Enhancing random walk algorithms with Personalized PageRank

In this paper, we explore an innovative approach to community detectionin graphs, combining Personalized PageRank (PPR) preprocessing withNode2vec embedding and Q-learning clustering. PPR preprocessing refines theinput graph, enhancing the local relevance of edges, which is then transformedinto low-dimensional representations using Node2vec. These embeddings capturethe structural nuances and similarities within the graph. Finally, weapply Q-learning, a reinforcement learning algorithm, to estimate number ofclusters and apply kmeans on these embeddings, optimizing community detection.Our method demonstrates betrer accuracy and robustness compared tobare random walk based approaches and traditional techniques, thus offeringan improvement on existing methods.