Explorations in Automatic Book Summarization

Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 380–389, Prague, June 2007. c⃝2007 Association for Computational Linguistics Explorations in Automatic Book Summarization Rada Mihalcea and Hakan Ceylan Department of Computer Science University of North Texas rada@cs.unt.edu, hakan@unt.edu Abstract Most of the text summarization research car- ried out to date has been concerned with the summarization of short documents (e.g., news stories, technical reports), and very lit- tle work if any has been done on the sum- marization of very long documents. In this paper, we try to address this gap and ex- plore the problem of book summarization. We introduce a new data set specifically de- signed for the evaluation of systems for book summarization, and describe summarization techniques that explicitly account for the length of the documents. 1 Introduction Books represent one of the oldest forms of written communication and have been used since thousands of years ago as a means to store and transmit information. Despite this fact, given that a large fraction of the electronic documents available online and elsewhere consist of short texts such as Web pages, news articles, scientific reports, and others, the focus of natural language processing techniques to date has been on the automation of methods tar- geting short documents. We are witnessing however a change: an increasingly larger number of books become available in electronic format, in projects such as Gutenberg (http://www.gutenberg.org), Google Book Search (http://books.google.com), or the Million Books (http://www.archive.org/details/millionbooks).project Similarly, a large number of the books published in recent years are often available – for purchase or through libraries – in electronic format. This means that the need for language processing techniques able to handle very large documents such as books is becoming increasingly important. In this paper, we address the problem of book summarization. While there is a significant body of research that has been carried out on the task of text summarization, most of this work has been concerned with the summarization of short doc- uments, with a particular focus on news stories. However, books are different in both length and genre, and consequently different summarization techniques are required. In fact, the straight-forward application of a current state-of-the-art summariza- tion tool leads to poor results – a mere 0.348 F- measure compared to the baseline of 0.325 (see the following sections for details). This is not surprising since these systems were developed specifically for the summarization of short news documents. The paper makes two contributions. First, we introduce a new data set specifically designed for the evaluation of book summaries. We describe the characteristics of a new benchmark consisting of books with manually constructed summaries, and we calculate and provide lower and upper perfor- mance bounds on this data set. Second, after briefly describing a summarization system that has been successfully used for the summarization of short documents, we show how techniques that take into account the length of the documents can be used to significantly improve …

Content truncated to 500 words