Context Sensitive Sentiment Analysis

Whether it automatically extracts it from annotated corpora, or it accesses it via subjectivity lexicons, sentiment analysis makes use of knowledge. Knowledge, however, is domain dependent, and validity of facts might change along with context switches. In spite of this, existing sentiment analysis systems are rather static, in that they are insensitive to context. We believe that opinion mining techniques could become much more effective, should they incorporate models of context. For this reason, we plan to investigate how sentiment analysis can be made context sensitive and to identify the added value of doing so. Our aim is to explore several dimensions of context (textual, social, temporal, spatial) and to figure out which ones can produce the highest boost in accuracy. To establish if our context aware models bring significant improvements, we plan to determine whether such augmented opinion mining techniques can prove beneficial for social network analysis and recommender systems.