Improving Fake Product Detection with Aspect-Based Sentiment Analysis

Jiaming Li, Yonghao Fu, Daoxing Liu, Ruifeng Xu

Abstract: With the development of e-commerce, the number of counterfeit products is increasing and the rights and interests of customers have been seriously infringed. A product can be evaluated by reviews and ratings objectively. However, the topics of reviews are diverse while customers tend to focus on only a few aspects, and many reviews have wrong scores that are inconsistent with the content. Natural language processing (NLP) is helpful in mining the opinion of reviews automatically. In this paper, the goal is to improve fake product detection through text classification technology. Precisely, we use CNN and LSTM models to judge whether the review is quality related or not, which can remove useless reviews, and aspect-based sentiment analysis with an attention mechanism to determine the sentiment polarity of the concerning aspect to get ratings for different aspects. We experiment on the Self-Annotated datasets and results show that by using text classification technology, the performance of fake product detection can be greatly improved.