Scientific papers summarization to extract research highlights
Scientific articles repositories provide a large amount of information about specific subjects. In order to facilitate the reader's search, it is common for those repositories to provide a reduced version of articles for free. Examples of that are the research highlights, key sentences that contain the article's main topics. Usually, the article's authors provide those sentences by hand and, because of that, we have the objective of generating automatically those highlights.
We interpreted the task of selecting highlights in a article as a task of automatic text summarization. Automatic summarization has the objective of reducing a text to a short form while maintaining the main points.
This work proposes an approach to represent sentences in a feature space and investigate techniques to identify the most important ones. The techniques considered were clustering, population based optimization, oneclass and multiclass classification.