Tracking Context Switches in Text Documents and Its Application to Sentiment Analysis

The emergence of social media platforms has provided voluminous amounts of opinionated data. As businesses look to understand and exploit these variegated means of online expression, it has become necessary to automate the process of detecting the panorama of underlying themes. A challenging task is to detect the gradual transitions of context that occur within a text. In this work, we introduce an innovative approach for analyzing textual content that conveys multiple themes. It focuses on efficiently segregating the context switches in text, and then accurately mining the different opinions present. We utilize three categories of features namely positional, lexical-semantic and sentiment polarity for theme co-referent text segmentation within a document. We obtain a Beta Cubed F1 score of 0.7208 for theme co-referent text segmentation of fifty text documents obtained from the question-answer website Quora. The topics corresponding to the theme of each of these segments are obtained by using a noun phrase extractor. We further present an application of the proposed approach to improve the efficiency of Sentiment Analysis. Experimental results demonstrate the proficiency of the proposed scheme to segregate textual content by themes, identify meaningful topics inherent in the themes, compute the polarity, and also suggest the applicability of the method to query-based retrieval systems.