engleski

Social Link Prediction from Homophilous Relationships Using Probabilistic Logic

The importance of homophily in network formation is a well-known fact, but link prediction models that use homophilous relations among nodes are rare compared to ones based on the network topology. This paper investigates the possibilities of using available homophilous relations to build a link prediction model using probabilistic logic techniques. A subset of egocentric networks of 10.021 actors was collected, together with basic socio-demographic data from which homophilous relations are deduced. The research demonstrates that the following relations are significant and useful when building a prediction model: physical proximity of actors, having common surname, being coworker, having same place of birth and generational distance. For every relation or combination, the probability function is estimated based on the training subset of the collected dataset. All homophilous relations are combined into an integrative model using a probabilistic logic framework. The predictive power of the model was evaluated and its performance was assessed as excellent.

Probabilitic Logic; Link Prediction

nije evidentirano