Sequential Recommendation with Controllable Diversification: Representation Degeneration and Diversity

TL;DR

The paper tackles the limited diversity in sequential recommendations caused by representation degeneration under long-tail user/item distributions. It introduces SPMRec, a differentiable framework that uses singular spectrum smoothing guided by a new diversification proxy AUSC to jointly regularize sequence and item embeddings, thereby increasing diversity without sacrificing accuracy. The method defines L_seq and L_item to smooth the spectrum on both sides and optimizes L = L_rec + λL_seq + βL_item, enabling a controllable trade-off between performance and diversity. Empirical results on four Amazon-category datasets show that SPMRec consistently improves Recall and NDCG while boosting diversity metrics, with pronounced benefits for short sequences and sparse data, indicating practical applicability for real-world SR systems.

Abstract

Sequential recommendation (SR) models the dynamic user preferences and generates the next-item prediction as the affinity between the sequence and items, in a joint latent space with low dimensions (i.e., the sequence and item embedding space). Both sequence and item representations suffer from the representation degeneration issue due to the user/item long-tail distributions, where tail users/ items are indistinguishably distributed as a narrow cone in the latent space. We argue that the representation degeneration issue is the root cause of insufficient recommendation diversity in existing SR methods, impairing the user potential exploration and further worsening the echo chamber issue. In this work, we first disclose the connection between the representation degeneration and recommendation diversity, in which severer representation degeneration indicates lower recommendation diversity. We then propose a novel Singular sPectrum sMoothing regularization for Recommendation (SPMRec), which acts as a controllable surrogate to alleviate the degeneration and achieve the balance between recommendation diversity and performance. The proposed smoothing regularization alleviates the degeneration by maximizing the area under the singular value curve, which is also the diversity surrogate. We conduct experiments on four benchmark datasets to demonstrate the superiority of SPMRec, and show that the proposed singular spectrum smoothing can control the balance of recommendation performance and diversity simultaneously.

Figures (10)

• Figure 1: A demonstrated strong correlation between the proposed spectrum smoothing regularization (weighted by $\beta$) and intra-list diversity on public Amazon review data. A challenging balance between recommendation performance and diversity can be achieved with the proposed method.
• Figure 2: Fast singular values decay of user sequence output and item embeddings of SASRec are observed, implying the representation degeneration exists in both users and items. Singular values are normalized by dividing the largest singular value.
• Figure 3: Intuition of Singular Spectrum Smoothing for addressing sequence and item representation degeneration.
• Figure 4: Relationship between item spectrum smoothing weight $\beta$ and diversity based on embeddings.
• Figure 5: Relationship between item spectrum smoothing weight $\beta$ and diversity based on coverage@100.