A Fuzzy-Rough Hybrid Approach to Multi-document Extractive Summarization

To generate a multi-document extractive summary,the measurement of sentence relevance is of vital importance.Earlier work, exploring statistics of textual terms at the word(surface) level, faces the problem that the textual terms may be synonymous or ploysemous. This may lead to misrank sentence relevance and may cause redundant information presented in the generated summary. Furthermore, the relationships between concepts expressed by natural languages are inherently fuzzy,which invites the use of fuzzy set and rough set theory. In this paper, we investigate some sentence features from a concept level space and apply a fuzzy-rough hybrid scheme to define a sentence relevance measure. Our approach is applied to the DUC2006 multi-document summarization tasks. The experimental results show our approach is promising and demonstrate the effectiveness of fuzzy set and rough set theory in the application of text summarization.