Emotion analysis of product reviews is a valuable research field, it can help customers and merchants make decisions. Based on various relations of attribute words and emotion words in product review texts, develops eight sets of feature selection rules, and applies SVM algorithm training model to judge the identification of attribute words and emotions words, then analyzes the emotion tendencies of attribute characteristics based on emotion words and negative words. The experimental results show: the proposed SVM-based recognition method achieves good classification effect in the identification of attributes and emotion words.