Two-Phase Machine Learning Approach for Extractive Single Document Summarization

摘要

Summarization is the process of condensing the information with minimal loss of information and its importance has grown with the increased information availability. This paper explores two-phase machine learning approach for Single document summarization where in Phase I sentence selection is done using Bayesian Network and in Phase II coherence summary is generated using Affinity Propagation. Phase I model has explored variety of features as incidence features, scoring features including psycholinguistic features. The proposed model is built on the benchmark dataset DUC 2001 and tested on DUC 2002 dataset using Rouge scores. The proposed two-phase machine learning approach gives significantly improved results compared to many existing baseline models.