Political discourse during conflict exhibits systematic behavioral changes that require multilayer network analysis to capture cross-dimensional coupling effects. We introduce a multilayer network framework that detects behavioral homogenization through three interconnected layers: toxicity similarity, political stance alignment, and social interactions. Analyzing 50,000 annotated messages from 10,375 users during the Russia-Ukraine conflict with inter-rater reliability \(\kappa > 0.85\) , findings reveal: (1) systematic density hierarchy with toxicity networks (density \(=\) 0.037, 1,999,000 edges) exceeding social networks (density \(= 3.2\times 10^{-6}\) , 173 edges) by 12,000-fold; (2) significant cross-layer coupling between stance and toxicity ( \(\rho = 0.347\) , \(p < 0.001\) ); (3) distinct group-specific behavioral signatures ( \(F(2{,}10372) = 847.3\) , \(p < 0.001\) , \(\eta ^2 = 0.140\) ). The multilayer framework improves the prediction of toxicity (AUC \(=\) 0.571 \(\rightarrow \) 0.621, accuracy 61.7% \(\rightarrow \) 66.5%) with Behavioral Homogenization Index (BHI \(=\) 0.412) indicating systematic convergence. This multilayer framework advances complex network methodologies for cross-dimensional behavioral analysis, with applications extending to social contagion, opinion dynamics, and conflict-driven network evolution. The demonstrated cross-layer coupling effects and systematic homogenization patterns establish new paradigms for analyzing multilayer networks in dynamic social systems.