Term-based methods suffer from the problems of polysemy and synonymy, and ontology-based approaches are often restricted in specific application domains. In this DDD107498 paper, we propose a novel method for multi-document summarization based on closed patterns, namely pattern-based summarization (PatSum), which simultaneously considers content coverage and non-redundancy. Compared to term-based methods, PatSum can capture more contextual semantic information than individual terms. Compared to ontology-based methods, PatSum not only can capture the correlations among the words, but also do not require additional information. Experimental results have shown that pattern-based summarization outperforms not only term-based methods and all participating systems on DUC2004 datasets, but also ontology-based method. Multi-document summarization can extract a particular individual\'s opinions in the form of closed patterns, from menstruation individual\'s documents shared in social networks, hence provides a useful tool for further analyzing the individual\'s influence in group activities. In the future, we plan to extend our work to query-based multi-document summarization and multi-document update summarization, with applications to behavior and influence analysis in social networks.
↧