Feature Extraction for Product Advertising Reviews Identification in Social Media
Jenq‐Haur Wang, Yu-Hsin Liu · 2015
The social Web has become an important venue for marketing in business. Many vendors hire writers to give product-using experience to promote their products. Promotional texts might give users misleading impressions which are totally far from the truth. However, distinguishing between real and advertising reviews is usually very difficult even for humans. In this paper, we investigate the potential features which might be useful for efficiently identifying advertising reviews (ad reviews) in social media. Based on observations in blog reviews, content, statistical, and social features such as ratio of positive opinion terms, ratio of praise words, and publishing date are extracted to train a SVM classifier for ad review identification. The experimental results on 2,150 reviews showed a good accuracy of 94.55% for classifying cosmetics ad reviews from general articles. Also, experiments on 16,017 cosmetics forum reviews showed the best accuracy of 87.79% in classifying ad reviews from non-ad comments. Our method outperforms conventional text classification using TF-IDF with much higher efficiency. This shows the feasibility of our proposed feature extraction methods in ad review identification.