Supervised Link Weight Prediction Using Node Metadata

Given that node metadata can provide key insights about the relationship between nodes, we investigate if incorporating it as a similarity feature (referred to as metadata similarity) between end nodes of a link can improve the accuracy of weight prediction when using common supervised learning methods. We compare the weight prediction accuracy when metadata similarity is added to a set of baseline topological similarity features to that of using only the topological features. The comparison is performed across four empirical datasets using regression-based and other supervised methods found in the literature. In this preliminary study, we find no significant evidence that metadata similarity improves prediction accuracy in the methods analyzed and within the experimental setup. We encourage further investigation in this research area.