Modeling Network with Topic Model and Triangle Motif

This paper describes a hierarchical model based on triangle motif and topic model, considering both network data and node attribute. The attribute of nodes we study here is text, so we choose document network as our research content. We represent the document network with triangle motif, which has good scalability on large amount of data. This representation makes the complexity of our approach grows linearly in the number of documents, and more relational with the max degree of the network. We extend hLDA by incorporating network data, remodeling the hLDA. Using non-parametric Bayesian model, our approach does not need pre-specification of the branch factor at each non-terminal. The model is suitable for large-scale network of academic abstract, web document and related news.