Contrast Pattern Based Collaborative Behavior Recommendation for Life Improvement

Positive attitudes and happiness have major impacts on human health and in particular recovery from illness. While contributing factors leading human beings to positive emotional states are studied in psychology, the effects of these factors vary and change from one person to another. We propose a behaviour recommendation system that recommends the most effective behaviours leading users with a negative mental state (i.e. unhappiness) to a positive emotional state (i.e., happiness). By leveraging the contrast pattern mining framework, we extract the common contrasting behaviours between happy and unhappy users. These contrast patterns are aligned with user behaviours and habits. We find the personalized behaviour recommendation for those with negative emotional states by placing the problem into the nearest neighborhood collaborative filtering framework. A real dataset of people with heart disease or diabetes is used in our recommendation system. The experiments conducted show that the proposed method can be effective in the health-care domain.