Interaction in face-to-face groups displays a number of structural properties, such as tendencies toward mutuality and triadic closure. We examine the extent to which such structural network properties operate within interactions that emerge in the internet social networking site, Facebook. Triads, which represent collections of 3 actors and the ties among them, are important to study because of their link to global structures (e.g., Davis and Leinhardt, 1972, and because of their role in network theories, such as those regarding the role of weak ties in the dissemination of information (Granovetter, 1973). Previous research demonstrated that information contained in triad censuses can be accounted for largely by lower network properties, such as network density and dyadic distributions (Faust, 2006; 2008). In this study, we examine the network ties among wall posts made in Facebook, using data gathered by Viswanath et al. (2009) for over 90K users. We select random nodes from the network, and then establish possible triads within the egocentric, subnetwork of the social graph. Triads consist of ego and two other nodes from friends of their friends. We investigate changes in these triads, and the component dyads, over a two year period, using information derived from successive triad censuses (Holland and Leinhardt, 1970). We distinguish between regular triads and "forbidden triads" (e.g., Johnsen, 1985). Our findings demonstrate that both tendencies towards mutuality, and the problem of low density in a large, internet social network, play important roles in determining the evolution of triads in Facebook interactions. [ABSTRACT FROM AUTHOR]