A method and apparatus for implementing multi-round social media deduction based on a large model

Abstract

This specification provides a method and apparatus for implementing multi-round social media simulation based on a large model. The method includes: setting social media simulation parameters according to user input and initializing the social media simulation kernel; performing multi-round multi-dimensional collaborative simulation of social media based on the social media simulation parameters through the social media simulation kernel, synchronizing the current simulation state of the social media simulation kernel through a real-time communication module, and adjusting the social media simulation parameters in real time; and obtaining the recommendation results of the social media simulation kernel.

Description

Technical Field

[0001] This document relates to the field of artificial intelligence technology, and in particular to a method and apparatus for implementing multi-round social media inference based on a large model. Background Technology

[0002] Existing technologies encompass public opinion propagation inference models based on traditional algorithms, social scenario simulation frameworks driven by multi-agent agents, and intelligent interaction generation systems empowered by large models. These systems generally rely on real sensitive user data or simplified role behavior rules, lacking the ability for high-fidelity propagation simulation without sensitive data support, refined multi-role game reconstruction, and dynamic linkage between response strategies and the inference process. On the one hand, traditional rule-based or simple probability model-based ABM tools (such as NetLogo and Mesa) cannot generate or understand natural language content, resulting in agent behavior lacking semantic depth and contextual adaptability. On the other hand, while emerging LLM intelligent agent frameworks (such as Generative Agents) possess language capabilities, they are limited to small-scale closed dialogue scenarios, lacking core social media mechanisms such as attention relationships, information flow, and likes / shares, making it difficult to support research on complex social dynamics.

[0003] Therefore, the existing technology has the following technical problems: current real social platform data is restricted by privacy regulations and APIs, making it difficult to obtain full interaction logs; artificial user experiments are costly and have limited samples; multi-agent simulation systems have dual shortcomings in terms of behavioral realism and platform structure. Summary of the Invention

[0004] The purpose of this invention is to provide a method and apparatus for implementing multi-round social media inference based on a large model, in order to solve the above-mentioned problems in the prior art.

[0005] This invention provides a method for implementing multi-round social media inference based on a large model, including: Set the social media inference parameters based on user input, and initialize the social media inference simulation kernel; Based on the social media simulation parameters, multiple rounds of multi-dimensional collaborative simulation of social media are performed through the social media simulation kernel. The current simulation state of the social media simulation kernel is synchronized through the real-time communication module, and the social media simulation parameters are adjusted in real time. Obtain the recommendation results from the social media inference simulation kernel.

[0006] This invention provides a device for implementing multi-round social media inference based on a large model, comprising: The initialization module is used to set social media inference parameters based on user input and to initialize the social media inference simulation kernel. The simulation module is used to perform multi-round multi-dimensional collaborative simulation simulation of social media based on the social media simulation parameters through the social media simulation kernel, and to synchronize the current simulation state of the social media simulation kernel through the real-time communication module, and to adjust the social media simulation parameters in real time. The output module is used to obtain the recommendation results of the social media inference simulation kernel.

[0007] This invention also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the above-described multi-round social media inference method based on a large model.

[0008] This invention also provides a computer-readable storage medium storing an information transmission implementation program, which, when executed by a processor, implements the steps of the above-described multi-round social media inference implementation method based on a large model.

