Automatic abstracting: a review and an empirical evaluation

The abstract is a fundamental tool in information retrieval. As condensed representations, they facilitate conservation of the increasingly precious search time and space of scholars, allowing them to manage more effectively an ever-growing deluge of documentation. Traditionally the product of human intellectual effort, attempts to automate the abstracting process began in 1958. Two identifiable automatic abstracting techniques emerged which reflect differing levels of ambition regarding simulation of the human abstracting process, namely sentence extraction and text summarisation. This research paradigm has recently diversified further, with a cross-fertilisation of methods. Commercial systems are beginning to appear, but automatic abstracting is still mainly confined to an experimental arena. The purpose of this study is firstly to chart the historical development and current state of both manual and automatic abstracting; and secondly, to devise and implement an empirical user-based evaluation to assess the adequacy of automatic abstracts derived from sentence extraction techniques according to a set of utility criteria. [Continues.]