Link Prediction on Evolving Data using Tensor Factorization Stephan Spiegel 1 , Jan Clausen 1 , Sahin Albayrak 1 , and J´ erˆ ome Kunegis 2 1 DAI-Labor, Technical University Berlin Ernst-Reuter-Platz 7, 10587 Berlin, Germany {stephan.spiegel,jan.clausen,sahin.albayrak}@dai-labor.de http://www.dai-labor.de 2 University of Koblenz-Landau Universit¨ atsstraße 1, 56072 Koblenz, Germany kunegis@uni-koblenz.de http://west.uni-koblenz.de/ Abstract. Within the last few years a lot of research has been done on large social and information networks. One of the principal challenges concerning complex networks is link prediction. Most link prediction algorithms are based on the underlying network structure in terms of traditional graph theory. In order to design efficient algorithms for large scale networks, researchers increasingly adapt methods from advanced matrix and tensor computations. This paper proposes a novel approach of link prediction for complex networks by means of multi-way tensors. In addition to structural data we furthermore consider temporal evolution of a network. Our approach applies the canonical Parafac decomposition to reduce tensor dimension- ality and to retrieve latent trends. For the development and evaluation of our proposed link prediction algo- rithm we employed various popular datasets of online social networks like Facebook and Wikipedia. Our results show significant improvements for evolutionary networks in terms of prediction accuracy measured through mean average precision. Keywords: Link Prediction Algorithm, Temporal Network Analysis, Evolving Data, Multi-way Array, Tensor Factorization 1 Introduction The growing interest in large-scale networks is originated from the increasing number of online social platforms and information networks. Many studies have scrutinized static graph properties of single snapshots, which lacks of information about network evolution [11, 14]. Especially for the challenge of link prediction it is necessary to observe trends within time, which are determined by the ad- dition and deletion of nodes. Most network evolution models are based on the