Information Enhancement for Joint Extraction of Entity and Relation
Weihu Guo, Shengyang Li, Yunfei Liu, Xipeng Fan, Miaobo Hu · 2023
Joint entity and relation extraction is an important research topic in natural language processing. However, the current work cannot clearly identify the boundaries of entity and relation when solving the triple overlapping problem where triple share the same entity. In addition, in joint entity and relation extraction, the transfer of feature information between entity recognition and relation extraction tasks is weak, which makes the ability of mutual learning between tasks low. This paper proposes a joint entity and relation extraction method based on information enhancement. By establishing the relation between entity and relation and type, the entity and relation feature of the input information is enhanced, and the entity and relation feature is cross-fused during the task progress stage. Two subtasks are carried out. Auxiliary learning and sharing feature information improves the model's ability to identify entity and relation boundaries, and at the same time enhances the model's ability to discriminate entity and relation features. The experimental result on four datasets(i.e., WebNLG, ADE, SciERC, SSUIE)show that our model is highly competitive for overlapped triples, and also verify the effectiveness of information enhancement.