-
LLMTaxo: Leveraging Large Language Models for Constructing Taxonomy of Factual Claims from Social Media
paper- This project aims to help understand and analyze vast and diverse social media content
- Used ClaimBuster to detect factual claims from social media posts. These claims were encoded using Sentence-BERT, and clustered using HDBSCAN to identify distinct claims by filtering out redundancy
- Leveraged LLMs (e.g., GPT-4o mini and Gemini 2.0 Flash) on TACC Lonestar6 supercomputer to generate multi-granularity topics (i.e., broad, medium, and detailed topics) for claims to automate the taxonomy construction
LLMTaxo Framework -
TrustMap: Mapping Truthfulness Stance of Social Media Posts on Factual Claims for Geographical Analysis
paper | demo | introductory video
- A truthfulness stance map for visualizing stance distributions across various topics and regions at varying levels of granularity
- Users can either select factual claims from user-selected topics or enter their own claims for exploration
- An LLM-based explanation of the truthfulness stance for a claim-tweet pair can be generated by clicking the generation button. A stance distribution report can also be genereated for the selected claim(s)
TrustMap User Interface -
Wildfire: a Twitter social sensing platform for layperson
paper | demo | introductory video
- A user-friendly social sensing platform enabling laypersons to collect and analyze Twitter data without coding skills or a data analytics background
- Features robust data analytics and visualization. Integrated ClaimBuster API, HuggingFace models, and GPT-3.5 to enable user ranking for expanded data collection
Wildfire Data Analytics Page -
Cherry-Picking Detection in News Reporting
paper | annotation interface
- An end-to-end system to detect cherry-picking in news reporting by identifying omitted but contextually important statements using fine-tuned and generative language models
- Designed and developed a data annotation interface to support efficient human labeling of data examples
Cherry-Picking Annotation Interface