Why Link Prediction?
link prediction aims at estimating the likelihood of the existenceof a link between two nodes, based on observed links and the attributes of nodes.Link prediction helps in analyzing networks with missing data, for example, manybiological networks, such as food webs, protein–protein interaction networks andmetabolic networks, whether a link between two nodes exists must bedemonstrated by field and/or laboratorial experiments, which are usually very costly.Our knowledge of these networks is very limited, for example, 80% of themolecular interactions in cells of Yeast [1] and 99.7% of human [2] are stillunknown. Instead of blindly checking all possible interactions, to predictbased on known interactions and focus on those links most likely to exist cansharply reduce the experimental costs if the predictions are accurate enough.In addition, link prediction algorithms can be used to predict the links thatmay appear in the future of evolving networks, such as friendshiprecommendation in online social networks and product recommendation in e-commerce web sites [3]. Further applications of link prediction include theidentification of spurious links, estimation of the competing mechanisms inevolving networks, classification of nodes, and so on.
Thanks to the fundamental theoretical importance in networksciences and the wide applications, the study of link prediction has attractedmuch attention recently. As a supportive evidence, a recent survey hascumulated 515 citations for only about 4 years till Aug. 2015, and a number ofpapers on this topic got published in prestigious journals [5-9].
Link prediction will be still a very hot topic in the near future.The major directions in the next five years include (but not being limited) inthe four following directions. (i) Some fundamental theoretical issues in linkprediction, such as the link predictability of networks. (ii) Link predictionfor different kinds of networks, such as directed networks and weightednetworks, as well as the prediction of direction and weights, in addition tothe existence of a link. (iii) Link prediction algorithms with some novel kindsof data, such as link prediction in social networks with human mobility data.(iv) The connection of link prediction algorithms to some other related topic,such as the understanding of networks evolution and the detection of communitystructure.
[1] H. Yu, et al.,Science 322 (2008) 104.
[2] L. A. N. Amaral, PNAS 105 (2008) 6795.
[3] L. Lü, et al.,Physics Reports 519 (2012) 1.
[4] L. Lü, et al.,Physica A 390 (2011) 1150.
[5] A. Clauset, et al.,Nature 453 (2008) 98.
[6] R. Guimera, et al.,PNAS 106 (2009) 22073.
[7] B. Barzel, et al.,Nature Biotechnology 31 (2013) 720.
[8] P. Bastiaens, et al.,Nature Biotechnology 33 (2015) 336.
[9] L. Lü, et al., PNAS 112 (2015) 2325.
转载本文请联系原作者获取授权，同时请注明本文来自周涛科学网博客。
链接地址：https://wap.sciencenet.cn/blog-3075-912975.html?mobile=1
收藏
