Personalized Ranking Model

  • 최적화 방식이 세 가지로 나뉘어짐
    • pointwise
      • Pointwise approaches considers a single interaction at a time and train a classifier or a regressor to predict individual preferences.
      • 예시) matrix factorization, AutoRec
    • pairwise
    • listwise
      • Listwise approaches approximate the ordering of the entire list of items, for example, direct optimizing the ranking measures such as Normalized Discounted Cumulative Gain (Normalized Discounted Cumulative Gain).
      • pointwise 나 pairwise 방식보다 복잡하며 계산 비용이 높다.

B) Related

C) References