Experimental study on fuzzy word memberships for multi-document summarization

Fuzzy co-clustering (FCC) is a technique that performs simultaneous fuzzy clustering of objects and features. Recently, several FCC algorithms have been proposed to handle clustering of high-dimensional datasets. The success of these FCC efforts is obvious as it results in both document and word clusters with fuzzy memberships. This paper reports our efforts made on multi-document summarization (MDS) using fuzzy co- clustering approach. The word-memberships are utilized in the MDS, which appear a good alternative interpretation to a document cluster comparing with the conventional frequency- based approaches. We explain the key differences between a summarizer based on memberships approach against the conventional approach and closely investigate on why in principle the fuzzy co-clustering approach has the high potential to outperform the frequency based approaches for MDS. Experiential study on benchmark dataset DUC 2004 shows very promising results, which encourages the further research in the area.