Optimized Label Propagation Community Detection on Big Data Networks

Identifying community structures and subnetwork patterns for complex networks provide us with great knowledge about network. Community detection has been getting lots of attention and interest in recent years. The application for such knowledge goes from target marketing to biology, social studies, and physics. The existing algorithms either lack accuracy or are too slow for Big Data graphs. Due to the rise of Big Data graphs, such solutions prove impractical for real-world datasets. In this study, we change the feed system for the Label Propagation algorithm from a random method to a degree-based system. In addition, we introduce a new convergence method that checks the membership for every node and flags them as converged when they meet the requirement. The main contributions of this work are twofold: (i) we optimize the Label Propagation algorithm, improving the accuracy by a factor of two. The results depend on the complexity of the graph; i.e. the denser a graph structure is, the better result the algorithm will achieve. (ii) We solved the inconsistency of identified communities of Label Propagation algorithm. The results are depicted using two well-known metrics known as the Normalized Mutual Information and the Adjusted Rand Index. We present that Optimized Label Propagation has better results in various real-world dataset and artificial datasets.