FARec: frequency domain enhanced filter for sequential recommendation

Zhibin Yang, Jiwei Qin, Jie Ma, Donghao Zhang

Abstract: Sequential recommendation aims to model user preferences by analyzing their historical behavioral data. However, most existing approaches focus on modeling user preferences in the time domain, ignoring the impact of various frequency patterns on user behavior. These frequency patterns are often intertwined in the time domain and are challenging to distinguish, which limits the model’s ability to capture users’ behaviors at different frequencies. In addition, based on the F-principle, deep learning models pay more attention to low-frequency information, which may lead to poor performance in high-frequency tasks. To alleviate these problems, we propose a novel frequency-enhanced filter (FARec) for sequential recommendation. The model uses a learnable filter as an encoder to capture user preference features and introduces a data augmentation module and a frequency ladder structure to improve the model’s ability to capture different frequency features. The data augmentation module draws on the magnitude spectrum in Fourier analysis to introduce two data augmentation methods to accommodate events of different frequency magnitudes: frequency masking and frequency mixing. Moreover, we use frequency domain regularization to align the enhanced view with the original view. The frequency ladder structure splits the original sequence spectrum into multiple frequency bands, allowing the encoder to focus on different spectrum to capture different frequency patterns. Finally, comprehensive tests conducted across four benchmark datasets reveal that FARec outperforms the leading baseline models in effectiveness.

External IDs:dblp:journals/computing/YangQMZ25