Tag ranking by propagating relevance over tag and image graphs

In this paper, we explore the problem of tag ranking by propagating relevance over community-contributed images and their associated tags. To rank the tags more accurately, we propose a novel tag ranking scheme through a two-stage graph-based relevance propagation approach. The first stage constructs a tag graph on each image and implements a random walk process on it in order to get the initial relevance of each tag for one image and the second stage builds a kNN-sparse image graph and propagates the relevance of tags among the web images. The proposed approach is purely data-driven, since the explicit relevance models between tags and images are not assumed. More importantly, compared to existing tag ranking approaches, we propose to leverage the relevance propagation over two graphs, which take into count not only the relationship among tags but also the relationship among images. Extensive experiments have conducted on the NUS-WIDE dataset have demonstrated the effectiveness of the proposed approach.