COMPUTATIONAL SOCIAL SCIENCE
MASIM: Multilingual Agent-Based Simulation for Social Science
MASIM introduces the first multilingual agent-based simulation framework for social science. It allows multi-turn interactions among generative agents with diverse sociolinguistic profiles, supporting global public opinion modeling and media influence analysis. The framework uses the MAPS benchmark, combining survey questions and demographic personas from global population distributions. Experiments confirm MASIM's ability to reproduce sociocultural phenomena and highlight the importance of multilingual simulation for scalable, controlled computational social science.
Executive Impact: MASIM's Enterprise Value
MASIM provides a robust platform for simulating complex social phenomena, offering insights for strategic decision-making and policy forecasting in diverse global contexts.
0
Simulated Agents
0
Supported Languages
0
Avg. RMSE (Native Lang.)
0
Empirical Findings
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Multilingual Agent Interaction
Global Public Opinion Modeling
Media Influence & Information Diffusion
Multilingual Agent Interaction
MASIM is the first framework designed to model multi-turn interactions among generative agents with diverse sociolinguistic personas, allowing agents to converse, influence, and react to one another across languages, addressing the lack of multilingual and cross-lingual interaction modeling in prior studies.
Global Public Opinion Modeling
Simulates how attitudes toward open-domain social science hypotheses evolve across languages and cultures, emulating user behavior on social platforms and aggregating responses via voting as a scalable alternative to traditional global surveys.
Media Influence & Information Diffusion
Incorporates autonomous news agents that dynamically generate content conditioned on institutional profiles and evolving discourse, facilitating controlled studies on information propagation and media effects without a priori manipulation.
15%
Lower RMSE with Native Language Simulation
Simulations conducted in agents' native languages consistently yield better calibration (lower RMSE) and more stable outcomes compared to English-only simulations, highlighting the importance of sociolinguistic context.
Enterprise Process Flow
Agent Initialization (t=0)
→
News/User Self-Intro & Post
→
Recommendation System
→
Agent Reads & Forms Memory
→
Agent Writes New Post
→
User Votes & Updates Attitude
| Feature | MASIM Approach (In-Context Learning) | Alternative Approaches (Fine-tuning/RL) |
|---|---|---|
| Cost-effectiveness |
|
|
| Scalability to hundreds of agents |
|
|
| Faithful real-world representation |
|
|
| Ease of extension |
|
|
Cultural Assimilation: South Korea's Trade Attitudes
South Korea ends up more supportive of free trade. South Korea, initially with a high attitude score (~0.8), shifted to ~0.3 after global communication, demonstrating increased support for international trade. This shift was largely influenced by early exposure to pro-trade content from Brazil and Peru, leading to opinion convergence despite initial disparities.
Calculate Your Potential AI Impact
Estimate the time savings and cost reductions your enterprise could achieve by implementing advanced AI solutions derived from MASIM's principles.
Annual Savings Potential
$0
Hours Reclaimed Annually
0
Your AI Implementation Roadmap
(Phases can range from 2-4 weeks each, depending on enterprise complexity)
Phase 1: Multilingual Agent Persona & Survey (MAPS) Dataset Construction
Integrated diverse sociolinguistic personas from World Values Survey (WVS) with global opinion survey questions from GlobalOpinionQA, establishing a grounded benchmark for cross-cultural simulations.
Phase 2: MASIM Framework Development
Designed and implemented the core simulation engine, including user and news organization agents, memory mechanisms (short-term & long-term), and a multilingual recommendation system for iterative interaction.
Phase 3: Real-World Calibration & Validation
Conducted extensive experiments on calibration against real-world survey data, global sensitivity to external signals, and local consistency of agent behaviors using LLM evaluators to ensure reliability and robustness.
Phase 4: Cross-Cultural Social Science Case Studies
Applied MASIM to investigate phenomena like cultural assimilation and normative diffusion across different countries, uncovering interpretable empirical findings and demonstrating the framework's potential for computational social science research.
Ready to Transform Your Enterprise with AI?
Leverage the power of MASIM's insights. Book a complimentary session with our AI strategy experts to discuss tailor-made solutions for your business challenges.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat