hard20 pts · first answer countsrecommender-systemsbias

Retraining on your own recommendations

Your recommender is retrained daily on clicks from items it chose to show. Over months, diversity falls and new items rarely get exposure. What is the core problem and fix?

A.No problem, since clicks are unbiased labelsB.Train only on items that were never shownC.Increase the embedding size of the item towerD.A feedback loop; add exploration and correct for exposure bias

Discussion 0

Join free to post in the discussion.

Retraining on your own recommendations

Discussion 0

Related questions