Topic Modeling for personalized recommendation using social media data

social network

In recent years, social media websites, such as Twitter, Facebook have gained in popularity and have become ubiquitous in our daily lives, where rich user-generated texts are propagated through social networks. Topic models, such as Latent Dirichlet Allocation (LDA), have been proposed and shown to be useful for text analysis. The existing topic models focus on traditional document collections, which consist of a relatively small number of long and high-quality documents. However, user-generated texts tend to be shorter and noisier than traditional content. Besides, the social networks have two novel features: context information on nodes, such as user features, and edges, such as relationship, which have not been considered by the existing topic models. In this work, we pose the problem of finding user topics in large-scale collection of documents from online social networks for recommending socially personalized content reflecting social relationship between social networking service (SNS) users.