Saliency detection in text documents using policy-driven reinforcement learning methodologies

Abstract

As the amount of information grows, it is challenging to find concise information. Thus it is necessary to build a system that could present human quality summaries. Saliency detection is a tool that provides abstracts or keywords of a given document. In this paper, three different approaches have been implemented for saliency detection. In all these three approaches, sentences are represented as a feature vector. In the first approach, features like root words, vocabulary intersections, words, and inclusion of numerical data use. This model is trained by using general Algorithms, Like Porter's Stemmer, Spell check. In the second approach, apart from the features used in the first approach, TF-IDF scores, Mean, Standard Deviation, and a Threshold value of a word is also used as features. In the third approach, Maximal Marginal Relevance (MMR) algorithm is used to generate a summary.