Social media allow for an unprecedented amount of interaction between people online. A fundamental aspect of human social behavior, however, is the tendency of people to associate themselves with like-minded individuals, forming homogeneous social circles both online and offline. In this work, we apply a new model that allows us to distinguish between social ties of varying strength, and to observe evidence of homophily with regards to politics, music, health, residential sector & year in college, within the online and offline social network of 74 college students. We present a multiplex network approach to social tie strength, here applied to mobile communication data - calls, text messages, and co-location, allowing us to dimensionally identify relationships by considering the number of communication channels utilized between students. We find that strong social ties are characterized by maximal use of communication channels, while weak ties by minimal use. We are able to identify 75% of close friendships, 90% of weaker ties, and 90% of Facebook friendships as compared to reported ground truth. We then show that stronger ties exhibit greater profile similarity than weaker ones. Apart from high homogeneity in social circles with respect to political and health aspects, we observe strong homophily driven by music, residential sector and year in college. Despite Facebook friendship being highly dependent on residence and year, exposure to less homogeneous content can be found in the online rather than the offline social circles of students, most notably in political and music aspects.