Practical Content-aware Session-based Recommendation: Deep Retrieve then Shallow Rank

Yuxuan Lei, Xiaolong Chen, Defu Lian, Peiyan Zhang, Jianxun Lian, Chaozhuo Li, Xing Xie

Keywords: Session-based Recommender Systems, Language Models, Tree Boosting System, KDDCup 2023

TL;DR: This paper presents the solution of our team unirec in the KDD Cup 2023 Multilingual Recommendation Challenge

Abstract:

This paper presents the solution of our team unirec in the KDD Cup 2023 Multilingual Recommendation Challenge.
The goal of the competition is to explore ways to improve session-based recommendation in real-world multilingual and imbalanced scenarios. Our method comprises a two-stage retrieval-then-rank strategy. In the first stage, advanced deep single models are used to score the full set of items, enabling us to obtain a smaller candidates set along with the corresponding session-item score features. In the second stage, we employ the shallow but powerful XGBoost algorithm for ranking to derive the final recommendation results. Our method ranks 3rd place in the final leaderboard of Task1. Our implementation using the recommendation library RecStudio and UniRec is publicly available at this link: https://gitlab.aicrowd.com/CXL/unirec-task1-amazon-kddcup-2023

Submission Number: 15