Link Prediction on Bitcoin OTC Network

Abstract

Link prediction is a common problem in many types of social networks, including small Weighted Signed Networks (WSN) where the edges have positive and negative weights. In this paper, we predict transactions between users in Bitcoin OTC Network, where the links represent the ratings (trust) that the users give to each other after each transaction. Before predicting, we transform the network where we convert negative weights into positive so that the feature scores, calculated by existing algorithms (such as Common Neighbours, Adamic Adar etc.) would improve the models performance in our link prediction problem. We consider two methods that will help us in our link prediction: attributes estimation based on similarity scores link prediction and link prediction as supervised learning problem. The first method can be used more as a way to determine which of the attributes (feature scores) are more important in link prediction. The second method is used for estimating attributes importance, but even more for actual prediction using the calculated feature scores as input to the machine learning and deep learning models. The predicted links can be interpreted as possible transactions between certain users.