Attention economics is a vital component of the Social Web, where the sheer magnitude and rate at which social data is published forces web users to decide on what content to focus their attention on. By predicting popular posts on the Social Web, that contain lengthy discussions and debates, analysts can focus their attention more effectively on content that is deemed more influential. In this paper we present a two-step approach to anticipate discussions in community forums by a) identifying seed posts - i.e., posts that generate discussions, and b) predicting the length of these discussions. We explore the effectiveness of a range of features in anticipating discussions such as user and content features, and present 'focus' features that capture the topical concentration of a user. For identifying seed posts we show that content features are better predictors than user features, while achieving an F1 value of 0.792 when using all features. For predicting discussion activity we find a positive correlation between the focus of the user and discussion volumes, and achieve an nDCG@1 value of 0.89 when predicting using user features.