A Combined Sentiment Analysis of Twitter for Identifying Users with Depression

In recent years, there has been an increased interest in understanding how sentiment analysis of social media users can be achieved using machine learning techniques. By examining online discourse, an assessment may be made of an individual's mood patterns over a period of time. This study will focus on users of Twitter because of the unique challenges associated with analysing of these short messages known as Tweets. They include words and other forms of media, such as emoticons, images and video. We propose a multimodal technique for evaluating the words, acronyms and emoticons of daily posts by 179 users to identify those experiencing symptoms of depression. Harmonic analysis was employed in a hybrid approach of word based and learning based linguistic content analysis. A support vector machine was then used to predict the valence of a Tweet to be either ‘Negative’, ‘Neutral’ or ‘Positive’ and compared this with other well known sentiment analysis approaches. Outcomes were aggregated daily to compute a user's mood for 30 days. Experimental findings demonstrated that this technique has a better accuracy of predicting mood than current state-of-the-art methods.