Popularity Bias

Multiple-choice questions (MCQs) are a standard format for evaluating large language models (LLMs), yet the popularity of answer options can confound evaluation. Modern LLMs systematically prefer popular but incorrect options over less popular correct ones, a vulnerability we call popularity bias. This pattern aligns with confidence miscalibration: model confidence remains high even as accuracy collapses for popular options. To systematically isolate this phenomenon, we introduce PopMCQ, a benchmark with six controlled strategies that vary option popularity while keeping the correct answer fixed. In our most adversarial setting, where all distractors are more popular than the correct option, models choose popular but wrong answers 66% of the time. To mitigate this bias, we propose PopDebias, a lightweight inference-time correction that estimates and removes a popularity prior from model predictions. It requires no fine-tuning, is label-free at test time (using only a small calibration split for parameter fitting), and adds negligible computational cost. Experiments on 22 open-source LLMs (0.5B to 32B parameters) show consistent improvements, with accuracy gains up to 54.1 percentage points under strong popularity pressure. The code and data are available https://github.com/DataScienceUIBK/PopMCQ

Large-scale recommender systems, particularly short-form video platforms, are often bottlenecked by massive popularity feedback loops. In such environments, as models recommend popular items, they generate an overwhelming amount of skewed training data for "head" items. This creates a self-reinforcing cycle where retrieval and ranking models memorize "head" item patterns at the expense of generalizing across the vast "tail" of the catalogue. While Curriculum Learning (CL) offers a powerful mechanism to break this feedback loop by systematically exposing models to progressively more difficult and less frequent examples, its adoption in industrial recommendation has been hampered by hardware utilization inefficiencies or the needs for complicated pre-processing techniques because dynamic data rejection algorithms tend to starve hardware accelearators (TPUs/GPUs) by becoming largely CPU-bound. In this work, we introduce a scalable Soft Curriculum Learning framework designed specifically for continuous training setups within industry-scale retrieval and ranking models. By utilizing loss annealing and in-graph weight adjustments rather than rigid data filtering, we break the popularity feedback and enable dynamic curriculum pacing without sacrificing system throughput. We demonstrate empirical evidence through applications across sequence-based retrieval models (such as SASRec), two-tower retrieval models, and large-scale continuous ranking models. Online A/B tests on our short-video platform demonstrate substantial lifts in both overall user satisfaction and fresh content consumption, all without degrading model throughput.

Popularity bias in recommendation systems arises when a majority user class generates disproportionate interaction data, causing the system to increasingly favour it while degrading recommendation quality for niche users. While extensive empirical evidence of popularity bias exists, the dynamics leading to its emergence are not well understood. In this work, we study the coupled evolution of recommender model updates and user engagement through the lens of dynamical systems. We formulate a stochastic process and analyse its asymptotic behaviour through an ordinary differential equation (ODE) framework grounded in two-time-scale stochastic approximation. We characterise the equilibrium points of this dynamical system, and derive conditions under which popularity bias is provably emergent, as well as conditions under which symmetric retention of all user classes is possible. We conduct experiments on synthetic data and real-world production logs derived from a large-scale commercial music recommendation platform to validate our theoretical results.

The two-tower model has been widely used for large-scale recommendation systems, particularly in the retrieval stage. Industry standards for training two-tower models typically involve in-batch and/or out-of-batch negative sampling. However, these methods often produce easy negatives that models can quickly learn, failing to sufficiently challenge the model. To address this issue, a novel self-supervised hard negative sampling technique is proposed that leverages a large language model (LLM) to generate hard negatives from the same cluster during model training. By utilizing the LLM to learn media representations, the proposed approach ensures that the generated negatives are more challenging and informative. This real-time sampling framework is designed for seamless integration into production models, capable of handling billions of training data points with minimal computational complexity. Experiments on public datasets, along with deployment to a large-scale online system, demonstrate that the proposed negative sampling technique outperforms widely used industry methods. Furthermore, analysis in industrial applications reveals that this sampling method can help break inherent feedback loops in recommendations and significantly reduce popularity bias.

Knowledge graph completion (KGC) models are commonly evaluated using rank-based metrics such as MRR and Hits@K, despite different users often requiring different evaluation perspectives. In this demo, we present PROBE-Web, an interactive system for probing diverse evaluation landscapes for KGC models. PROBE-Web enables users to flexibly evaluate KGC models by adjusting two critical perspectives: (P1) predictive sharpness and (P2) popularity-bias robustness. Through a user-friendly GUI, users easily evaluate multiple KGC models and analyze their strengths and weaknesses. PROBE-Web provides four key functionalities: (1) conventional evaluation toolkit, (2) flexible perspective-aware evaluation, (3) explainable case studies, and (4) evaluation landscape exploration. We believe that PROBE-Web can help users better understand KGC models aligning with their objectives.

