Citation Networks to Easily Identify Relevant Publications in Any Given Research Field

Abstract

The procedure is tested on 10 citation networks and the communities are identified and the dominant topic in each community is identified. Once the dominant topic of the identified communities is found out, then ranking its members is relatively easy, since the challenge before applying any graph-based page ranking algorithm lies in selecting a set of a most similar set of documents which can now be obtained using the proposed method. The work presents the application of community detection and topic modeling on citation networks to easily identify relevant publications in any given research field Community detection in networks is a method of dividing it into components where the vertices belonging to each component are strongly interconnected and the components themselves are relatively poorly connected. This work uses the application of consensus clustering for community detection. Following the identification of communities, topic modeling is performed to determine the underlying topic of every community. Latent Dirichlet Allocation (LDA) is used to perform topic modeling on the derived communities. The proposed procedure acts as a pre-processing stage for page ranking of citation network members.