Adaptive Recommendation and Learning Behavior Prediction of Textile English Teaching Resources Based on Transformer and Emotion Recognition Algorithms

To address the insufficient real-time perception of learners’ emotional states and learning behavior trajectories in textile English teaching, which limits the accuracy and dynamic adaptability of resource recommendation, this study develops a collaborative framework for adaptive recommendation and learning behavior prediction. A Transformer variant equipped with a multi-head emotional attention mechanism is employed to process learner interaction text and generate emotionally enhanced semantic representations. These representations are subsequently integrated with a lightweight Gated Recurrent Unit (GRU) predictor to infer learning behavior trajectories. By dynamically evaluating the semantic consistency between emotionally enhanced representations and predicted behavioral patterns, a personalized recommendation list is generated through a differentiable ranking optimization mechanism. Experimental results demonstrate that the proposed approach achieves recommendation matching accuracies of 77.2% in the fiber materials domain, 81.5% in the weaving technology domain, and 78.3% in the fashion design domain, confirming the effectiveness of collaborative optimization between resource recommendation and behavior prediction. Furthermore, the proposed framework provides a data-driven paradigm for multimodal cognitive state perception and intelligent information processing, offering potential methodological support for adaptive signal interpretation and intelligent sensing applications in advanced engineering systems.