User Interest Mining Based On User Photo Albums

The explosive growth of information technology and computer network has broughthuge social impact. Through internet, we almost can find any information we need. Butinstead of being helpless, we spend too much time on searching information we wantedin the ocean of information. How to provide users more personalized, more targeted ser-vice become more and more attractive among Internet companies. Top these problems isunderstanding user’s interest.There are various successful applications based on deep study of user interests, such aspersonalized search and personalized recommendation. The essence of these applications isthe widely collection of user online behavior, search history, shopping records to accuratelymodel user interest. With this model, it is efective to recommend user-related contentand products. Most interest model are based on user’s search history, seldom make useof multimedia content related to user, such as user photo album. With rapid developmentof photo sharing websites, we try to mining user interest based on user’s photo album.Our work has two contributions: we crawling million of web images which contain usercomments to build a large image database for image annotation usage; we build a top treebased on open directory project which can generate user interest distribution use annotatedimages as input. Firstly, we use a content base image retrieval system to find near duplicateimages on image database for each image from user’s photo album; we annotated imagewith annotations associated with search result images; find nearest topic which user photoimages may represent in topic tree, generate user interest. The experiments results indicateit’s a feasible way to mining user interest.