中央研究院 資訊科學研究所

The rapid explosions of social media and various online communities have significantly enhanced the level, the depth, and the scale of dynamic online interactions between people.  What are the impacts of such interactions regarding their needs for support?  How can we measure influence and identify leaders from these dynamic interactions?  How can we identify temporal causality related to the support function of an online health community? In collaboration with the American Cancer Society (ACS), we have developed and applied methodologies for a large-scale (10 year) analysis of an online forum for cancer survivors regarding these questions.  Using computational text analysis and machine learning methods, we developed models for automated classification of each online post’s sentiment. Based on the analysis of sentiment changes on threaded online discussions, computational models for identifying leaders were developed and compared.  Finally, we present a novel approach to identify the temporal causality for positive sentiment changes of support seekers.  These results not only provided, for the first time, evidence-based insights about the functions of health online communities, but also demonstrated the feasibility and the potential of predictive modeling and causality analysis toward personalized interventions using large-scale longitudinal data that leverages text data in social media.