期刊名称：International Journal of Computer Trends and Technology
电子版ISSN：2231-2803
出版年度：2013
卷号：4
期号：9-2
出版社：Seventh Sense Research Group
摘要：Largescale user contributed images with tags are easily available on photo sharing sites. However, the disturbance or inappropriate correspondence between the images and tags prohibits them from being leveraged for precise image retrieval and efficient management. To resolve the tag refinement problem, we proposing a Ranking Method based and Multicorrelation Tensor Factorization (RMTF), to jointly model the ternary relations inbetweenusers, imagetag, and nextimportantly rebuild the personalized imagetag associations as result. The user interest or background can be explored to eliminate the ambiguity of image tags so, the proposing RMTF is trusted to be guidelines to the formal solutions, which focus only on the binary format type image and tag correlation. When the model estimation is going, we use a ranking based optimization scheme to interpret the tagged data, which is according to pairing quality related difference between positive and negative samples is used, in the place of the point wise 0/1 trust. Clearly, the positive samples are directly decided by the observed userimagetag relations, when the negative samples are collected with respect to the most semantically and contextually irrelevant tags. Extensive experiments on a benchmark Flicker dataset demonstrate the effectiveness of the proposed solution for tagrefinement. We also exampledgood performances on two potential applications as the byproducts of the ternary relation analysis.
