An Ensemble Approach for Extractive Text Summarization

Abstract

Voluminous information spread with an ever-increasing number of articles, links and videos to choose from have made us struggle to make informed decisions quickly. The importance of semantic density has grown considerably. Utilizing machines to help understand and quickly give a pathway to informed decisions has been the emphasis of the computing world for many years now. Automatic text summarization is a common problem where voluminous text needs to be summarized either in an abstractive or an extractive way. So, machine learning algorithms need to be developed that can shorten longer texts automatically and deliver accurate summaries. Many machine learning models have been used by researchers across the globe but results haven’t reached the level of the desired accuracy. This work embodies experimentation based on Logistic Regression, Neural Network, Decision Tree, Random Forest, Naive Bayes, XGBoost, and SVM models, results thus obtained have been compared and finally, an ensemble approach is proposed, which yields better results. Experimentation has been performed using the standard dataset of BBC news. Recall-Oriented Understudy for Gisting Evaluation (ROUGE) metrics is used to validate our research claim which showcased significant improvements utilizing 1-gram, bigrams and longest common subsequence-based statistics.