Co-attention Based Feature Fusion Network for Spam Review Detection on Douban

Spam review detection is a popular research problem and researchers have paid much attention to online reviews of hotels, restaurants, hospitals, movies, and retail stores. Existing researches divide reviews into true reviews and spam reviews. In recent years, Douban has become one of the most popular Chinese film review platforms and been attacked by various spammers, including both hired spammers and spontaneously organized spammer formed by netizens who deliberately give high or low scores to movies. According to their attitudes towards movies, they can be divided into spam positive reviews and spam negative reviews. Their positions are different, and the consequences to the movies are also different, but they are both harmful and unfair. Therefore, we need to identify them and give movies fair evaluations. Therefore, we divide the reviews into four categories, namely, true positive reviews, spam positive reviews, true negative reviews, and spam negative reviews. We design a co-attention based neural network model, (including user attention and movie attention) which is very innovative because it can effectively fuse user features, movie features, text representation and matadata features, all of which are designed manually. We compare our model with the baselines to prove that our proposed model is more effective. Through ablation study and case study, we prove each of our attention modules is indispensable and attention mechanism can effectively focus on suspicious content.