Abstract— Social media sharing websites like Flickr allowusers to annotate images with free tags, which significantly contribute to thedevelopment of the web image retrieval and organization. Tag-based image searchis an important method to find images contributed by social users in suchsocial websites. However, how to make the top ranked result relevant and withdiversity is challenging. In this paper, we propose a social re-ranking systemfor tag-based image retrieval with the consideration of images relevance anddiversity[1]. We aim at re-ranking images according to their visualinformation, semantic information and social clues. The initial results includeimages contributed by different social users. Usually each user contributesseveral images. First we sort these images by inter-user re-ranking. Users thathave higher contribution to the given query rank higher. Then we sequentially implementintra-user re-ranking on the ranked users image set, and only the most relevantimage from each users image set is selected. These selected images compose thefinal retrieved results. We build an inverted index structure for the socialimage dataset to accelerate the searching process.
Keywords: Social Media, Tag-based Image Retrieval, Social Clues, Image search,Re-ranking.