Modelling Relationship Strength in Online Networks

You are here

The article aims to build an unsupervised model to estimate relationship strength based on profile similarity and interaction activity. The authors do this by formulating a latent variable model aimed at inferring hidden relationship strengths and developing a “coordinate ascent optimization” procedure for inference. Their model is tested and evaluated using real-world Facebook data. The authors’ underlying argument is that previous work in this area has assumed that online relationships are a binary variable: either existent or not, “strong” or “weak”. The authors argue that this approach has been problematic due to relationship strength being an important consideration; an approach that they argue has not been effectively applied in current models.

This article will be a useful resource for researchers and practitioners looking to understand relationship strength between users in online social networks, based on interaction activity (e.g., communication, tagging) and user similarity, and particularly using real-world data from Facebook. The authors formulate a link-based latent variable model along with a coordinate ascent optimization procedure for the inference. When evaluated using real-world data from Facebook, their approach demonstrated that the estimated link weights resulted in higher autocorrelation and lead to improved classification accuracy.

Rongjing Xiang, Jennifer Neville, Monica Rogati