Effective Machine Learning Approach to Track Polarity Change in Social Media Data

Abstract

An enormous amount of textual data is generated daily on various social media platforms, which may relate to election campaigns, political discussions, reviewing government policies for the public, launching new utility products, and respective review comments, etc. With these textual data, sentiment analysis helps to understand the behavior and attitudes of users and the public, whether they say good or bad things about the brand, products, and services. However, because we are such emotional creatures, our emotions influence everything we do and serve as the foundation for our future behavior. So, over time, prior positive polarity may change to negative due to some incidents or phenomena. In this paper, we aim to identify the change in polarity over time using a machine learning approach so that respective organizations can analyze possible incidents and circumstances to investigate the reasons behind the change. When the polarity moves from positive to negative, it is essential to identify the reason behind this and rectify or take necessary actions as soon as possible to overcome the negative impact. In other words, respective organizations must work to discover pain points for the polarity change, especially if the change is positive or negative. In this paper, we suggest a framework to identify the change in polarity, and in case a polarity change from positive to negative is detected, the proposed questionnaire helps to get insights into the responsible factors for this change. Results show that the proposed framework, studying daily statistics of the sentiment classification outcomes and the ML model’s interpretable capabilities, gives better accuracy for the change in polarity evaluation.