Robust Collaborative Filtering to Popularity Distribution Shift

Abstract

In leading collaborative filtering (CF) models, representations of users and items are prone to learn popularity bias in the training data as shortcuts. The popularity shortcut tricks are good for in-distribution (ID) performance but poorly generalized to out-of-distribution (OOD) data, i.e., when popularity distribution of test data shifts w.r.t. the training one. To close the gap, debiasing strategies try to assess the shortcut degrees and mitigate them from the representations. However, there exist two deficiencies: (1) when measuring the shortcut degrees, most strategies only use statistical metrics on a single aspect ( i.e., item frequency on item and user frequency on user aspect), failing to accommodate the compositional degree of a user-item pair; (2) when mitigating shortcuts, many strategies assume that the test distribution is known in advance. This results in low-quality debiased representations. Worse still, these strategies achieve OOD generalizability with a sacrifice on ID performance. In this work, we present a simple yet effective debiasing strategy, PopGo , which quantifies and reduces the interaction-wise popularity shortcut without any assumptions on the test data. It first learns a shortcut model, which yields a shortcut degree of a user-item pair based on their popularity representations. Then, it trains the CF model by adjusting the predictions with the interaction-wise shortcut degrees. By taking both causal- and information-theoretical looks at PopGo, we can justify why it encourages the CF model to capture the critical popularity-agnostic features while leaving the spurious popularity-relevant patterns out. We use PopGo to debias two high-performing CF models (MF [ 28 ], LightGCN [ 19 ]) on four benchmark datasets. On both ID and OOD test sets, PopGo achieves significant gains over the state-of-the-art debiasing strategies ( e.g., DICE [ 71 ], MACR [ 58 ]). Codes and datasets are available at https://github.com/anzhang314/PopGo.

Neighborhood-Enhanced Supervised Contrastive Learning for Collaborative Filtering

“…They proposed that high uniformity would lead to low tolerance, and make the learned model may push away two samples with similar semantics. Besides, the authors in the paper [49] studied how to utilize the popularity degree information to help the collaborative model automatically adjust the collaborative representations optimization intensity of any user-item pair [49]. The most related work to ours is SupCon [12].…”

General Debiasing for Graph-based Collaborative Filtering via Adversarial Graph Dropout

“…Given that propensity scores in IPS approaches can exhibit high variance, many studies [8,24] have turned to normalization or smoothing penalties to ensure model stability. Recent works [47,[66][67][68] have drawn inspiration from Stable Learning and Causal Inference. For instance, MACR [56] conducts counterfactual inference using a causal graph, postulating that popularity bias originates from the item node influencing the ranking score.…”

“…sDRO [57] integrates a Distributionally Robust Optimization (DRO) framework to minimize loss variances in long-tailed data distributions. PopGo [66] quantifies and reduces the interaction-wise popularity shortcut and is theoretically interpreted from both causality and information theory. However, devising these causal graphs and understanding the environmental context often hinge on heuristic insights from researchers.…”

“…In response to such negative influences, recent debiasing strategies [7,11,54,55,57,67,71] have emerged. However, we argue that there are several limitations: (1) Data Bias Perspective: Typically, only one specific factor is considered to be mitigated, such as item popularity [15,23,57,66], user conformity [71], and sensitive attributes [7]. However, interaction data is often riddled with various biases, whether predefined or arising from latent confounders.…”