Recommended or Not? Give Advice on Online Products

This paper introduces an opinion judgment system that automatically gives advice on whether to recommend this product and furthermore provides corresponding reasons.The core task of the system can be considered as a binary sentence sentiment classification problem. A novel "polarityword-target" related feature extraction method is proposed. An opinion judgment system is then built based on the sentiment of each sentence predicted by a maximum entropy (ME) classifier with the novel features. The experimental results on two domains show that the special feature extraction methods are promising; the opinion judgment system with 91.5% average accuracy is highly effective.