SwiftRank: An Unsupervised Statistical Approach of Keyword and Salient Sentence Extraction for Individual Documents

摘要

In this paper, we introduce an unsupervised stochastic statistical approach for ranking key-phrases, and identifying the salient sentences within a single document for generic extractive summaries. In particular, we propose a method to perceive the salient information of a text unit which is related to the corresponding title and its leverage depending on the sentence position in a text. Furthermore, the proposed method boosts not only the computational time and speed but it still comprehends the substantial information of a document. The experimental results suggest the proposed method well outperforms the baseline approaches significantly in both keyword extraction and summary sentence extraction.