Wavelet Enhanced Adaptive Frequency Filter for Sequential Recommendation

TL;DR

Sequential recommendation suffers from non-stationary, intertwined user preferences that are difficult to extract from time-domain data. The authors introduce WEARec, which combines Dynamic Frequency-domain Filtering to adaptively filter spectral content based on user context and Wavelet Feature Enhancement to recover non-stationary, short-term signals via Haar wavelets. The framework blends global, adaptive spectral information with enhanced local details in a transformer-like encoder, achieving strong performance and efficiency gains for long sequences. Extensive experiments on four public datasets show WEARec consistently outperforms state-of-the-art baselines, demonstrating the value of integrating adaptive spectral filtering with wavelet-based local reconstruction for sequential recommendations.

Abstract

Sequential recommendation has garnered significant attention for its ability to capture dynamic preferences by mining users' historical interaction data. Given that users' complex and intertwined periodic preferences are difficult to disentangle in the time domain, recent research is exploring frequency domain analysis to identify these hidden patterns. However, current frequency-domain-based methods suffer from two key limitations: (i) They primarily employ static filters with fixed characteristics, overlooking the personalized nature of behavioral patterns; (ii) While the global discrete Fourier transform excels at modeling long-range dependencies, it can blur non-stationary signals and short-term fluctuations. To overcome these limitations, we propose a novel method called Wavelet Enhanced Adaptive Frequency Filter for Sequential Recommendation. Specifically, it consists of two vital modules: dynamic frequency-domain filtering and wavelet feature enhancement. The former is used to dynamically adjust filtering operations based on behavioral sequences to extract personalized global information, and the latter integrates wavelet transform to reconstruct sequences, enhancing blurred non-stationary signals and short-term fluctuations. Finally, these two modules work to achieve comprehensive performance and efficiency optimization in long sequential recommendation scenarios. Extensive experiments on four widely-used benchmark datasets demonstrate the superiority of our work.

Figures (8)

• Figure 1: Number of users uniquely driven by each frequency component in the Sports and Beauty datasets.
• Figure 2: The model architecture of WEARec is similar to the transformer encoder. It first generates item embedding with positional embedding through the embedding layer , and then extracts user preference from the frequency domain by replacing the self-attention module with the wavelet feature enhancement module and dynamic frequency-domain filtering module. Their details are shown on both sides. Finally, a prediction layer computes a recommendation score for all candidate items.
• Figure 3: The HR@20 performance comparison of WEARec with FMLPRec, SLIME4Rec and BSARec at different sequence lengths $N$ on ML-1M and LastFM.
• Figure 4: The HR@20 and NG@20 performance achieved by WEARec variants on four datasets.
• Figure 5: Performance of WEARec on HR@20 with varying hyperparameters..