Predicting Teenager´s Future Stress Level from Micro-Blog

More and more teenagers today are overloaded with adolescent stress from different aspects: academic future, selfcognition, inter-personal, and affection. Long-lasting stress may lead to anxiety, withdrawal, aggression, or poor coping skills such as drug and alcohol use, threatening teenagers´ health and development. Hence, it is important for both teenagers and their guardians/teachers to be aware of the stress in advance, and manage the stress before it becomes severe and starts causing health problems. The current social media micro-blog offers an open channel for us to timely and unobtrusively sense teenager´s stress based on his/her tweeting contents and behaviors. This study describes a framework to further predict teenager´s future adolescent stress level from micro-blog, and discusses how we address the challenges (data incompleteness and multi-faceted prediction) using machine learning and multi-variant time series prediction techniques. Forthcoming events that may possibly influence teenager´s stress levels are also incorporated into our prediction method. Our experimental results demonstrate the effectiveness of considering correlated features and event influence in prediction. To the best of our knowledge, this is the first work on predicting teenager´s future stress level via micro-blog.