A Self Training Semi-Supervised Truncated Kernel Projection Machine for Link Prediction

Abstract:

With the large amount of complex network data becoming available in the web, link prediction has become a popular research field of data mining. We focus on the link prediction task which can be formulated as a binary classification problem in social network. To treat this problem, a sparse semi-supervised classification algorithm called Self Training Semi-supervised Truncated Kernel Projection Machine (STKPM), based on empirical feature selection, is proposed for link prediction. Experimental results show that the proposed algorithm outperformed several outstanding learning algorithms with smaller test errors and more stability.

• Periodical: Advanced Materials Research (Volume 580)
• Pages: 369-373
• DOI: https://doi.org/10.4028/www.scientific.net/AMR.580.369
• Online since: October 2012
• Authors: Liang Huang, Rui Xuan Li, Kun Mei Wen, Xi Wu Gu
• Keywords: Link Prediction, Self Training Semi-Supervised Truancated Kernel Projection Machine, Social Network

References

[1] G. Blanchard, P. Massart, R. Vert, L. Zwald : Proc. NIPS (2004), pp.1649-1656.

[2] X. Guo, D.X. Zhou : Appl. Comput. Harmon. Anal. 32 ( 2012).

[3] S. Smale, D. X. Zhou : Appl. Comput. Harmon. Anal. 19 (2005).

[4] M.E.J. Newman : Physical Review Letters E. 64 (2001).

[5] D. Liben-Nowelly, J. Kleinberg: CIKM'04 (2004), pp.556-559.

[6] H.H. Song, T.W. Cho, V. Dave, Y. Zhang, L.L. Qiu: IMC'09 (2009).

[7] H. Kashima, N. Abe : ICDM'06 (2006).

[8] Y.Q. Li, C.T. Guan, H.Q. Li, Z.Y. Chin : Pattern Recogonition Letters, 29 (2005).

[9] Information on http: /www. fml. tuebingen. mpg. de/Members/raetsch/benchmark.