other groups is high, the speech in the simulation process is more objective and reasonable due to the occupation reason, and the driving of the rumor propagation is weak.

[0082] As shown in the following table, Figure 4 The application provides a rumor simulation analysis system based on a large language model agent, which uses the rumor simulation analysis method based on the large language model agent as described above to perform rumor simulation analysis based on the large language model agent. According to the functions implemented, the rumor simulation analysis system based on the large language model agent 400 can include a role agent construction unit 410, a propagation simulation unit 420 and a role influence determination unit 430. The units of the application can also be referred to as modules, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete a fixed function, which are stored in the memory of the electronic device.

[0083] In the embodiment, the functions of each module / unit are as follows:

[0084] ​The role agent construction unit 410 is configured to determine initialized agent feature information and construct a predetermined number of role agents according to a preset propagation network topology structure; the initialized agent feature information includes to-be-processed speeches and role assignment information of agents corresponding to each network node.

[0085] The propagation simulation unit 420 is configured to obtain a mutated rumor generated by language propagation of the role agent at a current time step by simulating human thinking activities through a large language model based on to-be-processed speeches of the role agent at the current time step; and complete rumor propagation simulation based on the propagation network topology structure and the mutated rumor at each time step.

[0086] The role influence determination unit 430 is configured to determine a propagation probability between each role agent according to a propagation distance and a credibility of each role agent in each round of time step; determine role influence corresponding to each role agent according to the propagation probability and a viewpoint intensity of the role agent after multiple rounds of iteration; and determine a key network node in the propagation network topology structure according to the role influence.

[0087] The rumor simulation analysis system based on the large language model agent provided by the application fully considers subjective judgment, emotional expression and dynamic decision and other factors in the human language interaction process in the aspect of behavior modeling, successfully overcomes the limitations brought by simplifying individuals into homogenized nodes in the traditional model, and makes the behavior of the agent closer to the performance of real humans in rumor propagation. Different network topology structures such as small-world networks, scale-free networks and random networks are supported, and the propagation influence of different users on rumors is quantified through role assignment and credibility calculation. The rumor simulation analysis system based on the large language model agent can realize high-precision simulation of rumor propagation processes in complex social networks and provide a scientific basis for rumor governance.

[0088] More specific implementation manners of the rumor simulation analysis system based on the large language model agent described above can be referred to the description of the embodiments of the rumor simulation analysis method based on the large language model agent, and will not be described here one by one.

[0089] As shown in Figure 5 The application also provides an electronic device 1 for a rumor simulation analysis method based on a large language model agent.

[0090] The electronic device 1 can include a processor 10, a memory 11 and a bus, and can further include a computer program, such as a rumor simulation analysis program based on a large language model agent 12, stored in the memory 11 and executable on the processor 10. The memory 11 can include both an internal storage unit of the rumor simulation analysis system based on the large language model agent and an external storage device. The memory 11 can be used not only to store installed application software and various data, such as the code of the rumor simulation analysis program based on the large language model agent, but also to temporarily store data that has been output or will be output.

[0091] The memory 11 includes at least one type of readable storage medium, including flash memory, mobile hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 can also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 1. Further, the memory 11 can include both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can be used not only to store application software and various data installed on the electronic device 1, such as the code of the rumor simulation analysis program based on the large language model agent, but also to temporarily store data that has been output or will be output.

[0092] The processor 10 can be composed of an integrated circuit in some embodiments, such as a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPU), microprocessors, digital processing chips, graphics processors and combinations of various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, which connects all components of the electronic device through various interfaces and lines, executes or runs programs or modules stored in the memory 11 (such as the rumor simulation analysis program based on the large language model agent, etc.), and calls data stored in the memory 11 to perform various functions and process data of the electronic device 1.

[0093] The bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable connection and communication between the memory 11, the at least one processor 10, etc.

[0094] Figure 5 Only the electronic device with components is shown, and those skilled in the art can understand that, Figure 5 The structure shown does not constitute a limitation on the electronic device 1, and can include fewer or more components than shown, or combine certain components, or different component arrangements.

[0095] For example, although not shown, the electronic device 1 can also include a power supply (such as a battery) to power each component. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management system, so that the power management system can implement functions such as charge management, discharge management, and power consumption management. The power supply can also include one or more DC or AC power sources, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, etc. The electronic device 1 can also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which are not described here.

[0096] Further, the electronic device 1 can also include a network interface, which can optionally include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), and is typically used to establish a communication connection between the electronic device 1 and other electronic devices.

