Opinion-focused Summarization and its Analysis at DUC 2006

In this paper, we present our approach to opinion-focused summarization, its results with the DUC 2006 data, and additional analysis. We extend our approach previously proposed from DUC 2005 to achieve summarization responding multiple questions, assuming that given narrative consists of multiple questions, by segmenting the “narrative” into questions. Our new approach is based on sentence extraction, where sentence type annotation is used for weighting, and frequencies of terms with sentiment polarities are taken into account if question types are appropriate for this. In addition, we selected 15 topics related to opinionfocused summarization and analyzed sentences in original source documents which correspond to model summaries.