Combining local and global information for product feature extraction in opinion documents

• Local context information and global context information are defined.
• The studies mostly use only LCI, while GCI is not taken into account. This work combines LCI and GCI.
• It shows that feature extraction can benefit from GCI.
• It integrates LCI and GCI for extraction, which outperforms both LCI and GCI based approaches individually.
• After extraction, ranking features is also desirable as it helps user find important features.

Product feature (feature in brief) extraction is one of important tasks in opinion mining as it enables an opinion mining system to provide feature level opinions. Most existing feature extraction methods use only local context information (LCI) in a clause or a sentence (such as co-occurrence or dependency relation) for extraction. But global context information (GCI) is also helpful. In this paper, we propose a combined approach, which integrates LCI and GCI to extract and rank features based on feature score and frequency. Experimental evaluation shows that the combined approach does a good job. It outperforms the baseline extraction methods individually.