[0097] Optionally, the electronic device 1 can also include a user interface, which can be a display (Display), an input unit (such as a keyboard (Keyboard)), and optionally a standard wired interface, a wireless interface. Optionally, in some embodiments, the display can be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) touch, etc. The display can also be appropriately referred to as a display screen or a display unit, and is used to display information processed in the electronic device 1 and to display a visualized user interface.

[0098] It should be understood that the embodiments are only for illustration and are not limited in the scope of the patent application by this structure.

[0099] The rumor simulation analysis program 12 stored in the memory 11 in the electronic device 1 is a combination of a plurality of instructions, which, when running in the processor 10, can implement: determining initialization agent feature information and constructing a certain number of role agents according to a preset propagation network topology structure; the initialization agent feature information includes unprocessed speeches and role assignment information of agents corresponding to each network node; based on the unprocessed speeches of the role agents at the current time step, a large language model is used to simulate human thinking activities to obtain the mutant rumors generated by the role agents at the current time step for language propagation; based on the propagation network topology structure and the mutant rumors at each time step, rumor propagation simulation is completed; in each round of time step, the propagation probability between each role agent is determined according to the propagation distance and the credibility of each role agent; after multiple iterations, the role influence of each role agent is determined according to the propagation probability and the view intensity of the role agent; and according to the role influence, the key network nodes in the propagation network topology structure are determined.

[0100] Specifically, the specific implementation method of the processor 10 to the above instructions can refer to Figure 1 The description of related steps in the corresponding embodiments is not repeated here. Further, the modules / units integrated by the electronic device 1 can be stored in a computer readable storage medium if they are realized in the form of software function units and sold or used as independent products. The computer readable medium can include any entity or system that can carry the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory).

[0101] The embodiment of the present application also provides a computer readable storage medium, which can be nonvolatile or volatile, and stores a computer program, which is executed by a processor to realize the following: determining initialization agent feature information and constructing a certain number of role agents according to a preset propagation network topology; wherein the initialization agent feature information comprises to-be-processed speeches and role allocation information of agents corresponding to each network node; obtaining a variation rumor generated by the role agents in a current time step for language propagation through a large language model to simulate human thinking activities based on to-be-processed speeches of the role agents in the current time step; completing rumor propagation simulation based on the propagation network topology and the variation rumor in each time step; determining a propagation probability between the role agents according to a propagation distance and a credibility of each role agent in each time step; and determining role influence of each role agent corresponding to the role agent according to the propagation probability and a viewpoint intensity of the role agent after multiple rounds of iteration, and determining a key network node in the propagation network topology according to the role influence.

[0102] Specifically, the computer program is executed by the processor to specifically realize the method, and the description of the related steps in the rumor simulation analysis method based on the large language model agent in the embodiment can be referred to, and details are not described herein.

[0103] In the several embodiments of the present application, it should be understood that the disclosed device, system and method can be implemented in other ways. For example, the above-described system embodiments are merely schematic, for example, the division of the modules is only a logical function division, and another division mode can be used in actual implementation.

[0104] The modules described as separate components can or can not be physically separated, and the components displayed as modules can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment.

[0105] In addition, each functional module in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically independently, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of hardware plus software functional modules.

[0106] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0107] Therefore, embodiments should be considered in all respects as illustrative and not restrictive, the scope of the application being defined by the appended claims rather than the description given above, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. No reference signs in the claims shall be construed as limiting the scope of the claims.

[0108] Further, it is clear that the word "comprising" is not used exclusively in the sense of "consisting only of", that the word "comprising" does not exclude other elements or steps, and that the singular is not excluded in the sense of the plural. Multiple units or systems recited in a system claim can also be implemented by one unit or system by means of software or hardware.

[0109] However, those skilled in the art should understand that, for the rumor simulation analysis method based on the large language model agent and the rumor simulation analysis system based on the large language model agent proposed in the present application, various improvements can be made without departing from the content of the present application. Therefore, the protection scope of the present application should be determined by the contents of the appended claims.

