A comparative study on Chinese open domain suggestion mining

Abstract

As a new research task, suggestion mining increasingly gained attention in recent years. However, it is still open and challenging due to complex semantics, large diversity of domains, and the absence of large labeled and balanced datasets. More importantly, most of the research is focused on English in-domain suggestion mining. But as compared to English, Chinese suggestion has more abundant expression forms, showing many different characteristics, so it is the necessity to carry out suggestions mining research in Chinese environment. In this work, the performances of several classification models for Chinese suggestion mining were compared. Firstly, a Chinese suggestion mining corpus was constructed for open domain, and then trained several models for the suggestion mining, which included both traditional machine learning models (feature engineering-based models) and deep learning models. Our results demonstrated that these models can successfully classify Chinese sentences into two classes: suggestion and non-suggestion. The results of this study can guide future research in Chinese open domain suggestion mining.