[0009] This invention constructs a full-link post propagation simulation framework without sensitive data dependencies. It combines a public data verification mechanism of "basic base - round growth" to fundamentally avoid compliance risks. By establishing a dynamic parameter library adapted to the platform and post type, it achieves high-fidelity simulation of propagation fluctuations. This invention proposes a refined multi-role behavior rule system and a dynamic interaction chain generation mechanism for multi-agent collaboration to recreate the complex public opinion game ecosystem. It implements real-time control and data feedback functions based on WebSocket, supports dynamic adjustment of role weights and interaction parameters during the simulation process, and constructs a three-layer framework of "original post feature extraction - model output constraints - quality verification" to ensure consistent responses and scenario adaptability. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart of the multi-round social media inference implementation method based on a large model according to an embodiment of the present invention; Figure 2 This is a detailed flowchart of the multi-round social media inference implementation method based on a large model according to an embodiment of the present invention; Figure 3 This is a flowchart simulating the propagation lifecycle of an embodiment of the present invention; Figure 4 This is a flowchart illustrating the generation of multi-role interactions according to an embodiment of the present invention; Figure 5 This is a flowchart of the intelligent reply generation process according to an embodiment of the present invention; Figure 6 This is a flowchart illustrating the propagation wave simulation of an embodiment of the present invention; Figure 7 This is a schematic diagram of a multi-round social media inference implementation device based on a large model according to an embodiment of the present invention; Figure 8 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0013] Method Implementation Examples According to embodiments of the present invention, a method for implementing multi-round social media inference based on a large model is provided. The present invention aims to address the current limitations of privacy regulations and APIs on real social media platforms, making it difficult to obtain full interaction logs; the high cost and limited sample size of artificial user experiments; and the dual shortcomings of multi-agent simulation systems in terms of behavioral realism and platform structure. Figure 1 This is a flowchart of the multi-round social media inference implementation method based on a large model according to an embodiment of the present invention, such as... Figure 1 As shown, the multi-round social media inference implementation method based on a large model according to an embodiment of the present invention specifically includes: Step S101: Set the social media inference parameters according to the user's input, and initialize the social media inference simulation kernel; specifically including: The system obtains the target post data and the selected dissemination platform input by the user, performs format verification, and confirms that the verification is qualified. The target post data specifically includes: title, content, and type. Obtain the basic simulation parameters set by the user, wherein the basic simulation parameters specifically include: basic user volume, simulated account pool size, and simulation rounds; Obtain the reply strategy parameters set by the user, wherein the reply strategy parameters specifically include: automatic / intelligent reply enabled status, reply trigger threshold, style constraint label, and LLM output length limit.

[0014] A modular virtual social platform simulation kernel is loaded, wherein the virtual social platform simulation kernel specifically includes: a propagation simulation module, a multi-role intelligent agent module, an LLM response generation module, a fluctuation correction module, and a public data verification module; Structured action space parameters define action types, action trigger priorities, and action permissions for each character; Set constrained output protocol parameters, wherein the constrained output protocol parameters specifically include: JSONSchema field specifications and Constrained Decoding constraint threshold; Initialize the LLM parameters.

[0015] Step S102: Based on the social media inference parameters, perform multi-round, multi-dimensional collaborative simulation of social media through the social media inference simulation kernel, and synchronize the current inference state of the social media inference simulation kernel through the real-time communication module, and adjust the social media inference parameters in real time; specifically including: Based on the social media projection parameters, the propagation lifecycle is simulated through the propagation simulation module of the social media projection simulation kernel. The "basic base - round growth" two-factor algorithm is used to generate basic interaction data for each round, and the credibility calibration is completed in conjunction with the public data verification module. Based on the social media inference parameters, multi-role interaction is generated through the multi-role intelligent agent module of the social media inference simulation kernel. A multi-agent priority scheduling algorithm is adopted, and LLM drives each role intelligent agent to make decisions on interactive behavior, generating interactive actions that conform to the role's character setting. Based on the social media inference parameters, the LLM response generation module of the social media inference simulation kernel calls the LLM based on the structured action space and constrained output protocol to generate response content and complete quality verification. Based on the social media projection parameters, the fluctuation correction module of the social media projection simulation kernel integrates statistical algorithms and LLM hotspot identification capabilities to generate fluctuation interaction data that conforms to the real propagation laws.

[0016] The current simulation status is synchronized through the WebSocket real-time communication module, and the social media simulation parameters are adjusted in real time. The current simulation status specifically includes: round progress, interaction data, and response quality. The social media simulation parameters that can be adjusted in real time specifically include: basic simulation, role configuration, response strategy parameters, and manual comment addition parameters.

[0017] Step S103: Obtain the recommendation results from the social media inference simulation kernel. Specifically, this includes: Obtain the recommendation results from the social media simulation kernel, wherein the recommendation results specifically include: propagation trend curve data, role interaction ratio data, response quality score, simulation credibility score, fluctuation fit data, and propagation impact analysis data of manually added comments.

