Network versus content: The effectiveness in identifying opinion leaders in an online social network with empirical evaluation

Abstract

Network studies predict individuals with prominent positions in a social network to be more influential. However, such influence is mostly evaluated by propagation assumption that an individual disseminates information to others, while whether such information has impact on the receivers is not examined. This paper focuses on a detailed scenario of Yelp, an online review platform where users are voted as helpful or not by others. As such, the empirical number of votes can be an alternative ground truth for user influence, to complement the simulation-based propagation ability. We explore whether the network features or the content features of the users are more determinative for identifying opinion leaders. Results suggest that the network features can better predict users’ propagation influence, but fail to predict the empirical collective votes. The content features, on the other hand, though not able to explain the propagation influence, are better indicators for the voted opinion leaders. Via a generative model, we argue two possible mechanisms of users accumulating influence, namely the network contagion which can be well predicted by the network features, and the natural accretion which is determined by the quality of contents created by users. In most real-world systems, both mechanisms may take effect. Our study highlights the necessity of distinguishing such different mechanisms, and selecting appropriate network and content features for prediction accordingly.