Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reducing Popularity Influence by Addressing Position Bias

Andrii Dzhoha, Alexey Kurennoy, Vladimir Vlasov, Marjan Celikik

TL;DR

The paper tackles position bias in recommender systems, showing that debiasing can spread visibility across an assortment and reduce popularity skew even when short-term ranking gains are uncertain. It develops a theoretical view of how a position-induced feedback loop biases item popularity and demonstrates a position-aware learning approach that disentangles position from true relevance. Across offline and online evaluations on a large e-commerce platform, the method reduced skew and improved assortment utilization without degrading engagement or financial metrics, supporting long-term fairness and partnerships. The work highlights the practical value of debiasing for fairer rankings and broader ecosystem benefits for customers and the business.

Abstract

Position bias poses a persistent challenge in recommender systems, with much of the existing research focusing on refining ranking relevance and driving user engagement. However, in practical applications, the mitigation of position bias does not always result in detectable short-term improvements in ranking relevance. This paper provides an alternative, practically useful view of what position bias reduction methods can achieve. It demonstrates that position debiasing can spread visibility and interactions more evenly across the assortment, effectively reducing a skew in the popularity of items induced by the position bias through a feedback loop. We offer an explanation of how position bias affects item popularity. This includes an illustrative model of the item popularity histogram and the effect of the position bias on its skewness. Through offline and online experiments on our large-scale e-commerce platform, we show that position debiasing can significantly improve assortment utilization, without any degradation in user engagement or financial metrics. This makes the ranking fairer and helps attract more partners or content providers, benefiting the customers and the business in the long term.

Reducing Popularity Influence by Addressing Position Bias

TL;DR

The paper tackles position bias in recommender systems, showing that debiasing can spread visibility across an assortment and reduce popularity skew even when short-term ranking gains are uncertain. It develops a theoretical view of how a position-induced feedback loop biases item popularity and demonstrates a position-aware learning approach that disentangles position from true relevance. Across offline and online evaluations on a large e-commerce platform, the method reduced skew and improved assortment utilization without degrading engagement or financial metrics, supporting long-term fairness and partnerships. The work highlights the practical value of debiasing for fairer rankings and broader ecosystem benefits for customers and the business.

Abstract

Position bias poses a persistent challenge in recommender systems, with much of the existing research focusing on refining ranking relevance and driving user engagement. However, in practical applications, the mitigation of position bias does not always result in detectable short-term improvements in ranking relevance. This paper provides an alternative, practically useful view of what position bias reduction methods can achieve. It demonstrates that position debiasing can spread visibility and interactions more evenly across the assortment, effectively reducing a skew in the popularity of items induced by the position bias through a feedback loop. We offer an explanation of how position bias affects item popularity. This includes an illustrative model of the item popularity histogram and the effect of the position bias on its skewness. Through offline and online experiments on our large-scale e-commerce platform, we show that position debiasing can significantly improve assortment utilization, without any degradation in user engagement or financial metrics. This makes the ranking fairer and helps attract more partners or content providers, benefiting the customers and the business in the long term.

Paper Structure

This paper contains 13 sections, 11 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Recommender system and popularity of items
  3. Interaction model without position bias
  4. Interaction model within the feedback loop
  5. Position bias
  6. Interaction model
  7. Skew in the distribution of popularity of items
  8. Sample selection bias
  9. Influence on popularity
  10. Experiments and results
  11. Offline experiments
  12. Online experiment
  13. Conclusion

Figures (3)

  • Figure 1: Distribution of popularity of items. (a) Example of the histogram. (b) Discretized exponential distribution. The distribution with parameter $\lambda'$ has a stronger bias towards favoring more popular items illustrated by the expected values $\mu$ and $\mu'$ of the rank.
  • Figure 2: Model's life cycle within the feedback loop.
  • Figure 3: Position-aware learning.