NetGloVe: Learning Node Representations for Community Detection
Abstract
Community detection is a fundamental task in network analysis, with plenty of applications in social networking, biology and neuroscience. In the related literature, a variety of algorithms and methodologies have been proposed to identify the community structure of networks, including graph partitioning methods, hierarchical graph clustering, modularity optimization and spectral techniques (such as spectral clustering and modularity optimization).
The recent advances in representation learning techniques, have allowed us to represent graphs (or nodes) as vectors in a lower dimensional space, that can further be used in graph mining and learning tasks. That way, instead of “manually” extracting features that can be utilized by a graph learning algorithm, we can learn informative and discriminative feature representations by solving an optimization problem that takes into account the structural properties of the graph. To this direction, several network feature learning algorithms have been proposed, including node2vec [1] and LINE [4]. The goal of this work is to propose NetGloVe, a new representation learning method for graphs inspired from the domain of Natural Language Processing (NLP), and to examine its application to the task of community detection.
