EINE:Relation Classification by Enhancing the Impact of Non-Entity words
Xiaojian Li, Rong Gao, Hongchen Qin, Boqing Wen, Ye Tian, Lei Ma · 2022
Information extraction from unstructured text is an important task in natural language processing field, since the performance of relation extraction models directly affects the accuracy of information extraction, especially when joint information extraction is not suitable. However, recent approaches do not pay enough attention to the non-entity words in sentences. Existing works investigate entity words in text sequences, others by examining the semantics of the whole text sequence. In this paper, we propose a relation classification model called EINE, which stands for Enhancing the Impact of Non-Entity words, integrating relation information into word-level semantics and reasonably enhancing the impact of non-entity words. Experimental results show that relation classification not only relies on the representation of entity words in sentences, but also needs to use the non-entity words to avoid over-relying on the entity words. Meanwhile, the integration of relation information significantly improves the classification effect. Finally, our model obtains superior results on two datasets, SemEval-2010 Task 8 and WebNLG.