Enhancing relation extraction using multi-task learning with SDP evidence

Hailin Wang, Dan Zhang, Guisong Liu, Li Huang, Ke Qin

Abstract: Highlights•Research highlights 1 (Leveraging syntactic info): Multi-task learning improves RE by predicting SDP token positions and capturing valuable semantic relationships.•Research highlights 2 (Novel supervisory labels): Introducing SDP tokens and SDP matrix labels enhances model focus on critical tokens, improving syntactic knowledge and predictions.•Research highlights 3 (Comparable performance): SGA achieves state-of-the-art results in micro F1-score on SemEval2010-Task8 and KBP37 datasets and surpasses other models on NYT and WebNLG datasets, highlighting its efficacy in incorporating syntactic information for RE tasks.