Social Recommendation Based On Manifold Learning And Social Network User Information

With the rapid increasement of social multimedia information,it becomes increasingly difficult for users to mine their interested contents,which makes user customized recommendation an urgent technique for the further development of social multimedia.In investigating social multimedia computing,both multimedia semantic gap in multi-modal information and users' intention gap from information overload make obstacles to social multimedia computing.The target of this thesis is to investigate the role of visual semantic gap and multi-modal collaborative relationship in social recommendation.We attempt to construct cross-modal information correspondence by mining hidden correlation among social multimedia data for providing users personalized image recommendation.The main works are introduced in three parts as follows.1.We proposed manifold progressive propagation based image recommendation method.Considering variation in content and their complexity of visual information,we perform manifold propagation to infer user-image correlation progressively in order of increasing learning difficulty.Learning from simple to difficult samples helps optimize learning ability of the model gradually and simultaneously alleviate computational burden of model on manifold scale.2.We propose semantic manifold modularization-based ranking(MMR)for image recommendation.MMR attempts to leverage local semantic distribution of manifold to propagate users' preference.Semantic manifold modularization effectively decomposes the global visual manifold into semantic groups to avoid bias cross semantics and reduce computational burden on manifold learning.The final dense user-image correlations for recommendation are inferred by propagating users' historical records on semantic submanifolds sequentially,which keeps the merits of manifold learning in discriminative capturing and extends the scalability of manifold computing.It reduce bias cross semantics and computational burden simultaneously on global manifold.3.We propose cross-modal collaborative manifold propagation(CMP)for image recommendation.CMP leverages users' interest distribution to propagate images' user records,which lets users know the trend from others and produces interest-aware image candidates upon users' interests.Visual distribution is investigated simultaneously to propagate users' visual records along dense semantic visual manifold.Visual manifold propagation helps to estimate semantic detailed user-image correlations for the candidate images in recommendation ranking.