The analysis of large networks has become very usefull in a wide range of applications including social sciences, biology and complex systems. In graph mining, a key ingredient is to detect communities in these structured datasets. Graph clustering is usefull for enhanced data exploration and immediate visual insights of large networks.
LumenAI is specialized in real time machine learning. To tackle the challenge of graph mining, we have developed an innovative algorithm of community detection able to infer in real time the structure of a dynamic network. It is based on a real time optimization algorithm that maximizes a modularity index. The algorithm maintains a hierarchical clustering of nodes and allows to detect the merging of similar behaviours or the separation of two different ones in a complex system.
Below you can see the evolution of the algorithm in the case of a preferential attachment random graph, well-known to properly fit real-world properties.