Researchers have developed UNRESTSENT200K, a new dataset of approximately 200,000 Bangla comments from Facebook and YouTube related to the July-August 2024 Bangladesh uprising. This dataset is designed to evaluate crisis sentiment analysis in low-resource languages, covering five distinct phases of the event, including an internet blackout and a subsequent flood crisis. The study benchmarks various models, including fine-tuned encoders and large language models (LLMs) with and without LoRA tuning, finding that context from parent posts improves performance, while temporal shifts across crisis phases significantly degrade it. Despite strong performance from LLMs, challenges remain with sarcasm and implicit political references. AI
IMPACT This research provides a benchmark for understanding LLM capabilities in low-resource crisis sentiment analysis, crucial for monitoring social and political events.
RANK_REASON The cluster describes a new academic paper introducing a dataset and benchmarking LLMs for a specific task. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
