This document discusses predicting answering behavior in online question answering communities. It presents a method to represent individual users' question selection behavior using a matrix structure. It then uses learning to rank models to predict this behavior based on user, question, and thread features. The models achieved a mean reciprocal rank of 0.446, significantly outperforming baselines. Question features were found to be the most predictive, indicating questions from reputable users and with fewer existing answers are more likely to be selected.