UMBC ebiquity

Domain Independent Sentiment Analysis

Speaker: Justin Martineau

Start: Tuesday, March 01, 2011, 11:00AM

End: Tuesday, March 01, 2011, 12:15PM

Location: ITE 325 - B

Abstract: Domain independent sentiment signals are words or word pairs that are present and have the same sentimental orientation in multiple domains. These words can be easily identified if you have an accurate representation of their in-domain sentimental orientation. If you also have an accurate representation of their sentimental strength then you can use them to correctly classify out of domain documents with reasonable accuracy. In this talk I will present a method to identify domain independent sentiment signal words using multiple domains. My evaluation shows that these results are much better than current techniques and even better than establish bigram baselines.

Tags: text classification, learning, sentiment, natural language processing

Host: Tim Finin