Deep Embedding Clustering on Citation Networks for Community Detection

Citation networks are crucial for understanding academic knowledge organization and research dynamics, but traditional community detection algorithms struggle with their high complexity and sensitivity to noise. Therefore, our aim is to present a community discovery algorithm, known as deep embedding clustering (DECCN), with the purpose of enhancing the precision and efficiency of community detection in citation networks. Firstly, the k-core algorithm is employed for initial feature extraction in the citation network, resulting in a similarity matrix. Then, the dimensionality of the similarity matrix is reduced using a variational autoencoder (V AE) to obtain a low-dimensional node embedding representation. Finally, a Gaussian Mixture Model (GMM) is applied for clustering, categorizing nodes into different communities. We obtained multiple citation network datasets from literature databases and compared them with other community detection algorithms. Through relevant experiments, we demonstrated that the DECCN algorithm significantly improves accuracy and efficiency in community discovery on citation networks compared to traditional algorithms.