A Two-stage Ranking Framework for Multilingual Recommendation(Team:AIDA)

Shijie Liu, Chenliang Zhang, Xutao Han, Wen Wu, Wei Zhang

Keywords: mutilingual, recommendation, transfer learning

Abstract: In the KDD Cup 2023 multilingual recommendation challenge, we proposed a ranking framework, which consists of two main stage: recall and ranking. In the recall stage, we use a carefully-designed co-occurrence matrix for single-hop and multi-hop recall of candidate items. Moreover, an MLP model initialized with ``title'' information is also used in this stage. In the ranking stage, we designed many features and additional transfer features for task 2’s transfer learning.Then we use the XGBoost model for ranking. Our solutions rank 6th in task 1 and 4th in task 2. The code is available at https://github.com/karrich/KDD-CUP-2023-solution.

Submission Number: 3