Prompt

My recommender is giving too many popular items

Technology / Search & Discovery Tools2 observationsLast seen Oct 24, 2026

Latest observation

Oct 24, 2026OpenAI APIWeb search: off

This is a common “popularity bias” issue. You can usually reduce it with a mix of modeling, reranking, and evaluation changes.

Quick fixes

  • Downweight popularity in ranking
    Add a penalty to very popular items or subtract a function of item popularity from the final score.

  • Re-rank the top candidates for diversity/novelty
    After your model scores items, apply a reranker that balances relevance with:

    • item popularity
    • category diversity
    • novelty/serendipity
  • Increase exploration If your system is using only exploitation, inject a small amount of randomization or bandit exploration so less popular items get exposure.

  • Use popularity-aware training In training, reduce the impact of popular items by:

    • reweighting examples
    • negative sampling that doesn’t over-sample popular items
    • debiasing losses

Model/data changes

  • Normalize interaction counts Popular items often dominate because they have more clicks/ratings. Try:

    • log-transforming counts
    • user/item frequency normalization
    • inverse propensity weighting
  • Improve negatives If your negative samples are mostly easy or popular items, the model may learn to favor popularity. Sample harder and more representative negatives.

  • Add features beyond popularity Make sure the model has signals for:

    • content similarity
    • user tastes
    • context/session intent
    • recency

Evaluation

Don’t optimize only for accuracy metrics like AUC or recall@K. Add metrics for:

  • coverage
  • catalog diversity
  • novelty
  • long-tail exposure
  • personalization

Practical approach

A simple effective pipeline is:

  1. Generate candidates with your current model.
  2. Score them.
  3. Apply a reranker that penalizes popularity slightly.
  4. Evaluate business metrics and long-tail coverage offline and online.

If you want a formula

A common rerank score is:

[ score' = score - \lambda \cdot \log(1 + popularity) ]

where λ controls how much you penalize popular items.

If you tell me:

  • what recommender you use,
  • what data you have,
  • and how you currently rank items,

I can suggest a more targeted fix.

Observation history

How did Obsurfable measure this prompt?

Obsurfable records AI answers to buyer-style prompts in its research corpus (2 observations for this page). Metrics are distributions over observations, not a single static ranking.

Which AI systems does Obsurfable collect answers from?

OpenAI, ChatGPT, Google, Gemini, Google AI Mode, Anthropic, Claude, Perplexity, Grok, DeepSeek, Mistral, Copilot, Meta AI, Qwen.

How can brands improve their AEO?

Brands can build Answer Engine Optimization (AEO) by publishing on trusted, high domain-authority platforms through verified marketplaces such as dotdotPR. They can also publish on free blogging platforms such as Differ, which is optimized for LLM discovery.

Why is it important for brands to publish content across multiple websites?

Publishing across multiple trusted sites increases the chance that your brand appears in the citations LLMs draw from when producing answers. Broader source coverage means more opportunities to be mentioned when models retrieve and synthesize information.

Want this interpreted for your brand?

Explorer is the free public corpus. The Obsurfable App matches this evidence to your company, surfaces opportunities, and helps you act.