Knowledge graph completion (KGC) aims to predict missing facts from an observed knowledge graph (KG), playing a crucial role in a wide range of real-world applications such as drug discovery, recommender systems, and retrieval-augmented generation (RAG). Although numerous KGC models have been proposed, the evaluation of KGC remains underexplored, despite its critical role in reliably assessing model performance and selecting appropriate models for real-world applications. In this paper, we introduce two important perspectives for KGC evaluation that are overlooked by existing evaluation metrics, (P1) predictive sharpness and (P2) popularity-bias robustness. To address both perspectives, we propose a generalized evaluation framework, PROBE, which consists of a rank transformer (RT) that estimates the score of each prediction based on a desired level of predictive sharpness and a rank aggregator (RA) that determines the final evaluation score by aggregating all prediction scores according to a desired level of popularity-bias robustness. We theoretically analyze PROBE by defining six key properties for reliable KGC evaluation and prove that PROBE satisfies all the properties, while existing metrics fail to satisfy some. In particular, due to the open-world nature of KGs, an evaluation metric should preserve the relative performance of KGC models even when only incomplete facts are observed. We show that PROBE better maintains such consistency, providing a more reliable estimate of intrinsic model performance than existing metrics. Extensive experiments with six KGC models on six real-world KGs reveal that existing metrics may over- or under-estimate model performance depending on different evaluation perspectives, whereas PROBE enables a more comprehensive, flexible, and consistent evaluation of KGC models.

Recently, Generative Recommenders (GRs), characterized by a unified end-to-end framework, have exhibited astonishing potential in transforming the recommendation paradigm. Despite their effectiveness, we recognize that GRs are still susceptible to the long-standing issue of popularity bias that has pervaded the recommendation community. Although a few studies have attempted to extend traditional debiasing methods to GRs, their effectiveness is marginal, and the fundamental reason why GRs suffer from popularity bias remains under-explored. To bridge this gap, this study focuses on two core aspects in GRs: the optimization of generative framework and the item tokenization based on semantic index. Based on theoretical analyses, we identify that the severe popularity bias emerges from the confluence of a token-level optimization flaw and the undifferentiated property of item tokenization. Accordingly, this study develops a novel generative recommender system, called Ghost, by designing the asymmetric unlikelihood optimization and the skeleton-founded tokenization. Extensive empirical evaluations across three datasets, alongside multiple SOTA baselines, reveal that Ghost substantially alleviates popularity bias and promotes fairer recommendations, while incurring slight degradation to the overall recommendation utility.

Large language models (LLMs) exhibit systematic preferences for well-known entities, a phenomenon often attributed to popularity bias. However, the extent to which these preferences reflect real-world popularity versus statistical exposure during pretraining remains unclear, largely due to the inaccessibility of most training corpora. We provide the first direct, large-scale analysis of popularity bias grounded in fully observable pretraining data. Leveraging the open OLMo models and their complete pretraining corpus, Dolma, we compute precise entity-level exposure statistics across 7.4 trillion tokens. We analyze 2,000 entities spanning five types (Person, Location, Organization, Art, Product) and compare pretraining exposure against Wikipedia pageviews and two elicited LLM popularity signals: direct scalar estimation and pairwise comparison. Our results show that pretraining exposure strongly correlates with Wikipedia popularity, validating exposure as a meaningful proxy for real-world salience during the training period. More importantly, we find that LLM popularity judgments align more closely with exposure than with Wikipedia, especially when elicited via pairwise comparisons. This alignment is strongest for larger models and persists in the long tail, where Wikipedia popularity becomes unreliable. Overall, our findings demonstrate that popularity priors in LLMs are primarily shaped by pretraining statistics rather than external popularity signals, offering concrete evidence that data exposure plays a central role in driving popularity bias.

Collaborative filtering (CF) models based on graph neural networks (GNNs) achieve strong performance in recommender systems by propagating user-item signals over interaction graphs. However, they are susceptible to popularity bias, since skewed interactions and repeated message passing across high-order neighborhoods amplify the influence of popular items while suppressing long-tail ones. Existing debiasing approaches, including re-weighting objectives, regularization, causal methods, and post-processing, are less effective in GNN-based settings because they do not directly counteract bias propagated through the aggregation process, and recent in-aggregation weighting methods often rely on static heuristics or unstable embedding estimates. We propose Debiasing Popularity Amplification in Aggregation (DPAA), a popularity debiasing framework for GNN-based CF that integrates adaptive, representation-aware interaction weighting and layer-wise weighting directly into message passing. DPAA assigns interaction-level weights from a representation-based popularity signal, stabilized by a smooth transition from pre-trained to evolving model embeddings during training. It further introduces a layer-wise weighting that amplifies higher-order neighborhoods, surfacing long-range interactions with diverse and underexposed items. Experiments on real-world and semi-synthetic datasets show that DPAA outperforms state-of-the-art popularity bias correction methods for GNN-based CF.

