Relation as a Prior: A Novel Paradigm for LLM-based Document-level Relation Extraction

Recent advancements in Large Language Models (LLMs) have showcased their potential in document understanding, yet challenges remain, particularly in Document-level Relation Extraction (DocRE). Traditional methods often falter due to noise from unrelated entity pairs and misclassification of relation labels. To address these issues, the 'Relation as a Prior' (RelPrior) paradigm has been proposed, which strategically utilizes binary relations to filter out irrelevant entities and enhance the accuracy of relation predictions. Extensive experiments conducted on two benchmarks demonstrate that RelPrior not only mitigates the performance gaps identified in LLMs but also achieves state-of-the-art results, surpassing existing methods. This development is crucial as it paves the way for more effective document analysis, potentially transforming applications in various fields such as information retrieval and natural language processing.