Extending The Performance of Extractive Text Summarization By Ensemble Techniques

Abstract

Text summarization techniques help in automatically shortening the length of text data as well as fluently and accurately passing on the intended message. Extractive Text summarization methods have been well-researched but each such algorithm produces a potentially different summary. There is no standalone “best” algorithm using this method. The algorithm proposed in this paper gives a way to combine the strengths of all the existing algorithms to make text summarization more robust. Though ensemble optimization is a popular technique in machine learning, it has not been tried exhaustively in text summarization. Specifically, soft voting has not been attempted so far. In this paper, we extends the concept of voting classifiers in Machine Learning in text domain and propose a novel optimized ensemble based approach to text summarization. The quality of the summary generated is evaluated based on the Recall Oriented Understudy for Gisting Evaluation (ROUGE) metric. As seen in the results, the proposed model outperforms baseline extractive text summarization models such as textrank, lexrank, LSA, Luhn and KL summarizers by more than 15%.