Evaluating and Enhancing Large Language Models for Novelty Assessment in Scholarly Publications