Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Embedding Cultural Diversity in Prototype-based Recommender Systems

Armin Moradi, Nicola Neophytou, Florian Carichon, Golnoosh Farnadi

TL;DR

The paper tackles popularity bias in culturally rich recommender systems by focusing on demographic biases within prototype-based matrix factorization. It introduces two in-processing mechanisms for ProtoMF: Prototype K-filtering to emphasize locally relevant prototypes and a Prototype-Distributing Regularizer to encourage a uniform distribution of prototypes across the embedding space, using country of origin as a cultural cue. Across four datasets, the approach yields tangible fairness gains—reducing long-tail and underrepresented-item rankings—while maintaining or improving recommendation quality (e.g., HitRate@10 improvements). Additionally, the method enhances explainability by producing more culturally diverse prototype associations. Limitations include reliance on country-of-origin as a proxy for culture, potential generalizability constraints from data filtering, and a need for deeper theoretical and explainability analyses; future work could broaden cultural factors and provide formal fairness guarantees.

Abstract

Popularity bias in recommender systems can increase cultural overrepresentation by favoring norms from dominant cultures and marginalizing underrepresented groups. This issue is critical for platforms offering cultural products, as they influence consumption patterns and human perceptions. In this work, we address popularity bias by identifying demographic biases within prototype-based matrix factorization methods. Using the country of origin as a proxy for cultural identity, we link this demographic attribute to popularity bias by refining the embedding space learning process. First, we propose filtering out irrelevant prototypes to improve representativity. Second, we introduce a regularization technique to enforce a uniform distribution of prototypes within the embedding space. Across four datasets, our results demonstrate a 27\% reduction in the average rank of long-tail items and a 2\% reduction in the average rank of items from underrepresented countries. Additionally, our model achieves a 2\% improvement in HitRatio@10 compared to the state-of-the-art, highlighting that fairness is enhanced without compromising recommendation quality. Moreover, the distribution of prototypes leads to more inclusive explanations by better aligning items with diverse prototypes.

Embedding Cultural Diversity in Prototype-based Recommender Systems

TL;DR

The paper tackles popularity bias in culturally rich recommender systems by focusing on demographic biases within prototype-based matrix factorization. It introduces two in-processing mechanisms for ProtoMF: Prototype K-filtering to emphasize locally relevant prototypes and a Prototype-Distributing Regularizer to encourage a uniform distribution of prototypes across the embedding space, using country of origin as a cultural cue. Across four datasets, the approach yields tangible fairness gains—reducing long-tail and underrepresented-item rankings—while maintaining or improving recommendation quality (e.g., HitRate@10 improvements). Additionally, the method enhances explainability by producing more culturally diverse prototype associations. Limitations include reliance on country-of-origin as a proxy for culture, potential generalizability constraints from data filtering, and a need for deeper theoretical and explainability analyses; future work could broaden cultural factors and provide formal fairness guarantees.

Abstract

Popularity bias in recommender systems can increase cultural overrepresentation by favoring norms from dominant cultures and marginalizing underrepresented groups. This issue is critical for platforms offering cultural products, as they influence consumption patterns and human perceptions. In this work, we address popularity bias by identifying demographic biases within prototype-based matrix factorization methods. Using the country of origin as a proxy for cultural identity, we link this demographic attribute to popularity bias by refining the embedding space learning process. First, we propose filtering out irrelevant prototypes to improve representativity. Second, we introduce a regularization technique to enforce a uniform distribution of prototypes within the embedding space. Across four datasets, our results demonstrate a 27\% reduction in the average rank of long-tail items and a 2\% reduction in the average rank of items from underrepresented countries. Additionally, our model achieves a 2\% improvement in HitRatio@10 compared to the state-of-the-art, highlighting that fairness is enhanced without compromising recommendation quality. Moreover, the distribution of prototypes leads to more inclusive explanations by better aligning items with diverse prototypes.

Paper Structure

This paper contains 14 sections, 8 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Prototype-based Matrix Factorization
  4. Methodology
  5. Prototype K-filtering
  6. Prototype-Distributing Regularizer
  7. Evaluation Setup
  8. Datasets
  9. Baselines
  10. Metrics
  11. Results
  12. Discussion
  13. Conclusion & Future Directions
  14. Detailed Results

Figures (3)

  • Figure 1: The first group (a, b) presents the performance metrics. The second group (c, d) presents the fairness metrics. Our model with the highest HR@10 is chosen.
  • Figure 2: Results obtained for our various model configurations. All values are normalized per dataset, with higher values indicating better performance or fairness.
  • Figure 3: t-SNE visualizations of the item embedding space for (a) ProtoMF and (b) $\{\text{User-Item}\}_{\{k, \lambda\}}$ models. NL and US items are respectively from under and over-represented countries.