An Exploration of Impact of COVID 19 on mental health -Analysis of tweets using Natural Language Processing techniques

Executive Summery

Novel Corona virus’s spread and its impact on various aspects of national and individual’s well-being has been at the center of lot of discussions across micro-blogging sites and various social media platforms ever since it commenced in December 2019. Users are voicing their opinions on several topics related to covid-19. Social distancing as prescribed by Government and Local Administration We all are aware that the Novel Corona virus has significantly affected our physical health; however the current social distancing norms are taking a toll on the psychological well-being of individuals. The research paper presents a two-phased analysis of most recent 2000 tweets related to mental health pulled out twice over a span of one month on 28 June 2020 and 28 July2020 respectively, thereby analyzing 4000 tweets in total. The second phase analysis was conducted exactly after a gap of one month to validate the results generated by the first analysis. The intention is to analyze to what extent people have discussed about mental health in the past few months based on the information disseminated on Twitter. Data was extracted using Twitter’s search application programming interface (API) and Python’s tweepy library. A predefined keyword like ‘mental health’ was given to find out if Covid-19 emerges as a reason for the same. Several natural language processing (NLP) techniques like tokenization, removing URL and stop words, stemming and lemmatization were used to pre-process the text data and make it ready for analysis. These collected tweets were analyzed using word frequencies of single and double words (unigram, bigram). A very unique feature of this analysis includes a network diagram that shows interconnections between the set of most common words used in to its and the connections (if any) are represented through links. Topic modeling technique in NLP visualizes the top concerns of tweeters through a word cloud. At present we have many methods to do topic modeling. In this paper we are using the Latent Dirichlet Allocation (LDA) method which is a probabilistic approach of modeling given by Prof David H.B in 2003. This model deals with distribution of topics to tweets and allocation of those topics to documents and words to topics. Finally a sentiment analysis is done using text mining techniques to analyze the sentiment of the tweets in the form of positive, negative and neutral.