Anirban Saha Anik

This paper presents our system for Task 3 of the CLEF 2025 CheckThat! Lab, which focuses on verifying numerical and temporal claims using retrieved evidence. We explore two complementary approaches: zero-shot prompting with instruction-tuned large language models (LLMs) and supervised fine-tuning using parameter-efficient LoRA. To enhance evidence quality, we investigate several selection strategies, including full-document input and top-k sentence filtering using BM25 and MiniLM. Our best-performing model LLaMA fine-tuned with LoRA achieves strong performance on the English validation set. However, a notable drop in the test set highlights a generalization challenge. These findings underscore the importance of evidence granularity and model adaptation for robust numerical fact verification.

People affected by crises increasingly rely on social media platforms for real-time information and assistance. This underscores the need for a robust and reliable approach to provide accurate and timely information to affected individuals. Large Language Models (LLMs), which can understand user queries and generate responses, have the potential to act as assistants in crisis response. We explore various approaches, including instruction prompts, retrieval-augmented generation, and dynamic fusion of LLMs, to generate responses that address the information needs of affected individuals on social media platforms. Experiments demonstrate that the dynamic fusion approach produces better crisis responses across key evaluation dimensions, including professionalism, actionability, empathy, and relevance.

Xiaoying Song, Anirban Saha Anik, Eduardo Blanco, Vanessa Frias-Martinez, Lingzi Hong. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025.

Xiaoying Song, Anirban Saha Anik, Dibakar Barua, Pengcheng Luo, Junhua Ding, Lingzi Hong. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025.