Context-aware Community Detection in the Russia-Ukraine Conflict Network.

Community detection in online social networks is a challenging field of research. The involvement of context information in online user interactions enhances the challenge of effective community detection. The proposed work addresses this challenge by developing a semi-supervised learning-based node labeling approach that augments the pre-existing Louvain algorithm to make it context-aware. This approach is tested in comparison to the un-modified Louvain algorithm on a dataset acquired from Twitter that includes user interactions related to the ‘Russia-Ukraine war’. This data is modeled into a network based on user interactions in the form of replies and retweets. A primary network-level analysis is performed using the different centrality metrics to identify influential users of the network. It is also noted that modularity, which is the main factor in the Louvain algorithm, is increased by more than twice on incorporating the context information as edge weight in our approach. Thus, it is validated that our proposed approach is effective in proper community detection by considering context-rich real-life network data.