Personalized Image Tag Recommendation Algorithm for Web2.0 Platform Utilizing Tensor Factorization

In this paper, we a novel personalized image tag recommendation algorithm based on tensor factorization which is suitable to be used in the Web2.0 Platform. Firstly, the framework of the personalized image tag recommendation system is given, which is made up of three parts: 1) collecting the history of user tagging behaviors, 2) modeling the user interests from the tagging history, 3) obtaining the personalized image tag recommendation results through combining the image features and user interests together based on tensor factorization. Secondly, the tensor factorization based personalized image tag recommendation algorithm is given. The main work of this paper lies in that given a specific user, the proposed algorithm provide a set of personalized tags which are ranked according to the relative degree to user interests. Furthermore, the personalized image tags can be obtained by calculating the predictor through multiplying three feature matrices which represent the information of users, tags, and images respectively. Finally, experiments are conducted to make performance using the NUS-WIDE dataset under evaluation metric MRR, S@k, and P@k respectively. Experimental results demonstrate that the proposed algorithm can provide personalized image tags more accurately than other methods.