<p>Sequential recommendation is used to model user preferences by analyzing historical interaction data. However, existing approaches based on self-attention mechanisms usually consider user preferences only from the time-domain perspective, ignoring the inherent periodicity in user behavior. To address this challenge, we introduce an innovative solution called FDSRec, which combines frequency domain self-attention networks with contrastive learning techniques. Specifically, we propose a frequency domain self-attention encoder to capture periodic variations in user behavior data. We convert user representations into frequency domain representations by employing the Fast Fourier Transform technique. Subsequently, the frequency domain self-attention networks are used to learn the correlation and importance among different frequency domain components. In addition, we design time-domain and frequency-domain enhancement methods to address data sparsity and noise issues. Furthermore, we employ a multi-task technique to optimize sequential recommendation and contrastive learning objectives. Finally, extensive experiments are conducted on three publicly available datasets, demonstrating that FDSRec outperforms the state-of-the-art methods.</p>