Online social networks provide a platform forudsharing information at an unprecedented scale. Users generateudinformation which propagates across the network resulting inudinformation cascades. In this paper, we study the evolution ofudinformation cascades in Twitter using a point process modeludof user activity. We develop several Hawkes process modelsudconsidering various properties including conversational structure,udusers’ connections and general features of users including theudtextual information, and show how they are helpful in modelingudthe social network activity. We consider low-rank embeddingsudof users and user features, and learn the features helpful inudidentifying the influence and susceptibility of users. Evaluationudon Twitter data sets associated with civil unrest shows thatudincorporating richer properties improves the performance inudpredicting future activity of users and memes.
展开▼
