GPT4Rec - 리뷰

GPT4Rec: A Generative Framework for Personalized Recommendation and User Interests Interpretation

논문

• https://arxiv.org/pdf/2304.03879.pdf

코드

• 공개코드 없음

참고자료

• (논문 저자 발표자료) https://www.youtube.com/watch?v=iybsVdasiVI&ab_channel=JinmingLi

---

문제의식

• current models commonly treat items as mere IDs
  • incapable of fully leveraging the content information of items
  • unable to accommodate changing and growing item inventories

아이디어

• learn the user representation in the language space from the item interaction sequence

  • 아이템 ID가 아닌 콘텐츠가 담긴 아이템 Title 사용

• generates multiple queries that represent user interests

  • Fine-tuned GPT-2 model 사용 : last item title 예측

• [프로세스]
  

  1. given prompt with item titles(sequence), GPT generates queries(다음에 구매할만한 아이템 타이틀)

     • multi-query generation (that represent user’s interests) with beam search

  2. retrieve items by search engine

     • adopt BM25(search engine)
     • measure the similarity on the language space
     • (자체 제안) ranking based strategy to combine searching results of each query

---

모델구조

예시