Recommender Popularity Bias: The Rich-Get-Richer Feedback Loop

What Recommender Popularity Bias Is

Recommender popularity bias refers to a phenomenon where popular items become even more prominent due to positive feedback loops. This occurs when the system recommends the most popular items, which then attract more users and further increase their popularity.

This bias can be observed in various real-world scenarios, such as social media feeds, music streaming services, and e-commerce platforms.

Why It Happens

The rich-get-richer feedback loop is driven by the fact that popular items are more likely to be recommended because they have a higher historical engagement rate. This leads to an increase in their visibility, which in turn attracts even more users and recommendations.

This cycle can result in a concentration of user attention on a small set of highly popular items, potentially neglecting less well-known but equally valuable options.

Real-World Examples

A classic example is the recommendation system for music streaming services like Spotify. If an artist has a large number of listens and shares, they are more likely to be recommended to new users, leading to even higher popularity.

In e-commerce platforms, popular products often appear at the top of search results or recommendations, reinforcing their visibility and sales.

Mitigating Popularity Bias

To mitigate this bias, recommender systems can implement strategies such as diversification techniques that ensure a mix of popular and less-known items are recommended. Content-based filtering, collaborative filtering with exploration mechanisms, and diversity-aware algorithms are some approaches used to combat the rich-get-richer effect.

Exploration strategies like Thompson sampling or epsilon-greedy methods help balance between exploiting known popular items and exploring new options, thereby reducing the concentration of user attention on a few highly popular items.

Frequently asked questions

Try it live

Everything above runs in your browser — open Recommender Popularity Bias (2D): The Rich-Get-Richer Feedback Loop and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.