Propagation based product feature extraction

To extract the targets of product opinions, we proposed a novel propagation based approach to extract product features. This approach explored the syntactic relations of opinion words and product features, opinion words and opinion words, feature and features, and defined rules to extract the features. The relations between features and opinion words are described in dependency grammar. This propagation process of feature and opinion word extraction went on until no new features could be extracted. Meanwhile, to remove possible noises introduced in the propagation, three pruning strategies were introduced. Experimental results show that it outperforms two state-of-the-art existing approaches.