GUniER: GPT-Enhanced Joint Extraction of Entities and Relations Through Integrated Deep Bidirectional Semantics and Unified Modeling

Dongsheng Wang, Liming Wang, Kun Tang, Qile Bo

Abstract: Entity and relation extraction are key tasks in natural language processing (NLP) and knowledge graph construction. However, existing methods often overlook the complex interactions between entities and their relations. To address this problem, we propose GUniER, which uses a GPT enhancement module and deep bidirectional semantics to optimize the extraction of entities and relations while significantly reducing computational cost. Specifically, the GPT module unifies relation labels with sentence representations, while the BiGRU model captures deep bidirectional semantics to improve the model’s ability to handle complex relations and long-distance dependencies. Additionally, we introduce a two-dimensional interaction table to better model entity-relation interactions, and a cross-entropy loss function to address class imbalance. Experimental results on both open domain and domain-specific datasets show that GUniER achieves a notable improvement in accuracy and reduces computational time compared to state-of-the-art models. Specifically, GUniER achieved a 2.1% improvement in F1 score on the CNShipNet dataset and demonstrated reduced training and inference times compared to previous models, showcasing its efficiency. Our code and models are available at https://anonymous.4open.science/r/UniER.

External IDs:dblp:journals/access/WangWTB24