Extractive Text Summarization Based on Selectivity Ranking

Extractive summarization of text documents deals with automatic creation of a summary by combining the most salient sentences extracted from the original text document into a more concise form. In this paper, we introduce a novel graph-based method for extractive summarization that transforms a given text into a graph of interconnected sentences and employs computationally efficient selectivity measure to measure the importance of graph nodes. In turn, the text summary is build upon the sentences corresponding to the most important graph nodes. The edges in the graph are based on three measures of similarity between sentences, i.e., Mihalcea’s, Jaccard and Cosine. The first among the three leads to best performance, which also compares favorably to the performance of three alternative summarization methods of TF-IDF, centroid and TextRank. The results confirm that selectivity of nodes in a text graph build using the Mihalcea’s similarity between sentences is a computationally efficient and well performing method for extractive summarization.