Open Access System for Information Sharing

Login Library

 

Thesis
Cited 0 time in webofscience Cited 0 time in scopus
Metadata Downloads

Next Basket Recommendation with Hypergraph Neural Network

Title
Next Basket Recommendation with Hypergraph Neural Network
Authors
정은아
Date Issued
2024
Publisher
포항공과대학교
Abstract
본 논문에서는 아이템 조합의 상관관계를 모델링하는 하이퍼그래프 신경망(HyperGraph Neural Network)기반모델을 새롭게 제안한다. 장바구니 안 아이템들을 하이퍼엣지로 표현하여, 장바구니 단위로 나타나는 사용자의 선호를 학습한다. 또한, 바구니에 여러 번 등장하는 아이템 조합도 하이퍼엣지로 연결하여 서로 관련 있는 아이템을 추천하도록 한다. 마지막으로, 장바구니 추천 상황에 맞게 실제로 소비한 장바구니와 아이템이 소비하지 않은 장바구니와 아이템보다 더 높은 점수를 갖도록 설계한 손실 함수를 제안하여 모델 학습을 진행한다. 우리가 제안한 다음 장바구니추천모델은 2개의 실제 데이터 세트로 수행된 실험에서 기존 모델들보다 더 높은 성능을 달성하였다.
Next-Basket Recommendation(NBR) aims to recommend the set of items that a user is most likely to purchase together next, based on their purchase history of baskets. Existing methods in NBR do not directly consider the correlations among items within the basket, resulting in suboptimal performance due to inadequate representation of items. In this paper, we propose a novel model based on HyperGraph Neural Network (HGNN) to model the correlations among item combinations. We connect the items within the basket as hyperedges, enabling the learning of user preferences at the basket level. Additionally, we connect item combinations that appear multiple times in baskets as hyperedges to recommend related items. Finally, we propose a loss function designed to prioritize items and baskets that have been consumed over those that have not been consumed, which is tailored to the next-basket recommendation scenario. Our proposed next-basket recommendation model outperforms existing models in experiments conducted on two real-world datasets.
URI
http://postech.dcollection.net/common/orgView/200000807535
https://oasis.postech.ac.kr/handle/2014.oak/124078
Article Type
Thesis
Files in This Item:
There are no files associated with this item.

qr_code

  • mendeley

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Views & Downloads

Browse