Cross-network Association Based On Local Neighborhood Reconstruction

We introduce a novel cross-network collaborative problem in this work:given YouTube videos,to find optimal Twitter followees that can maximize the video promotion on Twit-ter.Since YouTube videos and Twitter followees distribute on heterogeneous spaces,we intro-duce a Cross-network Association based on local manifold reconstruction,match the YouTube vedio and candidate twitter followee in local manifold space.This article mainly contains three phases.The first stage is the topic modeling for YouTube and users topic modeling for Twitter.The second stage:Assuming YouTube vedio topic and Twitter user topic share similar intrinsic geometries.We can get the user topic distribution in twitter obtained by reconstructing the weight of the nearest neighbor in YouTube's local manifold.Applying the canonical correlation analysis(CCA)to maximize the topic correlation between in YouTube vedio and Twitter users,we make the global users'topic reconstruction using coherent and residual compensation in coherent subspace,then,we combine the two results.The properness by experiments can show that the reconstruction based on local neighbor is a more accurate association function.The third stage:given a YouTube video,we can get a followee.We use normalized Discounted Cumulative Gainas athe evaluation metic,which turns out the followee is better.Finally,we can recommend the vedio for the followee.