Predictive Tagging of Social Media Images using Unsupervised Learning

The popularity of online social media has provided a huge repository of multimedia contents. To effectively retrieve and store this multimedia content and to mine useful pattern from this data is a herculean task. This paper deals with the problems of social image tagging. Multimedia tagging i.e. assigning tags or some keywords to multimedia contents like images, audio, video etc. by users is reshaping the way the people generally search multimedia resources. This huge amount of data must be effectively mined and knowledge is discovered to find some useful patterns hidden in it. Some facts like Facebook which has more than one billion active users, and millions of photos are uploaded daily, YouTube has 490 million unique users who visit every month, People upload 3,000 images to Flickr (the photo sharing social media site) every minute, Flickr hosts over 5 billion images. Apart from their usage for general purpose search, they are also leading towards many diverse areas of research like land mark recognition, tag recommendation, tag relevancy, automatic image tagging or annotation. This paper addresses the problem of automatic image tagging or Predictive tagging of digital images in social network scenario. Predictive tagging aims to automatically predict tags and check the relevancy of tags associated with images. This can be accomplished by using unsupervised learning. General Terms Automatic image annotation, content based image retrieval, image annotation, image mining and multimedia data mining.