Ranking social emotions by learning listwise preference

Emotion modeling has received a great attention in recent years. This paper models the online social emotions that are the online users´ emotional responds when they are exposed to news articles. Specifically, we rank social emotion labels for online documents. Unlike the existing method, referred to as Pair-LR, which learns pairwise preference and adopts binary classification, we address the problem of ranking social emotions by learning listwise preference. In particular, a novel approach, referred to as List-LR, is proposed to learn a ranking model for social emotion labels of online documents by minimizing the listwise loss defined on instances. Empirical experiments show that the proposed approach outperforms Pair-LR and is also competitive to other two start-of-the-art approaches for label ranking.