Optimización multiobjetivo y paralelismo aplicados a la generación de resúmenes extractivos automáticos de múltiples documentos

Resumo

Automatic text summarization is a subject of great interest in many fields of knowledge, mainly as a consequence of the growth of the volume of information in recent years. The goal is to reduce the amount of information about a specific topic to make its reading and understanding easier. Specifically, the generic extractive multi-document text summarization tries to generate a summary that covers the main content of the documents at the same time that reduces the redundancy between its sentences. Due to the multiple objectives involved in the problem, the best way to address it is by using multi-objective optimization approaches, which are able of simultaneously optimizing a set of objective functions. In addition, it is necessary to consider parallelization techniques to reduce the high computing time. This Doctoral Thesis aims to solve different extractive multi-document text summarization problems through the application of multi-objective optimization and parallelism. Various multi-objective search strategies, based on dominance, indicator, and decomposition, have been researched, developed, and applied through several metaheuristic algorithms. Besides, different parallelized designs have been developed to reduce the computing time. The definitions of the problem have also been examined, analyzing different term-weighting schemes, different similarity measures, and even different criteria to be optimized. The generic summarization problem has received great attention, and the experience gained with this kind of problem has been used to address other related problems, such as query-focused summarization, sentiment-oriented summarization, or update summarization. The results obtained have demonstrated the validity and usefulness of the considered approaches.