With the growing prevalence of social media addiction, conventional psychological and algorithmic models fall short in fully capturing the complex interplay between user behavior, cognitive processing, and physical fatigue. This paper proposes a novel interdisciplinary framework that integrates biomechanical and neurobehavioral insights into the optimization of social media recommendation systems. By examining fine motor adaptations, dopamine-driven feedback loops, and user fatigue dynamics, the study enhances an SVD-based collaborative filtering algorithm through the incorporation of neuromechanical parameters, such as finger muscle fatigue and attention decay. In parallel, a BERT-LSTM-based rumor detection model is implemented to address content reliability under varying physiological states. Empirical results from 665 users demonstrate significant performance improvements. Compared with traditional recommendation algorithms, the optimized SVD model reduced average reaction time by 21.6%, increased operational precision by 7.6%, and decreased finger muscle fatigue by 13.6%. Additional ablation experiments highlight the contribution of cepstral features over fundamental frequency, and the critical role of 4-gram language models in enhancing melody and behavior recognition accuracy.
