A Southeast Asian benchmark for stance and coordination in social media
Multilingual models and an annotated benchmark to tell organized coordination apart from organic similarity in Southeast Asian social media.
About the project
A burst of similar social media posts can look like an organized campaign. In Southeast Asia, this judgment becomes harder when a message moves between languages, and even closely related languages. Many languages in Southeast Asia borrow from each other. This project develops multilingual models to tell these cases apart. Mentees will build an annotated benchmark of public social media conversations from Southeast Asia languages, and analyze their stance, similarity in languages, and also coordinating actors. This would involve multilingual fine tuning for semantic, linguistic and stance similarity analysis. It would also involve a graph network analysis of information diffusion and analysis.
What mentees potentially work on
- Curating, organizing and annotating multi-lingual and code-switched social media posts
- Build cross-language similarity and coordination benchmarks
- Analyze model errors across languages and types of coordinating evidence
Requirements and recommended background
- Comfort with Python
- Comfort and some prior knowledge of graph networks (or network science), linguistics, and reinforcement learning
- If none of these apply but you are interested, you are still welcome to apply!
There are no formal eligibility limits. See who can apply; you can still apply if you don't match every item.