OASIS: Open Agent Social Interaction Simulations with One Million Agents

Overview

OASIS is a scalable open‑source social‑media simulator that combines large language models with rule‑based agents to realistically emulate the behavior of up to one million users on platforms such as Twitter and Reddit. The simulator is designed to advance research on complex social phenomena (e.g., information diffusion, group polarization, herd behavior) and provides a versatile tool for exploring diverse social dynamics and user interactions in digital environments.

Key Features

• Scalability: Supports simulations with up to one million agents, enabling studies at a scale comparable to real‑world platforms.
• Dynamic Environment: Adapts to real‑time changes in social networks and content, offering flexibility for realistic simulations.
• Diverse Action Space: Agents can perform 21 actions (e.g., follow, comment, retweet), allowing rich multi‑faceted interactions.
• Integrated Recommendation System: Includes interest‑based and popularity‑based recommendation algorithms to mimic how users discover and engage with content on social media platforms.