Modeling and detecting change in user behavior through his social media posting using cluster analysis

Abstract

According to World Health Organization, one of the greatest health hazards of 21st century is mental disorder. Unlike any physical illness, mental illness is not that apparent to be recognized at early stages. Also, especially in India, patients do not come forward to seek help because of the social taboo or inferiority that is associated with these diseases. As per World Health Organization, 11 percent of the world's population suffer from mental disorders but only 1 percent of the population form the community of experts who can treat them, leading to the lack of sufficient man power to treat mental illness, and thus the treatments being very expensive. This calls for a strong need for a technique to automatically identify non-normal behavior of a person, which would serve as an indicator for early detection of mental illness. According to the census report of India 2011, the citizens from the age group of 18 to 30 is the majority having mental health problems, which is incidentally the age group which is very active on social networking sites. Online social networks serve as a valuable source of information about people through their published interests, attributes and social interactions and also a true mirror of their behavior. People who don't share their issues with friends and families then find a place on social media and open up their feelings there. Majority of work in current relevant literature talks about classifying tweets based on the sentiments. Whereas, our approach is to analy se the tweets of a person over a period of time to track the change in his behaviour if any. We have developed a new unsupervised technique for detecting change in behaviour of a person using the difference in the structural and behavioural feature vector and defining a threshold using iterative clustering. With a synthesized data, our models lead us to 92% accuracy and a precision 92 % and recall of 90 %.