In recent years, feature and opinion classification of Chinese product review is one of the most important research fields in Web data mining. A well-defined specification on data annotation for product named entities, features, opinions and boundaries was proposed and a hybrid tag representation was designed.By integrating linguistic features and POS features into automatic learning, a novel two-level Hierarchical HMMs (HHMMs ) framework was put forward. The HHMM-1 and HHMM-2 algorithms were advanced to identify features and opinion entities automatically. The experimental results showed that two-level HHMM works in a mutual complementation way, which makes the recall and F-score of our approach obviously outstanding.
