A Simple Yet Effective Approach for Diversified Session-Based Recommendation

TL;DR

This paper targets the diversity-accuracy trade-off in session-based recommender systems (SBRS) by introducing DCA-SBRS, a simple yet effective plugin that can be attached to existing accuracy-focused SBRSs. It comprises two components: a model-agnostic diversity-oriented loss (MDL) that encourages diverse category coverage via $L_{div} = -H(\widehat{P}_c)$ and a joint training loss $L = L_{acc} + \lambda L_{div}$, and a non-invasive category-aware attention (NCA) that leverages category information within the attention mechanism, e.g., $\alpha_{tj} = v^T \sigma(A_1 (h_t + c_t^s) + A_2 (h_j + c_j^s))$. Extensive experiments on three real-world datasets show substantial diversity gains (e.g., ILD@10 up to ~138% on some sets) and balanced improvements in comprehensive metrics, with only modest drops in accuracy, demonstrating the framework's practical effectiveness. The work argues that this end-to-end, model-agnostic approach provides a lightweight, adaptable path to more trustworthy recommender systems by mitigating filter bubbles without sacrificing performance. It further discusses metric limitations and proposes adaptions (e.g., F$_{\beta}$) to better capture joint accuracy-diversity performance in practical deployments.

Abstract

Session-based recommender systems (SBRSs) have become extremely popular in view of the core capability of capturing short-term and dynamic user preferences. However, most SBRSs primarily maximize recommendation accuracy but ignore user minor preferences, thus leading to filter bubbles in the long run. Only a handful of works, being devoted to improving diversity, depend on unique model designs and calibrated loss functions, which cannot be easily adapted to existing accuracy-oriented SBRSs. It is thus worthwhile to come up with a simple yet effective design that can be used as a plugin to facilitate existing SBRSs on generating a more diversified list in the meantime preserving the recommendation accuracy. In this case, we propose an end-to-end framework applied for every existing representative (accuracy-oriented) SBRS, called diversified category-aware attentive SBRS (DCA-SBRS), to boost the performance on recommendation diversity. It consists of two novel designs: a model-agnostic diversity-oriented loss function, and a non-invasive category-aware attention mechanism. Extensive experiments on three datasets showcase that our framework helps existing SBRSs achieve extraordinary performance in terms of recommendation diversity and comprehensive performance, without significantly deteriorating recommendation accuracy compared to state-of-the-art accuracy-oriented SBRSs.

Figures (7)

• Figure 1: NARM vs NARM+MTL. Note: +MTL denotes the variant of NARM via leveraging item categories as input and adopting the common multi-task learning framework.
• Figure 2: An Overview of Our Proposed DCA-SBRS.
• Figure 3: The Unbalanced Grouping Induced by the Category (the symbol '$\times$' denotes the outliers with a mass of involved items).
• Figure 4: The Impact of MDL in Diversity w.r.t. ILD@$10$.
• Figure 5: The Impact of MDL for NARM+MDL with $N=10$.