ABSTRACT
Multimedia social search refers to a class of problems, such as tagging, retrieval, and recommendation, where both multimedia content and social context (i.e., information distilled from the web that is correlated to the content) are exploited. Ranking is the heart of multimedia social search. The motivation behind multimedia social search is that retrieval or recommendation based only on content yields frequently unsatisfactory results due to the well known semantic gap. Thanks to online social sharing sites (e.g., Flickr [https://www.flickr.com], Lastfm [https://www.last.fm], etc.) multimedia come with additional metadata, including, ownership, tags, geo-location, etc. Such metadata offer rich complementary information worth exploiting. A new transductive learning framework has recently emerged that addresses the aforementioned problems as ranking on hypergraphs by
jointly analyzing the content and its associated context defined by the metadata in a unified manner.