Claims

1. A rumor simulation analysis method based on a large language model intelligent agent, applied to electronic devices, characterized in that, include: Based on the preset propagation network topology, the initial agent feature information is determined and a set number of role agents are constructed; wherein, the initial agent feature information includes the speech to be processed and the role allocation information of the agent corresponding to each network node; the roles include ordinary users, opinion leaders and rumor creators; when there are two or more propagation network topologies, the propagation network topologies are dynamically switched. Based on the speech to be processed by the agent at the current time step, the variant rumors generated by the agent at the current time step through language propagation are obtained by simulating human thought processes using a large language model; the rumor propagation simulation is completed based on the propagation network topology and the variant rumors at each time step. In each time step, the propagation probability between each agent is determined based on the propagation distance and the credibility of each agent. The implementation process of the opinion strength evolution mechanism includes: the propagator sets an initial value for opinion strength; the opinion strength decays due to propagation in subsequent propagations; and in each time step, infected nodes attempt to spread rumors to their neighbors. Based on the propagation probability, combined with the propagation strength, a threshold is set to control whether nodes mutate. The propagation trigger mechanism for agents spreading rumors is that the farther the distance or the lower the credibility of the propagator, the lower the propagation probability. After multiple iterations, the role influence of each role agent is determined based on the propagation probability and the viewpoint strength of the role agent; the key network nodes in the propagation network topology are determined based on the role influence; wherein, the credibility is a quantitative indicator of the role agent's influence on the propagation of rumors, which is jointly determined by the node network attributes and role weight.

2. The rumor simulation analysis method based on a large language model intelligent agent according to claim 1, characterized in that, The propagation network topology includes one or more of the following: small-world network, scale-free network, and random network.

3. The rumor simulation analysis method based on a large language model intelligent agent according to claim 1, characterized in that, Based on the speech to be processed by the agent at the current time step, the mutated rumors generated by the agent's language propagation at the current time step are obtained by simulating human thought processes using a large language model, and are achieved through the following formula: Among them, the large language model is the LLM large language model. As a character intelligent agent, The basic rumor is the unprocessed statement of the agent at the current time step, and the context information includes historical propagation records and node interaction behaviors; the information of the agent includes age, gender, and job status.

4. The rumor simulation analysis method based on a large language model intelligent agent according to claim 1, characterized in that, The propagation probability between each agent is determined based on the propagation distance and the credibility of each agent, using the following formula. in, For the network The role of intelligent agents in the game; For nodes The neighbors; For the role intelligent agent Credibility; This represents the shortest path distance between nodes. This is the spatial attenuation coefficient.

5. The rumor simulation analysis method based on a large language model intelligent agent according to claim 4, characterized in that, The credibility of each agent is determined by the following formula. in, As the character's influence coefficient, For the network Intelligent agents in the context.

6. The rumor simulation analysis method based on a large language model intelligent agent according to claim 1, characterized in that, The viewpoint strength of the agent is achieved through the following formula. Among them, when the role intelligent agent To the neighbors When a rumor is successfully spread, The strength of the viewpoint is expressed as Strength of viewpoints ∈[0,1] represents the degree of subjective acceptance of rumors by the agent; For the role intelligent agent Credibility, Weighting for character acceptance.

7. A rumor simulation analysis system based on a large language model intelligent agent, comprising performing rumor simulation analysis using the rumor simulation analysis method based on a large language model intelligent agent as described in any one of claims 1-6; including: The role-based intelligent agent construction unit is used to determine the initial intelligent agent feature information and construct a set number of role-based intelligent agents according to the preset propagation network topology. The initial intelligent agent feature information includes the speech to be processed and the role allocation information of the intelligent agent corresponding to each network node. The roles include ordinary users, opinion leaders and rumor creators. When there are two or more propagation network topologies, the propagation network topologies are dynamically switched. The propagation simulation unit is used to simulate the speech to be processed by the role agent at the current time step, obtain the mutated rumors generated by the role agent in the current time step through language propagation by simulating human thought activities using a large language model; and complete the rumor propagation simulation based on the propagation network topology and the mutated rumors at each time step. The role influence determination unit is used to determine the propagation probability between each role agent in each time step based on the propagation distance and the credibility of each role agent. The implementation process of the opinion strength evolution mechanism includes: the propagator setting an initial value for opinion strength; the opinion strength decaying during subsequent propagation; and in each time step, infected nodes attempting to spread rumors to their neighbors; controlling node mutation based on the propagation probability, combined with the propagation strength, and setting a threshold. The propagation trigger mechanism for rumor propagation is that the greater the distance or the lower the propagator's credibility, the lower the propagation probability. After multiple iterations, the role influence corresponding to each role agent is determined based on the propagation probability and the opinion strength of the role agent. Key network nodes in the propagation network topology are determined based on the role influence. The credibility is a quantitative indicator of the role agent's influence on rumor propagation, determined by the node's network attributes and role weight.

8. An electronic device, characterized in that, The electronic device includes a memory, a processor, and a rumor simulation analysis program based on a large language model intelligent agent stored in the memory and executable on the processor. When the rumor simulation analysis program based on a large language model intelligent agent is executed by the processor, it implements the rumor simulation analysis method based on a large language model intelligent agent as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the rumor simulation analysis method based on a large language model intelligent agent as described in any one of claims 1 to 6.

Images