Extractive Summarization of Text Using Weighted Average of Feature Scores

Abstract

In the era of information overload, need for applications to comb through huge number of documents to extract important information is increasing. This information is helpful in assessing whether or not a document is relevant. Automatic text summarization is one of the solutions to the problem of extracting useful information from huge collection of textual data. A summarizer converts a lengthy document into a short summary by extracting important sentences from it without losing the crucial information. A summarizer can be either abstractive or extractive. An extractive summarizer relies on the statistical features of the input text to create a summary by merely copying the important sentences, whereas an abstractive summarizer tries to understand the context of the document and generates a summary which may contain new sentences not part of the original document. This paper focuses on extractive summarization technique. An approach for generating short and precise summary from a single document using weighted average of feature scores has been proposed. Sentences are ranked based on their scores, and top 40% sentences are selected to form the summary. Experiments were carried out on 250 documents from BBC News summary dataset. The results were compared with existing online summarizers and the proposed summarizer gave better average recall, precision and F-measure values.