[0018] In summary, this invention introduces a Large Language Model (LLM) as the cognitive and decision-making center of the intelligent agent and constructs a highly modular virtual social platform simulation kernel. Supported by a structured action space (such as post, comment, follow) and constrained output protocols (such as JSON Schema + Constrained Decoding), it solves the compliance risks and insufficient scenario coverage issues of traditional inference relying on real sensitive data. By generating propagation lifecycle data through a dedicated module and linking it with publicly available data for verification, it achieves high-fidelity compliance simulation with a deviation of ≤15%, breaking through the bottleneck of homogenization in multi-role interactions, restoring a complex public opinion game ecosystem, and achieving an interaction content repetition rate of less than 20%. Simultaneously, it overcomes the pain points of large-scale model intelligent reply topic deviation and information distortion. By leveraging dual protocol constraints to anchor the core and style of the original post, it suppresses model "illusions" and achieves dynamic linkage between reply strategies and the inference process, solving the problem of rigid interaction simulation parameters. Through kernel adaptation to differentiated parameters of different platforms and post types, it integrates the advantages of traditional algorithms and large-scale models to simulate propagation fluctuations, improving the consistency between simulation and real-world patterns. Combined with WebSocket... The real-time communication module supports dynamic parameter adjustment and visual monitoring during the simulation process, significantly enhancing the system's practical value and implementation capability.

[0019] The technical solutions of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] This invention focuses on addressing the core pain points of traditional post propagation simulation. It relies on a large language model (LLM)-driven intelligent agent cognitive decision-making center and a highly modular virtual social platform simulation kernel as its core architecture. Supported by a structured action space (post / comment / follow, etc.) and a constrained output protocol (JSON Schema + Constrained Decoding), it achieves highly compliant, highly realistic, and highly controllable post propagation simulation. The technical solution of this invention optimizes the standardized design of style constraint tags (Neutral / Positive / Negative), adds a function to manually add comments containing emojis / images / GIFs during the simulation process, and, combined with WebSocket real-time communication technology, controls the deviation between simulation data and real-world patterns to ≤15%, and the repetition rate of multi-role interaction content to ≤20%, significantly improving the system's practicality and deployment capability. Figure 2As shown, the specific processing includes the following: Step 1, Pre-configuration (Post / Parameter Input): Users configure core parameters through the front-end interface, and the system completes parameter validity verification (value range, format specifications). 1. Upload / Selection of Target Post Data: The user completes the upload / selection of target post data (including title, content, and type), selects the dissemination platform, and all basic parameters are entered and pass the format validation.

[0021] 2. Basic simulation parameters: basic user volume (100-10000, step size 100), simulated account pool size (500-5000, step size 500), simulation rounds (1-10 rounds, each round corresponds to the propagation stage: rounds 1-3 initial diffusion, rounds 4-7 mid-term stabilization, rounds 8-10 late-stage decay).

[0022] 3. Response strategy parameters: Automatic / intelligent response enabled status, response trigger threshold (e.g., if negative comments account for ≥30%, activate the appeasement strategy), style constraint tags (Neutral / Positive / Negative, choose one), LLM output length limit (50-200 characters).

[0023] Step 2, Simulation Kernel Initialization: Load the highly modular virtual social platform simulation kernel, complete the initialization of core modules, deployment of the LLM cognitive decision-making center and configuration of constraint protocols, and ensure that the behaviors generated by LLM are both semantically rich and system reliable, solving the problems of chaotic and uncontrollable behavior generation in existing technologies.

[0024] 1. Modular kernel configuration: Loading propagation simulation module, multi-role intelligent agent module, LLM response generation module, fluctuation correction module, and public data verification module.

[0025] 2. Structured Action Space Parameters: Define action types (post / comment / follow / like / share / collect, etc.), action triggering priority (core interactive actions comment / like have priority 1, auxiliary actions follow / collect have priority 2), and action permissions for each role (e.g., water army has high-frequency comment permission enabled by default, and ordinary users have follow permission enabled randomly).

[0026] 3. Constrained output protocol parameters: JSON Schema field specifications (including 12 required fields such as action type, execution role, timestamp, content summary, and style tags (Neutral / Positive / Negative)), ConstrainedDecoding constraint thresholds (semantic similarity ≥ 0.8, format compliance rate = 100%, information distortion rate = 0); 4. LLM initialization parameters: model selection, inference temperature (balancing stability and diversity).

[0027] Step 3: Multi-dimensional collaborative simulation and deduction.

[0028] 1. Propagation Lifecycle Simulation: The dedicated propagation simulation module in the kernel is activated, and the "baseline size - round growth" two-factor algorithm is run to generate basic interaction data for each round. This is then linked with the public data verification module to complete credibility calibration. Details are as follows: Figure 3 As shown.

