Chinese Opinion Target Extraction Based on MRC Framework

In this paper, we formulate the Opinion target extraction (OTE) task as a machine reading comprehension (MRC) task. By formalizing the task of OTE as extracting answer spans to the question “Which opinion target is mentioned in the comment text?”, the prior semantic knowledge of the opinion target is introduced to improve the effect of OTE. We conduct experiments on the method proposed in this paper on pre-trained language models such as Bert and use the Lexical Analysis of Chinese (LAC) lexical analysis tool to correct the results predicted by the model. We conduct experiments on the Chinese evaluation object extraction datasets Baidu, Mafengwo, and Dianping, and the optimal F1 scores were 91.42%, 91.33%, and 93.72%, respectively. The experimental results show that the Pre-trained Model-MRC-LAC Model proposed in this paper can effectively improve the extraction effect of Chinese opinion target extraction.