Social media platforms like X (formerly Twitter) play a central role in public discourse but are also exploited for influence operations (IO) through coordinated inauthentic behavior (CIB). This study proposes a method to detect IO-related coordinated communities during the August 2023 release of Advanced Liquid Processing System (ALPS)-treated water from the Fukushima Daiichi Nuclear Power Plant. Using reposting data, we construct a graph of user communities based on network science techniques, incorporating both intra- and inter-community features. A Graph Neural Networks (GNN) is trained on these structures to classify communities as abnormal or normal. The model achieves F1 = 0.97, outperforming baseline methods. By automating early detection of coordinated communities, our method supports timely countermeasures in IO. While effective, further work is needed to capture communities with varied intents and improve attribution. The results demonstrate the potential of graph-based learning in real-time monitoring of influence activities on social platforms.