A New Memetic Algorithm for Multi-document Summarization Based on CHC Algorithm and Greedy Search

Multi-document summarization has been used for extracting the most relevant sentences from a set of documents, allowing the user to more quickly address the content thereof. This paper addresses the generation of extractive summaries from multiple documents as a binary optimization problem and proposes a method, based on CHC evolutionary algorithm and greedy search, called MA-MultiSumm, in which objective function optimizes the lineal combination of coverage and redundancy factors. MA-MultiSumm was compared with other state-of-the-art methods using ROUGE measures. The results showed that MA-MultiSumm outperforms all methods on the DUC2005 dataset; and on DUC2006 the results are very close to the best method. Furthermore in a unified ranking MA-MultiSumm only was improved on by the DESAMC+DocSum method, which requires as many iterations of the evolutionary process as MA-MultiSumm. The experimental results show that the optimization-based approach for multiple document summarization is truly a promising research direction.