Predicting depression levels using social media posts

Abstract

Depression is a frequent and dangerous medical ailment that affects how you feel, think, and act. It has a negative effect on people’s feelings, thoughts, and actions. Severe depression expresses itself in a variety of ways, including insomnia, anger, hopelessness, and even suicide attempts. There will be many reasons for depression like abuse, certain treatments, genes, deaths, or losses etc. The COVID-19 pandemic is one of the major health crises that has changed the life of millions of people globally. The COVID – 19 pandemics has affected 220 countries around the world and have reported 190,347,496 confirmed cases. Also, due to COVID – 19, people's mental health was reduced significantly, and this was found by one of the research groups at Boston University. Loneliness, worry, economic instability, and the daily hearing of bad news caused by the coronavirus epidemic are all taking a toll on people's mental health and may be feeling depressed or anxious. Also, in the last few decades, social media usage was increased drastically, and there tends to share people’s emotions in public through social media. Many types of research revealed that people's mental health can be measured using social media data. This research aims to prove that we can use social media data to predict depression in a pandemic situation like COVID – 19. As well as to prove that there is an increment in the number of depressed people or there is reduction in mental health in a COVID-19 like situation compared to the normal period using the social media data. For this purpose, I have used a combination of social media posts on Facebook and Vkontakte social media. Vkontakte's data was taken from the available online dataset. The Facebook dataset was developed by scraping the Facebook posts from publicly available Facebook pages. In order to train and test the model I have used that dataset. Also, different classification methods like trees, naïve Bayes, neural nets, rules, and logistic regression are used to train the dataset. Moreover, for each classification method has many sub-classifications. Emotion, sentiment, linguistic style, depression language and combination of all features (emotion, sentiment, linguistic style, depression language) are used as the features. The results were interpreted that the higher the number of features used, the higher the F-measure scores in detecting depression users. The highest f – measure was acquired by all features. When considering each individual features linguistic style feature obtains the highest f- measure. Then I have collected another separate Facebook dataset to show that there is a reduction in people's mental health before COVID and in the COVID period. That dataset was scraped by the Facebook user profiles and that user group was selected based on the people who work in Information Technology (IT) industry. The main reason to select people working in the IT industry was due to the new normal of work from home and the working stress they were facing. Moreover, due to those reasons, there is a probability of falling into depression easily. Also, before accessing their Facebook profiles, their consent was taken through a Google Form, and a questionnaire was provided to them to answer. The questionnaire contains screening test questions for the diagnosis of depression. After collecting the dataset, the dataset was used to predict the depression of each respondent using the previously developed model. Based on the results of this dataset it shows a significant difference in mental health before the COVID period and after the COVID period. Result of this dataset reveals that there is reduction in mental health. Also, it reveals that social media data can be used to predict depression in a pandemic situation like COVID – 19. Furthermore, results obtained by the questionnaire and prediction model are similar for 66.67%. That reveals the prediction model gives moreover correct result. This is a data-driven method and predictive approach for the early detection of depression. The main contribution of this research is the explore that the impact of some features to predict depression and to prove that can be used the social media data to predict people's mental health during pandemic situations.