Index 2024 Talk: How Uber Eats Built a Recommendation System Using Two Tower Embeddings

Поделиться
HTML-код
  • Опубликовано: 23 янв 2025

Комментарии • 3

  • @marcgrab
    @marcgrab 2 месяца назад +1

    Very good talk. You can learn a lot.

  • @jianglonghe5188
    @jianglonghe5188 Месяц назад

    10:36 i think it should be O(q+m) -> O(q), it is not originally quadratic

    • @ChuanChihChou
      @ChuanChihChou 28 дней назад

      I think he meant that if the model makes pairwise predictions (i.e. given a query and an item, predicts say the probability that the user will order the item if shown) then we need to make q*M inferences given q queries and M items.