PhD Proposal : Automatic Domain Adaptive Sentiment Analysis


Wednesday, September 30, 2009, 13:00pm - Wednesday, September 30, 2009, 15:00pm

ITE 325 B

Sentiment analysis is the automatic detection and measurement of opinions and emotions expressed in text segments by machines. However, sentiment is highly domain dependent. This is particularly troubling given the scale and variety of topics seen on the web. Providing sentiment analysis on the web requires more than the standard single domain machine learning approach. In this talk I describe a plan to overcome domain dependence by breaking down documents into three different types of signals. Two of these signals can be processed beforehand to create a general-purpose sentiment model. However, the third type of signal can only be analyzed at query time. To capture the information in this signal we must develop algorithms that can rapidly adapt our general-purpose model to match the query specific domain.

Tim Finin (chair)
Marie desJardins
Tim Oates
James Mayfield (JHU)
Akshay Java (Microsoft)

Tim Finin

OWL Tweet

UMBC ebiquity