Project
Multi-Agent Rumor Intervention Simulation
A social-computing competition project that modeled rumor spread and reduced the propagation index through measurable intervention.
Python · Qwen2.5-14B-Instruct · Multi-Agent Simulation · NetworkX · Social Computing
Product Snapshot
- Role
- Team lead and core engineer
- Focus
- Role agents, propagation and resistance, centrality and community detection, plus deletion, persuasion, edge-breaking, and banning strategies.
- Validation
- Team project records report a 0.3 reduction in the rumor spread index under combined intervention.
- Public Proof
- BDSC2025 materials document 200 nodes, 2,000 edges, and the intervention metric; it was not independently rerun for this portfolio.
Next Step
Project
This was the project I led for the BDSC2025 Social Computing Innovation Competition.
We treated rumor spread as a multi-agent social system rather than a single classification problem. The interesting part was not making each agent look smart, but shaping meaningful collective behavior across a network.
My role
I handled the heaviest engineering and coordination work:
- set up the server and runtime
- deployed
Qwen2.5-14B-Instruct - translated the task brief into an execution plan
- coordinated the team across modules
- organized role agents, propagation and resistance, and the core intervention logic
Result
Project materials document 200 nodes and 2,000 edges. Team records report that the combined intervention reduced the rumor spread index by 0.3.
The original experiment was not rerun for this portfolio, so the number remains explicitly labeled as a team-project result rather than a newly reproduced claim.
Configure the public Giscus environment variables to open discussion on the public site.