The rapid diffusion of online social networks has shifted a large proportion of human interaction away from traditional means of communication. The purpose of this work is to assess concerns regarding the quality of the relationships supported through online social networks. Properties of interaction are assumed to be a reasonable benchmark for the quality of relationships among people. Therefore, we analyze interactions on Facebook to examine how affective dimensions of messages determine the quantity and kind of interaction they experience. We find that (1) that posting sad messages encourages peers to respond verbally, while (2) articulating positive emotions reduces verbal responses, but leads to more feedback in terms of received 'likes.' Text analysis shows that the sentiment of verbal responses is significantly determined by the sentiment of the message: (3) positive emotions trigger positive feedback, and (4) negative emotions trigger negative responses. We discuss implications of our findings with respect to contemporary theories of close relationships.