Network structure is thought to play a crucial role in a variety of social processes, ranging from the diffusion of information to the spread of social behaviors and norms. As such, understanding the way individuals are interconnected in social networks is of prime significance to predict their collective outcomes. Leveraging data from a knowledge-sharing website, we develop several quantitative measures to quantify a node's social context diversity in directed networks and investigate the role of social context diversity in predicting one's social reputation online. We show that social reputation is tightly controlled by the structural diversity of one's social neighbors, rather than merely neighbor size. To more accurately capture the social context diversity of individuals, we introduce a conceptually new and computationally efficient diversity measure which is obtained via "$k$-clip decomposition" on networks and outperforms some other diversity measures based exclusively on existing weak or strong connectivity between nodes. In the light of "social bridges", we further demonstrate the potential utility of common neighbors to depict the social context diversity of individuals which would go unnoticed otherwise. Our results highlight the importance of social context diversity and have practical implications in the domains of collective intelligence and social influence.