Semantics Preserving Dimensionality Reduction of Citation Network Using Topic Models

This research focuses on constructing a more compact network from a large citation network using Latent Dirichlet Allocation (LDA) and evaluating the efficiency and structural characteristics of the resulting network. A comparison is made between the compact LDA-derived network and the original citation network across various graph-based metrics. The results demonstrate that, despite its reduced size, the LDA-derived network retains or even improves key structural properties, enhancing efficiency while preserving critical information.