Towards a Principled Multi-Agent Approach to Simulating Social Platforms

Stephanie Anneris Malvicini, Gerardo Simari, Maria Vanina Martinez

Keywords: Hybrid Agent Architectures, Social Platform Simulation

Abstract:

Simulating social platforms has become increasingly important, as access to real-world data is becoming more limited, and controlled experimentation on real platforms is often impractical, expensive, or simply infeasible.
Traditional agent-based models (ABM) and Belief-Desire-Intention (BDI) architectures offer transparency and a theoretical grounding; however, they are not especially suited to operate in environments in which rich natural language is used.
Recent Large Language Model (LLM)–driven agentic approaches enable natural language interaction, but often at the expense of interpretability, control, and reproducibility.
In this work, we propose a general Social Platform Simulator leveraging a Social Agent model that supports a spectrum of agent implementations, ranging from fully LLM-based to structured symbolic architectures.
The framework explicitly models key components of social platforms, including the social network, content management, filtering algorithms, and shared knowledge, enabling modular, controlled, and extensible simulations of social media–like environments.
We demonstrate the expressiveness of the framework by analyzing both a minimalist LLM-based instantiation and a hybrid architecture that allows LLMs to be part of all internal processes of a classical BDI framework, which we refer to as ``BDI+'', and represents a balance between the control and transparency of BDI architectures and the expressive power of LLMs. The proposed framework supports reproducible social simulations and the generation of synthetic datasets that can be used for benchmarking in social data ecosystems.
By means of a case study on affective polarization, we show how BDI+ facilitates structured interventions, heterogeneous agent roles, and the explicit analysis of internal agent states.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 12