Millised õiglusprobleemid tekivad soovitustes?

In recommendation systems, fairness problems are related to the unfair distribution of recommendations among users or objects. Main problems:

• Popularity bias: the system tends to recommend popular objects more often, reducing the visibility of less popular but potentially relevant objects.

• Imbalance between user groups: some user groups (e.g., by age, gender, region) may receive lower quality or less diverse recommendations.

• Imbalance between content creators: authors with less recognition or from minorities may receive fewer impressions and interactions.

• Feedback and closed loop: recommendations influence user behavior, which can reinforce existing biases and injustices.

To address these problems, methods such as balancing, regularization, and specific fairness metrics are used during model training.