Social image tag enrichment based on textual similarity modeling

摘要

In social image sharing websites, users provide several descriptive tags to annotate their shared images. Usually, the user annotated tags are noisy, biased and incomplete. How to improve tag quality is very important for tag based applications. The content relevant tags have certain similarities or connections with each other. Thus from some highly relevant tags, we can infer the other content relevant tags for an image. In this paper, a social image tag enrichment approach is proposed. Considering the diversity of content relevant tags for the image, we first determine some seed tags which are highly relevant to image content and cover wide range of semantics. Then the seed tags are utilized to adopt semantic similarity tags for the input image. Experiments demonstrate the effectiveness of the proposed approach.