Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
80 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
7 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

The Unfairness of Popularity Bias in Recommendation (1907.13286v3)

Published 31 Jul 2019 in cs.IR

Abstract: Recommender systems are known to suffer from the popularity bias problem: popular (i.e. frequently rated) items get a lot of exposure while less popular ones are under-represented in the recommendations. Research in this area has been mainly focusing on finding ways to tackle this issue by increasing the number of recommended long-tail items or otherwise the overall catalog coverage. In this paper, however, we look at this problem from the users' perspective: we want to see how popularity bias causes the recommendations to deviate from what the user expects to get from the recommender system. We define three different groups of users according to their interest in popular items (Niche, Diverse and Blockbuster-focused) and show the impact of popularity bias on the users in each group. Our experimental results on a movie dataset show that in many recommendation algorithms the recommendations the users get are extremely concentrated on popular items even if a user is interested in long-tail and non-popular items showing an extreme bias disparity.

The Unfairness of Popularity Bias in Recommendation

The paper, "The Unfairness of Popularity Bias in Recommendation," addresses a critical issue often overlooked in recommender systems: the bias towards popular items. The authors focus on the discrepancy between user expectations and actual recommendations, aiming to identify how popularity bias affects different user segments.

Recommender systems have become ubiquitous in various domains, from movies to music and online dating. The primary purpose of these systems is to assist users in discovering relevant items. Most widely used recommendation algorithms, such as collaborative filtering, inherently suffer from a popularity bias. This bias results in frequently rated items receiving excessive exposure, overshadowing less popular, long-tail items. While long-tail items are crucial for capturing the richness of user preferences and providing a fair opportunity for diverse products, most algorithms tend to exacerbate this imbalance.

The researchers propose a novel perspective by categorizing users into three distinct groups based on their interest in popular items: Niche, Diverse, and Blockbuster-focused. Their experimentation with the MovieLens 1M dataset reveals that recommendation algorithms tend to over-concentrate their suggestions on popular items, regardless of user preference profiles. Notably, niche users, who have minimal interest in popular items, are most adversely affected by this bias. This observation suggests that algorithms generally fail to tailor recommendations to meet the expectations of users whose preferences are inclined towards non-popular items.

Several recommendation algorithms were analyzed, including User KNN, Item KNN, SVD++, and Biased Matrix Factorization. The results indicated significant discrepancies in how these algorithms handle popularity bias. For example, algorithms like Biased MF and SVD++ better align with user expectations about the popularity of recommendations, while algorithms based on straightforward popularity metrics exacerbate bias.

In terms of implications, the findings suggest a need for developing algorithms that account for user preference toward item popularity. Addressing this bias can enhance user satisfaction, especially for niche users who contribute significantly to systems with richer interaction data. Future developments in AI could focus on evaluating algorithms not solely based on accuracy but also on their ability to meet diverse user expectations in terms of item popularity.

To conclude, this paper highlights the importance of considering user-specific biases in recommendation systems. It challenges existing systems to broaden their focus beyond majority interests, thus improving fairness and diversity in recommendations. Future research might explore integrating content-based approaches or hybrid methods to mitigate popularity bias more effectively across various domains.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Himan Abdollahpouri (25 papers)
  2. Masoud Mansoury (27 papers)
  3. Robin Burke (40 papers)
  4. Bamshad Mobasher (34 papers)
Citations (205)