Spotle AI-thon Top 10 Showcase - Analysing Mental Health Of India - Team Jarvis - IIT Roorkee and CMI
0 likes•107 views
The document analyzes the mental health of India during the COVID-19 pandemic using Twitter data, employing natural language processing and sentiment analysis to classify tweets by emotion. The study found that fear was the predominant sentiment expressed, followed by neutral tweets providing information, with joy, anger, and sadness constituting smaller proportions. It concludes that the emotional landscape during the pandemic reflects public responses to the crisis and government actions.
Spotle AI-thon Top 10 Showcase - Analysing Mental Health Of India - Team Jarvis - IIT Roorkee and CMI
- 1. ANALYSIS OF THE MENTAL HEALTH OF INDIA DURING COVID-19 USING TWITTER DATA Team : JARVIS AVIRUP CHAKRABORTY Chennai Mathematical Institute ROHAN GUPTA Indian Institute of Technology Roorkee Spotle AI-Thon 2020
- 2. INTRODUCTION ❖Social Media has become a powerful platform for the public to voice their concerns and beliefs. Among them one such platform is Twitter has been a popular platform for microblogging in the past few years. ❖Due to the COVID-19 pandemic, there has been a wide change in sentiments of various sectors of the Indian public towards the government policies/actions. Studying the sentiment of the people on the epidemic and government decisions is very important as it acts as a sanity check for the effectiveness of the adopted government policies. ❖In this context, Natural Language Processing and Sentiment Analysis are extremely useful in social media monitoring as they allow us to gain an overview of the wider public opinion behind certain topics.
- 3. PROBLEM STATEMENT ❖Use Natural Language Processing and Sentiment Analysis to classify COVID 19 tweets based on emotions. ❖Perform Exploratory Data Analysis and draw Conclusions to analyze the Mental Health of India amidst Covid-19.
- 4. METHODOLOGY ❖ To build a dataset for training the initial model(s), we considered three different datasets (DailyDialog, ISEAR, and Emotion-Stimulus) containing dialogues/situations from individuals and corresponding state of emotions they depict. We chose five classes: Anger, Fear, Joy, Sad, Neutral ❖ We followed mainly three approaches to train the data. The performances of these approaches were compared using their corresponding F1-score which were as follows: ❖ Based on the F1 scores, we chose the CNN model for predictions. Approach F1-score SVM 0.727 LSTM 0.739 CNN+Embedding 0.832 BERT 0.801
- 5. EXPLORATORY DATA ANALYSIS Word Cloud for top prevalent words in India Tweet Count Trend in India
- 7. RESULTS & FINDINGS ❖Model trained was used to make predictions on the given dataset. Some examples of the predictions on the cleaned tweets are :
- 8. ❖The counts for each of the emotion class and their corresponding percentages are tabulated below: Emotion Count Percentage (%) Fear 52383 55.84 Neutral 9687 15.01 Joy 8751 11.63 Anger 8111 9.91 Sadness 5314 7.61
- 9. CONCLUSION ❖Most tweets have been classified in ‘Fear’ class. This can be attributed to the fear in the minds of people when they tweet about the unprecedented rise in COVID- 19 cases. It is quite comprehensible in this situation of pandemic that the dominant emotion is - fear. ❖Next majority tweets belong to ‘Neutral’ class. These tweets serve as a medium of information transfer about recent statistics of deaths, active cases and recoveries in the pandemic. ❖The other three classes, namely ‘Joy’, ‘Anger’ and ‘Sadness’, share almost equal proportions which can respectively be attributed to: ➢increase in recoveries and corona vaccine ➢mismanagement of the government or the insensitive behaviour of citizens towards pandemic ➢some personal loss or deaths of people in general.
- 10. THANK YOU