Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploiting Inter-Session Information with Frequency-enhanced Dual-Path Networks for Sequential Recommendation

Peng He, Yao Liu, Yanglei Gan, Run Lin, Tingting Dai, Qiao Liu, Xuexin Li

TL;DR

This paper addresses the bottleneck in sequential recommendation where self-attention smooths high-frequency user signals and prior frequency-based models ignore inter-session dependencies. It introduces FreqRec, a Frequency-Enhanced Dual-Path Network that jointly models inter-session and intra-session dynamics via a learnable complex-valued Fourier transform, consisting of a Global Spectral Aggregator and a Local Spectral Refiner. A frequency-domain consistency loss aligns predicted spectral coefficients with ground-truth signatures, enabling explicit recovery of high-frequency patterns. Across three real-world benchmarks, FreqRec achieves state-of-the-art results and demonstrates robustness to data sparsity and noisy logs, underlining the value of cross-session spectral modeling for SR.

Abstract

Sequential recommendation (SR) aims to predict a user's next item preference by modeling historical interaction sequences. Recent advances often integrate frequency-domain modules to compensate for self-attention's low-pass nature by restoring the high-frequency signals critical for personalized recommendations. Nevertheless, existing frequency-aware solutions process each session in isolation and optimize exclusively with time-domain objectives. Consequently, they overlook cross-session spectral dependencies and fail to enforce alignment between predicted and actual spectral signatures, leaving valuable frequency information under-exploited. To this end, we propose FreqRec, a Frequency-Enhanced Dual-Path Network for sequential Recommendation that jointly captures inter-session and intra-session behaviors via a learnable Frequency-domain Multi-layer Perceptrons. Moreover, FreqRec is optimized under a composite objective that combines cross entropy with a frequency-domain consistency loss, explicitly aligning predicted and true spectral signatures. Extensive experiments on three benchmarks show that FreqRec surpasses strong baselines and remains robust under data sparsity and noisy-log conditions.

Exploiting Inter-Session Information with Frequency-enhanced Dual-Path Networks for Sequential Recommendation

TL;DR

This paper addresses the bottleneck in sequential recommendation where self-attention smooths high-frequency user signals and prior frequency-based models ignore inter-session dependencies. It introduces FreqRec, a Frequency-Enhanced Dual-Path Network that jointly models inter-session and intra-session dynamics via a learnable complex-valued Fourier transform, consisting of a Global Spectral Aggregator and a Local Spectral Refiner. A frequency-domain consistency loss aligns predicted spectral coefficients with ground-truth signatures, enabling explicit recovery of high-frequency patterns. Across three real-world benchmarks, FreqRec achieves state-of-the-art results and demonstrates robustness to data sparsity and noisy logs, underlining the value of cross-session spectral modeling for SR.

Abstract

Sequential recommendation (SR) aims to predict a user's next item preference by modeling historical interaction sequences. Recent advances often integrate frequency-domain modules to compensate for self-attention's low-pass nature by restoring the high-frequency signals critical for personalized recommendations. Nevertheless, existing frequency-aware solutions process each session in isolation and optimize exclusively with time-domain objectives. Consequently, they overlook cross-session spectral dependencies and fail to enforce alignment between predicted and actual spectral signatures, leaving valuable frequency information under-exploited. To this end, we propose FreqRec, a Frequency-Enhanced Dual-Path Network for sequential Recommendation that jointly captures inter-session and intra-session behaviors via a learnable Frequency-domain Multi-layer Perceptrons. Moreover, FreqRec is optimized under a composite objective that combines cross entropy with a frequency-domain consistency loss, explicitly aligning predicted and true spectral signatures. Extensive experiments on three benchmarks show that FreqRec surpasses strong baselines and remains robust under data sparsity and noisy-log conditions.

Paper Structure

This paper contains 26 sections, 15 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Sequential Recommendation
  4. DFT for Sequential Modeling
  5. Preliminaries
  6. Methodology
  7. Historical Sequence Encoding
  8. Dual-Path Representation Learning
  9. Contextual-enhanced Representation.
  10. Frequency-enhanced Network.
  11. Training Objective
  12. Cross-Entropy Loss.
  13. Frequency Domain Loss.
  14. Experiments
  15. Experimental Setup
  16. ...and 11 more sections

Figures (4)

  • Figure 1: The distribution of Pearson correlation coefficient between sessions sharing at least one item on Sports & Outdoors and Beauty datasets, respectively.
  • Figure 2: The overall framework of FreqRec. User sequences are embedded and then processed by two parallel paths: a self-attention branch and the FreqNet branch. FreqNet contains a Global Spectral Aggregator that performs batch-axis DFT $\rightarrow$ FreqMLPs $\rightarrow$ IDFT to distill cohort-level signals, while a Local Spectral Refiner applies the same pipeline along the temporal axis to refine user-specific cues. The two spectral modules can be fused in either parallel or serial form. Training minimizes a hybrid objective that couples standard cross-entropy prediction with frequency-domain consistency loss.
  • Figure 3: Demonstration of Frequency-Domain MLP.
  • Figure 4: Sensitivity test of hyper-parameters on the Beauty and Toys & Games datasets.