[0029] The core parameters of the two-factor model are: base number (= the base number of users configured in advance), round growth factor (marketing posts 0.6-0.8, news posts 0.8-1.0, lifestyle posts 0.4-0.6), and decay coefficient (0.1-0.3, increasing with each round).

[0030] Verification parameter: The deviation threshold (≤15%) between simulation data and publicly available industry data (third-party public opinion reports, platform dissemination index).

[0031] The algorithm involved is the "base number - round growth" two-factor algorithm (formula: interaction amount in the nth round = base number × (growth factor)^(n-1) × (1 - decay coefficient × (n-1) / total number of rounds)).

[0032] 2. Multi-role interaction generation: The multi-role agent module is activated, and the multi-agent priority scheduling algorithm is run. The LLM drives the interaction behavior decisions of each role agent, generating interactive actions that conform to the role's character profile. Specific details are as follows: Figure 4 .

[0033] Character behavior parameters: probability of interaction from ordinary users (30%-50%), probability of interaction from paid commenters (80%-100%), probability of interaction from official accounts (20%-40%), and percentage of comments guided by paid commenters (≥70%).

[0034] Homogenization control parameters: threshold for repetition rate of interactive content for the same character (≤20%), LLM generated content diversity coefficient (0.4-0.7).

[0035] Algorithm involved: Multi-agent priority scheduling algorithm (dynamically allocates interactive resources based on role weight and round requirements).

[0036] 3. Intelligent Response Generation: The LLM response generation module is activated, invoking the LLM based on the structured action space and constrained output protocol to generate response content and complete quality verification. Specific details are as follows: Figure 5 As shown.

[0037] LLM input parameters: original post topic summary, role tags, style tags (Neutral / Positive / Negative), and response trigger scenario (such as negative comments / product inquiries).

[0038] Quality verification parameters: topic deviation threshold (≤10%), semantic similarity threshold (≥0.8), and format compliance rate (100%).

[0039] Algorithm involved: LLM output constraint algorithm based on JSON Schema + Constrained Decoding.

[0040] 4. Propagation Fluctuation Simulation: Activate the fluctuation correction module, integrating traditional statistical algorithms with LLM hotspot identification capabilities to generate fluctuation interaction data that conforms to real propagation patterns. Specific details are as follows... Figure 6 .

[0041] Platform adaptation parameters: Weibo repost rate weight 0.4, Xiaohongshu collection rate weight 0.5, Douyin completion rate weight 0.6.

[0042] Volatility parameters: random volatility factor (based on normal distribution N(0,0.15)) and LLM hotspot correction coefficient (1.2-1.5 during hotspot periods and 1.0 during non-hotspot periods).

[0043] Algorithms involved: Traditional statistical algorithm (basic fluctuation calculation) + LLM hotspot identification and correction algorithm.

[0044] Step 4, Real-time Control: Synchronize the current simulation status (round progress, interaction data, response quality) through the WebSocket real-time communication module.

[0045] 1. Adjustable parameters: Overall basic simulation / role configuration / response strategy parameters, propagation lifecycle simulation deviation >15%, multi-role interaction generation repetition rate >20%, or intelligent response generation response quality is substandard.

[0046] 2. Manually add comment parameters: Manually add comments (supports inputting text + inserting emoticons, uploading JPG / PNG format images, GIF animations, file size ≤ 5MB), and associate post ID (automatically associate with the post of the current projection round).

[0047] Step 5: Conclusion and Result Output.

[0048] Output parameters include: propagation trend curve data, role interaction ratio data, response quality score (≥80 points is acceptable), simulation credibility score (≥85 points is acceptable), fluctuation fit data, and propagation impact analysis data with manually added comments. These parameters provide users with intuitive and usable inference results, support decision-making for propagation strategy optimization, and enhance the practical value of the system.

[0049] In summary, the embodiments of this invention rely on the core architecture of "LLM-driven intelligent agent hub + highly modular simulation kernel + structured action space + constrained output protocol", combined with style tag standardization and flexible interaction design, forming a systematic advantage over existing technologies and accurately solving four core pain points, as follows: 1. Overcome the problem of dependence on sensitive data and achieve compliant high-fidelity simulation: Virtual data is generated by disseminating simulation modules, and parameters are optimized by combining publicly available data to completely get rid of dependence on real sensitive data and avoid privacy compliance risks from the root; at the same time, it can be adapted to different types of posts and platforms, significantly improving the scenario coverage capability, and the deviation between simulation data and real patterns is ≤15%.

