Dropping Fake Favorable Feedback for Better Sentiment Analysis

Abstract

Fake favorable feedback is a big obstacle for customers to get better experiences on shopping websites. As far as we know, the effects of fake favorable feedback on natural language processing models have not been reported. To investigate the distraction of fake feedback, this paper developed a method to resolute fake feedback (RFF). The proposed RFF first analyzes the tokenizes of feedback, and then several short texts are generated to replace some long texts. Experimental results on the raw data and processed data show the effectiveness of our proposal.