Identifying key user influence in social topic networks is critical for public opinion management, advertising, and disaster prevention. This paper presents a novel approach for identifying key influencers in social topic networks by analyzing multi-dimensional propagation patterns. We develop a comprehensive model that integrates both vertical and horizontal propagation characteristics to enhance influence discovery accuracy. For vertical propagation, we propose DS-U2Vec, which employs random walks incorporating user intimacy and propagation depth to generate node representations. For horizontal propagation, we introduce MB-Link for community detection and GCN-based node characterization. These representations are then fused through an attention mechanism and processed by a multilayer perceptron for final influence assessment. Experimental results demonstrate our model’s superior performance in uncovering latent user relationships and identifying influential users, significantly outperforming conventional methods in capturing the complex dynamics of information diffusion.