[0050] 2. Break through the bottleneck of homogenization of multiple roles and restore the real public opinion ecology: Using LLM as the cognitive decision-making center of intelligent agents, combined with standardized style tags (Neutral / Positive / Negative) to generate differentiated interaction intentions for multiple roles, coupled with multi-agent scheduling algorithm to dynamically allocate resources, combined with repetition rate control mechanism (repetition rate ≤20%), to accurately restore the behavioral differences of different roles and the complex public opinion game process.

[0051] 3. Overcome the pain point of uncontrolled quality of intelligent replies and achieve accurate style adaptation: By using JSON Schema + Constrained Decoding dual-constraint LLM output, and superimposing Neutral / Positive / Negative standardized style tag constraints, the "illusion" of large models is effectively suppressed, ensuring that the reply topic is consistent, the information is true, and the style is accurately matched; at the same time, it supports dynamic linkage between reply strategy and inference process, improving the flexibility of scenario adaptation.

[0052] 4. Solve the problem of rigid simulation and improve realism and interactivity: Integrate traditional algorithms with LLM hotspot identification capabilities, build a platform-post type differentiated parameter library, introduce random fluctuation factors to simulate real propagation fluctuations, and solve the problem of parameter rigidity; add a manual comment addition function (supporting emoticons / images / GIFs), and with WebSocket real-time communication technology, realize dynamic control and visual monitoring of the simulation process, greatly improving operational flexibility and practical value.

[0053] 5. Enhance system scalability and deployment capability: Based on a highly modular simulation kernel, functional modules can be added or removed as needed (such as adding new role types and platform adaptation rules); standardized protocol and parameter design reduce subsequent maintenance costs; real-time interaction and multimedia support functions enhance user experience and help technical solutions be quickly deployed and applied.

[0054] Device Example 1 According to embodiments of the present invention, a device for implementing multi-round social media inference based on a large model is provided. Figure 7 This is a schematic diagram of a multi-round social media inference implementation device based on a large model according to an embodiment of the present invention, such as... Figure 7 As shown, the multi-round social media inference device based on a large model according to an embodiment of the present invention specifically includes: Initialization module 70 is used to set social media inference parameters based on user input and to initialize the social media inference simulation kernel; specifically, it is used for: The system obtains the target post data and the selected dissemination platform input by the user, performs format verification, and confirms that the verification is qualified. The target post data specifically includes: title, content, and type. Obtain the basic simulation parameters set by the user, wherein the basic simulation parameters specifically include: basic user volume, simulated account pool size, and simulation rounds; Obtain the reply strategy parameters set by the user, wherein the reply strategy parameters specifically include: automatic / intelligent reply enabled status, reply trigger threshold, style constraint label, and LLM output length limit; A modular virtual social platform simulation kernel is loaded, wherein the virtual social platform simulation kernel specifically includes: a propagation simulation module, a multi-role intelligent agent module, an LLM response generation module, a fluctuation correction module, and a public data verification module; Structured action space parameters define action types, action trigger priorities, and action permissions for each character; Set constrained output protocol parameters, wherein the constrained output protocol parameters specifically include: JSONSchema field specifications and Constrained Decoding constraint threshold; Initialize the LLM parameters; The simulation module 72 is used to perform multi-round, multi-dimensional collaborative simulation of social media based on the social media simulation parameters through the social media simulation kernel, and to synchronize the current simulation state of the social media simulation kernel through the real-time communication module, and to adjust the social media simulation parameters in real time; specifically, it is used for: Based on the social media projection parameters, the propagation lifecycle is simulated through the propagation simulation module of the social media projection simulation kernel. The "basic base - round growth" two-factor algorithm is used to generate basic interaction data for each round, and the credibility calibration is completed in conjunction with the public data verification module. Based on the social media inference parameters, multi-role interaction is generated through the multi-role intelligent agent module of the social media inference simulation kernel. A multi-agent priority scheduling algorithm is adopted, and LLM drives each role intelligent agent to make decisions on interactive behavior, generating interactive actions that conform to the role's character setting. Based on the social media inference parameters, the LLM response generation module of the social media inference simulation kernel calls the LLM based on the structured action space and constrained output protocol to generate response content and complete quality verification. Based on the social media inference parameters, the fluctuation correction module of the social media inference simulation kernel integrates statistical algorithms and LLM hotspot identification capabilities to generate fluctuation interaction data that conforms to the real propagation law. The current simulation status is synchronized through the WebSocket real-time communication module, and the social media simulation parameters are adjusted in real time. The current simulation status specifically includes: round progress, interaction data, and response quality. The social media simulation parameters that can be adjusted in real time specifically include: basic simulation, role configuration, response strategy parameters, and manual comment addition parameters. Output module 74 is used to obtain the recommendation results of the social media inference simulation kernel. Used for: Obtain the recommendation results from the social media simulation kernel, wherein the recommendation results specifically include: propagation trend curve data, role interaction ratio data, response quality score, simulation credibility score, fluctuation fit data, and propagation impact analysis data of manually added comments.

