Topic Modelling, Classification and Characterization of Critical Information

Abstract

The misinformation, spread on the social media sites such as Twitter, overshadows the utility of such platforms, especially during times of crisis. Fake content is spread to popularise unauthorized treatments or downgrade the efficacy of preventative measures and treatments, resulting in spread of anxiety, depression and chaos amongst society. It is need of the hour, therefore, to apply the technologies like deep learning, natural language programming, and data mining, to develop automated systems that can discern false information from the real information, characterize it for better understanding, and mine it to derive actionable knowledge, that helps to check the spread of misinformation. This work proposes an automated framework that uses a combination of NLP & descriptive and predictive machine learning techniques. COVID-19 related messages on the social media sites are classified as appropriate or misleading using a deep learning model. The classified social media information is characterized based on its sentimental valence, sentimental intensity and emotional acceptance in public, for better understanding. Critical information is retrieved from the authentic information, analysed for better comprehension, and put in an actionable form ready to be leveraged. The popular fake information, such as myths or rumours, also equally important to be identified are retrieved and understood, in order to develop counter-strategies for curbing their spread. Results demonstrate that the framework developed in this paper is able to successfully classify information as fake or real; sentimentally and emotionally characterize it, and churn out novel, actionable and interesting knowledge, crucial for the policymakers, to curb the spread of misinformation.