Personalized Topic Graph Generation Method Using Image Labels in Image-Sharing SNS

Due to the rapid growth of social networking services (SNSs), many researchers have conducted research to find the users’ interests from them in order to recommend potential friends or contents related to their interests. However, it is very difficult to discover the users’ interests because of noisiness and sparsity of the SNS data. To overcome the difficulty, we propose a modified Latent Dirichlet Allocation (LDA) model named LDA-IL that utilizes relatively low-noise image data compared with text documents. In addition, it is important to find implicit interests as well as explicit interests of the users. To do so, we propose a method of generating the personalized topic graph that represents the users’ interests. To prove the validity of the LDA-IL model and the personalized topic graph, we developed an illustrative scenario and performed experiments.