Using Triadic Closure to Predict Social Roles in Online Networks
The study of social roles in online networks is key to properly targeting audiences for advertising campaigns and for recommending new contacts to existing users. However, many current models are skewed by missing, outdated, or non-standard data in users’ profiles. To create a more reliable model, researchers studied five sociological theories in the context of the IT industry’s Linkedin network: homophily, triadic closure, reach, tie strength and trust, and structural holes. It was found that all five theories provided distinct data about the nature of social relations. Thus, the Social Roles and Statuses model was developed which incorporated each of these five theories. This model incorporates two functions, the Node Feature Function and Edge Feature Function. The Node Feature Function uses a set of probabilistic equations to determine the influences of local users on social roles, and the Edge Feature Function incorporates the frequency of connections between users to determine the influence of homophily on the network. It was found that the SRS model most accurately predicted social roles in networks where missing, inaccurate, or atypical data confounded previous models’ accuracy.
In lecture, we have not analyzed networks of this complexity and evaluated the effects of all five of the aforementioned sociology theories. However, we have discussed the fundamental importance of triadic closure within social networks. To quantify the strength of interpersonal relationships in this network, a Local Clustering Coefficient (LCC) was introduced. Interestingly, the triadic closure data showed that individuals have different social functions online than in the working world. The findings showed that triadic closure alone cannot predict the social roles of users, though it provides a lot of intriguing insight to social relations in a network.
This article can be found via the Cornell Library website (“Inferring Social Roles and Statuses in Social Networks”):