Abstract:
Unearthing a set of users’ collective emotional reactions to news or posts in social media has many useful applications and business implications. For instance, when one reads a piece of news on Facebook with dominating “angry” reactions, or another with dominating “love” reactions, she may have a general sense on how social users react to the particular piece. However, such a collective view of emotion is unable to answer the subtle differences that may exist among users. To answer the question “which emotion who feels about what” better, therefore, we formulate the Personalized Social Emotion Mining (PSEM) problem. Solving the PSEM problem is non-trivial in that: (1) the emotional reaction data is in the form of ternary relationship among user-emotion-post, and (2) the results need to be interpretable. Addressing the two challenges, in this paper, we develop an expressive probabilistic generative model, PROMO, and demonstrate its validity through empirical studies.
