Diversity Recommendation via Causal Deconfounding of Co-purchase Relations and Counterfactual Exposure

TL;DR

Cadence tackles diversity in recommendations through causal deconvolution of co-purchase signals and counterfactual exposure. It introduces Unbiased Asymmetric Co-purchase Relationship (UACR) to build a deconfounded item graph, refines embeddings via a novel graph aggregation, and uses a two-stage candidate selection with counterfactual exposure to uncover under-exposed yet relevant items. The approach maintains accuracy while improving diversity, supported by complexity- and stability-focused analyses, and reinforced by theoretical guarantees on embedding norms relative to item popularity. Empirical results across real-world datasets demonstrate superior diversity-accuracy trade-offs and strong statistical significance against state-of-the-art baselines, with good transferability and efficiency.

Abstract

Beyond user-item modeling, item-to-item relationships are increasingly used to enhance recommendation. However, common methods largely rely on co-occurrence, making them prone to item popularity bias and user attributes, which degrades embedding quality and performance. Meanwhile, although diversity is acknowledged as a key aspect of recommendation quality, existing research offers limited attention to it, with a notable lack of causal perspectives and theoretical grounding. To address these challenges, we propose Cadence: Diversity Recommendation via Causal Deconfounding of Co-purchase Relations and Counterfactual Exposure - a plug-and-play framework built upon LightGCN as the backbone, primarily designed to enhance recommendation diversity while preserving accuracy. First, we compute the Unbiased Asymmetric Co-purchase Relationship (UACR) between items - excluding item popularity and user attributes - to construct a deconfounded directed item graph, with an aggregation mechanism to refine embeddings. Second, we leverage UACR to identify diverse categories of items that exhibit strong causal relevance to a user's interacted items but have not yet been engaged with. We then simulate their behavior under high-exposure scenarios, thereby significantly enhancing recommendation diversity while preserving relevance. Extensive experiments on real-world datasets demonstrate that our method consistently outperforms state-of-the-art diversity models in both diversity and accuracy, and further validates its effectiveness, transferability, and efficiency over baselines.

Figures (2)

• Figure 1: Parameter sensitivity on $\alpha$ and $K_g$
• Figure 2: Performance comparison with error bars across three datasets. The error bars represent the standard deviation over five independent runs with different random seeds.