<p>Sequential recommendation technology is of great value in personalized recommendations by modeling temporal dependencies, yet existing methods face challenges such as data sparsity, noise interference, and user interest drift. While contrastive learning partially alleviates these issues, traditional sequential architectures struggle to capture complex behavioral patterns through single-view modeling. This paper proposes a Dual-View Graph-Frequency Contrastive Learning for Sequential Recommendation (GFCLSRec) that optimizes semantic space through dual-view representation fusion. The framework reconstructs user interaction sequences into item graphs, establishing dual representations from sequential and graph-structural views to respectively capture explicit temporal dependencies and implicit high-order correlations. We design a collaborative enhancement mechanism integrating graph topology perturbation and spectral transformation to generate semantically consistent contrastive pairs. A multi-task learning architecture enables contrastive learning between graph-structural features and temporal-frequency characteristics, effectively distilling essential behavioral patterns from noise signals. Extensive experiments on four public datasets demonstrate significant performance improvements over state-of-the-art baselines, offering a novel perspective for multi-view contrastive learning in sequential recommendation systems.</p>