Discriminative Features Fusion with BERT for Social Sentiment Analysis

The need for sentiment analysis in social networks is increasing. In recent years, many studies have shifted from author sentiment research to reader sentiment research. However, the use of words that hinders sentiment analysis is very diverse. In this paper, we provide a model that combines the latest and most recent contextual text embedding technology and feature selection to more accurately detect the emotional intent of an article. We named it DF2BERT (Discriminative Features Fusion with Bert), and extensively applied datasets in different languages and different text classification tasks to validate our method, and compared it with several well-known approaches. Experimental results show that our model can effectively predict sentiment behind the text which outperform comparisons.