Opinion analysis is concerned with extracting information about attitudes, beliefs, emotions, opinions, evaluations and sentiment expressed in texts. To date, research in the area of opinion analysis has focused on developing methods for the automatic extraction of opinions and their attributes. While this opinion information is useful, its true potential can be realized only after it is consolidated (summarized) in a meaningful way: the raw information contained in individual opinions is often incomplete and their number is overwhelming.

Until now, the task of domain-independent opinion summarization has received little research attention. We address this void by proposing methods for opinion summarization. Toward that end, we formulate new approaches for the problems of determining what opinions should be attributed to the same source ( source coreference resolution) and whether opinions are on the same topic (topic identification/coreference resolution). Additionally, we introduce novel evaluation metrics for the quantitative evaluation of the quality of complete opinion summaries. Finally, we describe and evaluate OASIS, the first opinion summarization system known to us that produces domain-independent non-extract based summaries. Results for the individual components are encouraging and the overall summaries produced by OASIS outperform a competitive baseline by a large margin when we put more emphasis on computing an aggregate summary during evaluation.