Casevo - An open-source social communication simulation system developed by the Communication University of China

What is Casevo?

Casevo (Cognitive Agents and Social Evolution Simulator) is an open-source social communication simulation system jointly developed by the School of Data Science and Intelligent Media Communication and the State Key Laboratory of Media Convergence and Communication at the Communication University of China. Combining large language models and multi-agent technology, it simulates human cognition, decision-making, and social interaction to understand and predict social communication phenomena. Casevo uses a modular architecture to support a complete simulation framework, from scenario setting to complex social network modeling, and uses a round-up update mechanism to advance the simulation process. The Casevo system has broad application potential and is suitable for fields such as news communication, social computing, and public policy, helping researchers with theory building, hypothesis testing, and strategy optimization, thus promoting the development of the "AI For Social Science" research paradigm.

Casevo's main functions

• Social interaction simulationSimulates complex social interaction processes, such as election debates and public opinion dissemination, to reproduce the interaction and information exchange between individuals.
• Dynamic social network modelingIt supports the construction and dynamic adjustment of social network structures, reflecting the evolution of individual relationships, and is applicable to research scenarios such as information dissemination and social influence.
• Individual Behavior and Decision SimulationBased on mechanisms such as Chain-of-Track (CoT) and Retrieval-Enhanced Generation (RAG), the agent can perform multi-step reasoning and decision-making based on historical memory, simulating the behavioral choices of individuals in complex situations.
• Massive parallel processingIt features a parallel optimization module, which efficiently handles the parallel behavior and decision-making of large-scale agents, improving the efficiency and performance of simulation.
• Flexible scene customizationUsers can customize simulation scenarios according to their needs, including agent customization, network topology, and external event intervention, to meet diverse research requirements.

Casevo's technical principles

• Discrete event simulationBased on a discrete event simulation mechanism, this method uses a polling update approach to manage agent behavior and event scheduling, ensuring the synchronization of system behavior and the orderly arrangement of events. It is suitable for progressively evolving social dynamic simulation scenarios.
• Large Language Models (LLMs) IntegrationIntegration with LLMs allows agents to generate natural language text, make human-like decisions and communicate, enhancing the realism and complexity of the simulation.
• Chain Thinking (CoT)The CoT mechanism supports agents in performing multi-step reasoning, considering multiple factors before making decisions, and simulating strategic behaviors such as planning, negotiation, and alliance building.
• Search Enhancement Generation (RAG)The RAG memory system enables agents to recall past interactions and decisions, generating more detailed and context-sensitive behaviors based on historical data, simulating long-term strategic thinking and memory-dependent decision-making in humans.
• Modular architectureCasevo employs a modular design, dividing functions such as model setup, agent behavior definition, parallel optimization, and network management into independent modules. This enables the system to achieve high flexibility and scalability, facilitating customization and expansion according to specific needs.

Casevo's project address

• GitHub repository:https://github.com/rgCASS/casevo
• arXiv technical paper:https://arxiv.org/pdf/2412.19498

Casevo's application scenarios

• Social science researchSimulate the election process, analyze changes in voter preferences, predict election results, and provide data support for election research.
• Behavioral predictionSimulates consumer purchasing decisions, analyzes influencing factors, and helps businesses develop marketing strategies to improve market competitiveness.
• educateTo build a virtual chemistry laboratory, simulate chemical reactions and experimental operations, assist in chemistry teaching, and improve learning effectiveness and safety.
• Entertainment and Game DevelopmentDesign NPCs with complex interactions so that they can react in diverse ways based on player behavior, thereby enhancing the game's playability and immersion.
• Emergency ManagementSimulate emergency response to disasters such as earthquakes, analyze the effectiveness of emergency strategies, improve emergency response efficiency, and reduce disaster losses.