FEDIN: Frequency-Enhanced Deep Interest Network for Click-Through Rate Prediction

2026-05-03

AI Summary

This work addresses the challenge of modeling periodic user interest patterns in sequential recommendation, which are often obscured by temporal behavioral noise. To this end, the authors propose a deep interest network that jointly leverages time and frequency domains. By analyzing the distributional discrepancy in spectral entropy of user attention under positive and negative samples, they design a target-aware spectral filtering mechanism to effectively extract low-entropy periodic interest signals while suppressing noise. The model employs a dual-branch architecture integrating both time-domain and frequency-domain representations, complemented by a spectral entropy–guided signal selection strategy. Experimental results on three public datasets demonstrate that the proposed method significantly outperforms state-of-the-art baselines, exhibiting superior robustness to noise and consistently stable click-through rate prediction performance.

Abstract

Sequential recommendation models often struggle to capture latent periodic patterns in user interests, primarily due to the noise inherent in time-domain behavioral data. While frequency-domain analysis offers a global perspective to address this, existing approaches typically treat user sequences in isolation, overlooking the crucial context of the target item. In this work, we present a novel empirical observation: user attention scores exhibit distinct spectral entropy distributions when conditioned on positive versus negative target items. Specifically, true user interests manifest as highly concentrated spectral patterns with lower entropy in the frequency domain, whereas irrelevant behaviors appear as high-entropy noise. Leveraging this insight, we propose the Frequency-Enhanced Deep Interest Network (FEDIN). FEDIN introduces a frequency-domain branch that utilizes a target-aware spectrum filtering mechanism to isolate these periodic interest signals. Extensive experiments on three public datasets demonstrate that FEDIN consistently outperforms state-of-the-art sequential recommendation baselines, demonstrating superior robustness against noise. We have released our code at: https://github.com/otokoneko/FEDIN.

Problem

• sequential recommendation
• periodic patterns
• frequency-domain analysis
• spectral entropy
• click-through rate prediction

Innovation

• frequency-domain analysis
• spectral entropy
• target-aware filtering
• periodic interest patterns
• click-through rate prediction