We present a graph-based semi-supervised approach for learning user-preferred travel schedules. Assuming a setting in which a user provides a small number of labeled travel schedules, we classify schedules into desirable and non-desirable. This task is non-trivial since only a small number of labeled points is available. It is further complicated by the fact that each schedule is comprised of multiple components or aspects which are different in nature. For instance in our case arrival times are modeled by probability distributions to account for uncertainty, while other aspects such as waiting times are given by a feature vector. Each aspect can thought of as a different type of observation for the same schedule While existing label propagation approaches can exploit vast amounts of unlabeled data, they cannot handle multi-aspect data. We propose Multi-Aspect Label Propagation (MALP), a novel approach which extends label propagation to handle multiple types of observations.