PalClaim: Evidence-Grounded Arabic Claim Verification with Calibrated Abstention
Supervisor Name
Basar Tahayna
Supervisor Email
bashar.tahayna@najah.edu
University
Najah National University
Research field
Computer Science
Bio
Dr Bashar Tahayna is a researcher and part-time lecturer in Computer Engineering at An-Najah National University, Palestine. His research interests include machine learning, natural language processing, sentiment analysis, and computer vision. His publications investigate context-aware sentiment analysis, automatic annotation of idiomatic expressions, transfer learning, and video classification. His teaching and supervision focus on applied artificial intelligence and data-driven systems. For the proposed PalUROP project, he brings relevant experience in text classification, dataset annotation, and experimental evaluation.
Description
This project investigates whether an Arabic claim-verification system can provide reliable, evidence-supported predictions while recognizing when available evidence is insufficient. Using existing Arabic claim–evidence datasets, including ARAFA, the students will compare classical machine-learning baselines, an Arabic transformer model, and a retrieval-based verification pipeline. A small, manually reviewed evaluation set of Palestine-related public-interest claims will examine performance on locally relevant language and topics. The system will classify claim–evidence relationships as supported, refuted, or not enough information, with an additional confidence-based abstention mechanism that refers uncertain cases to human review. Evaluation will measure classification performance, evidence retrieval, confidence calibration, and the trade-off between coverage and error when abstaining. The contribution will be a reproducible undergraduate research study on the reliability and limitations of evidence-grounded Arabic claim verification.