LAL-JER: Label-Aware Learning for Adaptive Joint Entity and Relation Extraction with LLM data augmentation
Meng He, Yunli Bai · 2023
Joint entity and relation extraction has achieved great improvements in Natural Language Processing (NLP) and has been widely applied, such as constructing knowledge graph, query understanding and question answering. Existing methods usually spend long time on fitting the models on certain datasets with given label type, which greatly lacks the ability of generalization. The model cannot make prediction on label types that have not seen in the training set. To address this issue, we propose to use prompt to incorporate the semantic meaning of the label type description. Furthermore, we use large language model to perform data augmentation to improve the robustness of our model during training. Extensive experiments and ablation study on two joint entity and relation extraction validates the effectiveness of our work on that: 1. Our methods achieved states of art performance on joint entity and relation extraction benchmark based on pretrained language model bert. 2. Our methods can help the model make predictions on label type unseen before given prompts.