Dense Vector Embedding Based Approach to Identify Prominent Disseminators From Twitter Data Amid COVID-19 Outbreak

Abstract

The unprecedented transmission of the Coronavirus COVID-19 across the globe has grown to be a matter of prime concern for researchers, authorities, and healthcare professionals alike. Owing to the unavailability of vaccination, educating people is reckoned to be of utmost importance to mitigate the risk. With a plethora of unstructured data available on social media, it becomes crucial to comprehend information and use it effectively to combat COVID-19. A fine-grained knowledge base could be advantageous in developing a reliable social network for pandemic situations. However, there has been no prior finding related to the identification of disseminators forCOVID-19 and hence, there is a need to build a computationally intelligent system that utilizes the potential of a massive amount of data to disseminate information more effectively. In this work, we gathered Twitter data of 3.2 million unique users, consisting of over 12 million tweets. We divided our work into four parts. Firstly, by employing dense vector embedding, one of the techniques of the neural network, to generate semantically similar keywords. Secondly, we classified the collected data into three awareness categories i.e., information, prevention, and action. Thereafter, we used the statistical physics of complex networks to recognize prominent disseminators w.r.t. the identified categories. Finally, we sub-categorized the prominent disseminators into media, people, and organizations based on their profile information. From the result, we concluded that data generated broadly fall into information and prevention categories, whereas the print media, politicians, and health organizations are the forerunners of the selected prominent disseminators.