<p>Learning emotion propagation in smart teaching environments exhibits dynamic fluctuations along the instructional sequence of knowledge point instruction, typically characterized by short cascade paths and complex interactions among nodes. To address these challenges, we propose ECTL, a deep learning model for predicting learning emotion propagation depth in smart teaching scenarios. Specifically, the model introduces a hybrid temporal encoding module that integrates linear, sine, and cosine mappings to capture multi-scale temporal dependencies. In addition, a multi-head attention mechanism incorporating positional encoding and distance-aware bias is developed and combined with an LSTM module to jointly model the structural dependencies and temporal dynamics of emotion propagation sequences. To further enhance feature representation, a multi-head parallel feed-forward network is employed to effectively extract cascade features and improve prediction accuracy. Finally, the predicted propagation depth is used as a constraint in a propagation path prediction algorithm to uncover potential latent pathways of learning emotion diffusion. Experimental results demonstrate that the proposed model achieves state-of-the-art performance in the emotion propagation depth prediction task within smart teaching environments. Moreover, comparative experiments conducted on multiple public cascade datasets further confirm that the underlying architecture possesses strong generalization capability in modeling complex network topologies and cascade structures.</p>