Topic Models such as Latent Dirichlet Allocation (LDA) have been successfully applied as a data analysis and dimensionality reduction tool. With the emergence of social networks, many datasets are available in the form of a network with typed nodes (documents, authors, URLs, publication dates,...) and edges (authorship, citation, friendship,...). We propose a network-aware topic model that integrates rich, heterogeneous, network-based information, representing them using pathtyped random walks. In more detail, the proposed model is based on Dirichlet multinomial regression, an extension of LDA, as well as on random walks for exploiting network information; each document node is characterized by its connectivity to other nodes in the graph through a given set of random walks. A set of sparse latent parameters relate this characterization to topic assignments. Being sparse, the latent parameters give insight into the effect of different network features on the extracted topics.