Generative recommendation (GeneRec) has introduced a new paradigm that represents items as discrete semantic tokens and predicts items in a generative manner. Despite its strong performance across multiple recommendation tasks, existing GeneRec approaches still suffer from severe popularity bias and may even exacerbate it. In this work, we conduct a comprehensive empirical analysis to uncover the root causes of this phenomenon, yielding two core insights: 1) imbalanced tokenization inherits and can further amplify popularity bias from historical item interactions; 2) current training procedures disproportionately favor popular tokens while neglecting semantic relationships among tokens, thereby intensifying popularity bias. Building on these insights, we propose CRAB, a post-hoc debiasing strategy for GeneRec that alleviates popularity bias by mitigating frequency imbalance among semantic tokens. Specifically, given a well-trained model, we first rebalance the codebook by splitting over-popular tokens while preserving their hierarchical semantic structure. Based on the adjusted codebook, we further introduce a tree-structured regularizer to enhance semantic consistency, encouraging more informative representations for unpopular tokens during training. Experiments on real-world datasets demonstrate that CRAB significantly improves recommendation performance by effectively alleviating popularity bias.

Popularity bias is a pervasive problem in recommender systems, where recommendations disproportionately favor popular items. This not only results in "rich-get-richer" dynamics and a homogenization of visible content, but can also lead to misalignment of recommendations with individual users' preferences for popular or niche content. This work studies popularity bias through the lens of user-recommender alignment. To this end, we introduce Popularity Quantile Calibration, a measurement framework that quantifies misalignment between a user's historical popularity preference and the popularity of their recommendations. Building on this notion of popularity alignment, we propose SPREE, an inference-time mitigation method for sequential recommenders based on activation steering. SPREE identifies a popularity direction in representation space and adaptively steers model activations based on an estimate of each user's personal popularity bias, allowing both the direction and magnitude of steering to vary across users. Unlike global debiasing approaches, SPREE explicitly targets alignment rather than uniformly reducing popularity. Experiments across multiple datasets show that SPREE consistently improves user-level popularity alignment while preserving recommendation quality.

Online reviews and recommendation systems help users navigate overwhelming choice, but they are vulnerable to self-reinforcing distortions. This paper examines how a single malicious reviewer can exploit popularity-biased rating dynamics and whether behavioral heterogeneity in user responses can reduce the damage. We develop a minimal agent-based model in which users choose what to rate partly on the basis of currently displayed averages. We compare broad attacks that perturb many items with sparse attacks that selectively boost low-quality items and suppress high-quality items. Additional analyses not shown here indicate that sparse attacks are substantially more harmful than broad attacks because they better exploit popularity-based exposure. The main text then focuses on sparse attacks and asks how their effects change as the fraction of contrarian users increases. Three results stand out. First, attack-induced damage is strongest when prior honest reviews are scarce, revealing a transition from a fragile low-information regime to a more robust high-information regime. Second, sparse attacks are especially effective at artificially promoting low-quality items. Third, moderate contrarian diversity partially buffers these distortions, primarily by suppressing the rise of low-quality items rather than fully restoring high-quality items to the top. The findings suggest that recommendation robustness depends not only on attack detection and predictive accuracy, but also on review density, popularity feedback, and user response heterogeneity.

Recommender systems shape individual choices through feedback loops in which user behavior and algorithmic recommendations coevolve over time. The systemic effects of these loops remain poorly understood, in part due to unrealistic assumptions in existing simulation studies. We propose a feedback-loop model that captures implicit feedback, periodic retraining, probabilistic adoption of recommendations, and heterogeneous recommender systems. We apply the framework on online retail and music streaming data and analyze systemic effects of the feedback loop. We find that increasing recommender adoption may lead to a progressive diversification of individual consumption, while collective demand is redistributed in model- and domain-dependent ways, often amplifying popularity concentration. Temporal analyses further reveal that apparent increases in individual diversity observed in static evaluations are illusory: when adoption is fixed and time unfolds, individual diversity consistently decreases across all models. Our results highlight the need to move beyond static evaluations and explicitly account for feedback-loop dynamics when designing recommender systems.

Large language models (LLMs) often produce incorrect answers with high confidence, yet the factors associated with such overconfidence remain insufficiently understood. We study this problem through the lens of knowledge popularity. Using entity-centric factual QA derived from Wikidata triplets, we characterize popularity through question entity popularity, answer popularity, and question-answer co-occurrence. We find two consistent patterns. First, hallucinated answers are far from random: compared with ground-truth answers, they tend to be more popular or more frequently associated with the question entity. Second, confidence is strongly tied to the popularity of generated answers: even among incorrect predictions, more popular answers or those with higher question-answer co-occurrence receive higher confidence. Together, these findings suggest that popular but wrong alternatives may contribute to overconfidence. We further show that popularity-related signals can mitigate overconfidence and improve overall confidence estimation. Across six models and three datasets, incorporating knowledge popularity reduces average confidence on incorrect answers from 0.765 to 0.254 and overall ECE from 0.356 to 0.050, while improving Alignment from 77.08% to 83.72%.