[0055] The embodiments of the present invention are device embodiments corresponding to the above method embodiments. The specific operation of each module can be understood with reference to the description of the method embodiments, and will not be repeated here.

[0056] Device Example 2 This invention provides an electronic device, such as... Figure 8As shown, it includes: a memory 80, a processor 82, and a computer program stored on the memory 80 and executable on the processor 82, wherein the computer program, when executed by the processor 82, performs the steps as described in the method embodiment.

[0057] Device Example 3 This invention provides a computer-readable storage medium storing an information transmission implementation program, which, when executed by a processor 82, performs the steps described in the method embodiment.

[0058] The computer-readable storage media described in this embodiment include, but are not limited to, ROM, RAM, disk, or optical disk.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for implementing multi-round social media inference based on a large model, characterized in that, include: Set the social media inference parameters based on user input, and initialize the social media inference simulation kernel; Based on the social media simulation parameters, multi-round multi-dimensional collaborative simulation of social media is carried out through the social media simulation kernel, and the current simulation state of the social media simulation kernel is synchronized through the real-time communication module, and the social media simulation parameters are adjusted in real time. Obtain the recommendation results from the social media inference simulation kernel.

2. The method according to claim 1, characterized in that, The social media projection parameters are set based on user input, including: The system obtains the target post data and the selected dissemination platform input by the user, performs format verification, and confirms that the verification is qualified. The target post data specifically includes: title, content, and type. Obtain the basic simulation parameters set by the user, wherein the basic simulation parameters specifically include: basic user volume, simulated account pool size, and simulation rounds; Obtain the reply strategy parameters set by the user, wherein the reply strategy parameters specifically include: automatic / intelligent reply enabled status, reply trigger threshold, style constraint label, and LLM output length limit.

3. The method according to claim 1, characterized in that, The initialization of the social media simulation kernel specifically includes: A modular virtual social platform simulation kernel is loaded, wherein the virtual social platform simulation kernel specifically includes: a propagation simulation module, a multi-role intelligent agent module, an LLM response generation module, a fluctuation correction module, and a public data verification module; Structured action space parameters define action types, action trigger priorities, and action permissions for each character; Set constrained output protocol parameters, wherein the constrained output protocol parameters specifically include: JSON Schema field specifications and Constrained Decoding constraint threshold; Initialize the LLM parameters.

4. The method according to claim 1, characterized in that, Based on the aforementioned social media inference parameters, the multi-round, multi-dimensional collaborative simulation inference of social media through the social media inference simulation kernel specifically includes: Based on the social media projection parameters, the propagation lifecycle is simulated through the propagation simulation module of the social media projection simulation kernel. The "basic base - round growth" two-factor algorithm is used to generate basic interaction data for each round, and the credibility calibration is completed in conjunction with the public data verification module. Based on the social media inference parameters, multi-role interaction is generated through the multi-role intelligent agent module of the social media inference simulation kernel. A multi-agent priority scheduling algorithm is adopted, and LLM drives each role intelligent agent to make decisions on interactive behavior, generating interactive actions that conform to the role's character setting. Based on the social media inference parameters, the LLM response generation module of the social media inference simulation kernel calls the LLM based on the structured action space and constrained output protocol to generate response content and complete quality verification. Based on the social media projection parameters, the fluctuation correction module of the social media projection simulation kernel integrates statistical algorithms and LLM hotspot identification capabilities to generate fluctuation interaction data that conforms to the real propagation laws.

