Social image tag recommendation by concept matching

Tags associated with social images are valuable information source for superior image search and retrieval experiences. In this paper, we propose a novel tag recommendation technique that exploits the user-given tags associated with images. Each candidate tag to be recommended is described by a few tag concepts derived from the collective knowledge embedded in the tag co-occurrence pairs. Each concept, represented by a few tags with high co-occurrences among themselves, is indexed as a textual document. Then user-given tags of an image is represented as a text query and the matching concepts are retrieved from the index. The candidate tags associated with the matching concepts are then recommended. Leverages on the well studied Information Retrieval (IR) techniques, the proposed approach leads to superior tag recommendation accuracy and lower execution time compared to the state-of-the-art.