Tagging Opinion Phrases and their Targets in User Generated Textual Reviews
N. K. Gupta · 2013
We discuss a tagging scheme to tag data for training information extraction models which can extract the features of a product/service and opinions about them from textual reviews, and which can be used across different domains with minimal adaptation. A simple tagging scheme results in a large number of domain dependent opinion phrases and impedes the usefulness of the trained models across domains. We show that by using minor modifications to this simple tagging scheme the number of domain dependent opinion phrases are reduced from 36 % to 17%, which leads to models more useful across domains. 1