Automatic text summarization with Maximal Frequent Sequences

Abstract

In the last two decades, an exponential increase in the available electronic information causes a big necessity to quickly understand large volumes of information. It raises the importance of the development of automatic methods for detecting the most relevant content of a document in order to produce a shorter text. Automatic Text Summarization (ats) is an active research area dedicated to generate abstractive and extractive summaries not only for a single document, but also for a collection of documents. Other necessity consists in finding method for ats in a language and domain independent way. In this book we consider extractive text summarization for single document task. We have identified that a typical extractive summarization method consists in four steps. First step is a term selection where one should decide what units will count as individual terms. The process of estimating the usefulness of the individual terms is called term weighting step. The next step denotes as sentence weighting where all the sentences receive some numerical measure according to the usefulness of its terms. Finally, the process of selecting the most relevant sentences calls sentence selection. Different extractive summarization methods can be characterized how they perform these steps. In this book, in the term selection step, we describe how to detect multiword descriptions considering Maximal Frequent Sequences (mfss), which bearing important meaning, while non-maximal frequent sequences (fss), those that are parts of another fs, should not be considered. Our additional motivation was cost vs. benefit considerations: there are too many non-maximal fss while their probability to bear important meaning is lower. In any case, mfss represent all fss in a compact way: all fss can be obtained from all mfss by bursting each mfs into a set of all its subsequences.New methods based on graph algorithms, genetic algorithms, and clustering algorithms which facilitate the text summarization task are presented. We have tested different combinations of term selection, term weighting, sentence weighting and sentence selection options for language-and domain-independent extractive single-document text summarization on a news report collection. We analyzed several options based on mfss, considering them with graph, genetic, and clustering algorithms. We obtained results superior to the existing state-ofthe- art methods. This book is addressed for students and scientists of the area of Computational Linguistics, and also who wants to know recent developments in the area of Automatic Text Generation of Summaries.