Paper 1: Performance based Comparison between Several Link Prediction Methods on Various Social Networking Datasets (Including Two New Methods)
Abstract: This work extends my previous work on link prediction in Social Networks. In this research, I used two additional datasets, Twitter dataset and Facebook Social Circles Dataset and I ran link prediction methods on these datasets. In my previous work, I performed experiment on the Facebook dataset and proposed two new link prediction methods: Neighbors Connectivity and Common Neighbors of Neighbors (CNN). As in my previous work, in this work, I ran the link prediction methods for several training and testing sizes. Results showed that For Facebook dataset, random had the highest precision, followed by Neighbors Connectivity, then Preferential Attachment, followed by Jaccard/CC, Adamic-Adar, finally CNN. For Twitter dataset, random achieved the highest precision. Preferential Attachment achieved the next highest precision, and Adamic-Adar achieved the least precision. For Facebook Social Circles dataset, Preferential-Attachment achieved the highest precision of 1.08891 followed by random for a training and testing sizes of (1535, 2504) respectively. That is said with slight variation on the orders depending on the training and testing size. The low precision values achieved with Facebook and Twitter datasets are due to the graph types which are sparse as indicated in the datasets websites which confirms Kleinberg finding.
Keywords: Social networks; link prediction; comparison; experiment