LAUSR

Network alignment is an emerging research topic with great utility value. Its purpose is to recover the hidden alignment between different networks for subsequent cross-network research. Its applications benefit numerous domains including online social networks, biological researches, and so on. Hence, recently, it draws plenty of research attention. Existing attempts indicate that it is a promising direction. However, due to its recency, there are still several limitations. Existing literature mostly focuses on network alignment in specific scenarios and employs heuristically designed functions; thus, it does not provide general solutions. Besides, they only capture static information but not interactive information that is based not only on the isolated networks but also the alignment itself. In this paper, we propose a bootstrapping framework for network alignment (BoNA). Specifically, the framework starts from an imperfect seed alignment and then iteratively refines it by capturing interactive information until convergence. Within each iteration, we calculate the likelihoods of pairwise alignment by using supervised learning techniques; hence, heuristic functions are no longer required. Furthermore, we extend the framework by starting from multiple seed alignments to combine their advantages. We conduct experiments in both biology and online social networks to demonstrate the generality of our framework. Results indicate that BoNA outperforms existing aligners significantly in all evaluated scenarios in terms of both node correctness and topology conservation.