Popular News Always Compete for the User's Attention! POPK: Mitigating Popularity Bias via a Temporal-Counterfactual

TL;DR

This paper tackles popularity bias in news recommendations by introducing POPK, a temporal-counterfactual approach that modifies negative sampling to include competing popular articles at time $t$ via a function $ extbf{P}(t, popk)$. It supports two popularity logics, acc and ptb, and multiple popularity definitions (e.g., top accumulated clicks, top per-hour clicks, and variants by click ratio or variance), all integrated into a metric-learning loss. Through experiments on Nikkei (Japanese), MIND-small (English), and Adressa (Norwegian), POPK consistently improves accuracy and, in many settings, diversity across three strong baselines (NRMS, NAML, LSTUR), with results depending on dataset and configuration. The method is practical, lightweight to implement, and adaptable to different objectives, making it suitable for real-world systems seeking to reduce popularity skew while preserving user personalization; future work includes learning popularity measures that adapt over time.

Abstract

In news recommendation systems, reducing popularity bias is essential for delivering accurate and diverse recommendations. This paper presents POPK, a new method that uses temporal-counterfactual analysis to mitigate the influence of popular news articles. By asking, "What if, at a given time $t$, a set of popular news articles were competing for the user's attention to be clicked?", POPK aims to improve recommendation accuracy and diversity. We tested POPK on three different language datasets (Japanese, English, and Norwegian) and found that it successfully enhances traditional methods. POPK offers flexibility for customization to enhance either accuracy or diversity, alongside providing distinct ways of measuring popularity. We argue that popular news articles always compete for attention, even if they are not explicitly present in the user's impression list. POPK systematically eliminates the implicit influence of popular news articles during each training step. We combine counterfactual reasoning with a temporal approach to adjust the negative sample space, refining understanding of user interests. Our findings underscore how POPK effectively enhances the accuracy and diversity of recommended articles while also tailoring the approach to specific needs.

Figures (8)

• Figure 1: POPK overall idea.
• Figure 2: Process of selecting the $popk$ most popular news articles.
• Figure 3: Detailed information per news article at a given time $t$.
• Figure 4: Diversity increase $D_{ctg}@5/10$ between original baseline model and POPK version.
• Figure 5: Category log-heatmap NRMS model ($popk$ = 1 to 3, acc).