5. The method according to claim 1, characterized in that, The real-time communication module synchronizes the current simulation state of the social media inference kernel and adjusts the social media inference parameters in real time, specifically including: The current simulation status is synchronized through the WebSocket real-time communication module, and the social media simulation parameters are adjusted in real time. The current simulation status specifically includes: round progress, interaction data, and response quality. The social media simulation parameters that can be adjusted in real time specifically include: basic simulation, role configuration, response strategy parameters, and manual comment addition parameters.

6. The method according to claim 1, characterized in that, Obtaining the recommendation results from the social media inference simulation kernel specifically includes: Obtain the recommendation results from the social media simulation kernel, wherein the recommendation results specifically include: propagation trend curve data, role interaction ratio data, response quality score, simulation credibility score, fluctuation fit data, and propagation impact analysis data of manually added comments.

7. A device for implementing multi-round social media deduction based on a large model, characterized in that, include: The initialization module is used to set social media inference parameters based on user input and to initialize the social media inference simulation kernel. The simulation module is used to perform multi-round multi-dimensional collaborative simulation simulation of social media based on the social media simulation parameters through the social media simulation kernel, and to synchronize the current simulation state of the social media simulation kernel through the real-time communication module, and to adjust the social media simulation parameters in real time. The output module is used to obtain the recommendation results of the social media inference simulation kernel.

8. The apparatus according to claim 7, characterized in that, The initialization module is specifically used for: The system obtains the target post data and the selected dissemination platform input by the user, performs format verification, and confirms that the verification is qualified. The target post data specifically includes: title, content, and type. Obtain the basic simulation parameters set by the user, wherein the basic simulation parameters specifically include: basic user volume, simulated account pool size, and simulation rounds; Obtain the reply strategy parameters set by the user, wherein the reply strategy parameters specifically include: automatic / intelligent reply enabled status, reply trigger threshold, style constraint label, and LLM output length limit; A modular virtual social platform simulation kernel is loaded, wherein the virtual social platform simulation kernel specifically includes: a propagation simulation module, a multi-role intelligent agent module, an LLM response generation module, a fluctuation correction module, and a public data verification module; Structured action space parameters define action types, action trigger priorities, and action permissions for each character; Set constrained output protocol parameters, wherein the constrained output protocol parameters specifically include: JSON Schema field specifications and Constrained Decoding constraint threshold; Initialize the LLM parameters; The deduction module is specifically used for: Based on the social media projection parameters, the propagation lifecycle is simulated through the propagation simulation module of the social media projection simulation kernel. The "basic base - round growth" two-factor algorithm is used to generate basic interaction data for each round, and the credibility calibration is completed in conjunction with the public data verification module. Based on the social media inference parameters, multi-role interaction is generated through the multi-role intelligent agent module of the social media inference simulation kernel. A multi-agent priority scheduling algorithm is adopted, and LLM drives each role intelligent agent to make decisions on interactive behavior, generating interactive actions that conform to the role's character setting. Based on the social media inference parameters, the LLM response generation module of the social media inference simulation kernel calls the LLM based on the structured action space and constrained output protocol to generate response content and complete quality verification. Based on the social media inference parameters, the fluctuation correction module of the social media inference simulation kernel integrates statistical algorithms and LLM hotspot identification capabilities to generate fluctuation interaction data that conforms to the real propagation law. The current simulation status is synchronized through the WebSocket real-time communication module, and the social media simulation parameters are adjusted in real time. The current simulation status specifically includes: round progress, interaction data, and response quality. The social media simulation parameters that can be adjusted in real time specifically include: basic simulation, role configuration, response strategy parameters, and manual comment addition parameters. The output module is specifically used for: Obtain the recommendation results from the social media simulation kernel, wherein the recommendation results specifically include: propagation trend curve data, role interaction ratio data, response quality score, simulation credibility score, fluctuation fit data, and propagation impact analysis data of manually added comments.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the multi-round social media deduction implementation method based on a large model as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an implementation program for information transmission, which, when executed by a processor, implements the steps of the multi-round social media inference implementation method based on a large model as described in any one of claims 1 to 6.

Images