An embedding and interactions learning approach for ID feature in deep recommender system

作者：

Highlights：

• Recommendation models use the same method to embedding all features including IDs.

• The embedding and learning approach reduces accuracy and generalization ability.

• Embedding ID tags independently and learning their interactions efficiently.

• Optimizing prediction accuracy and convergence time of deep recommendation models.

摘要

•Recommendation models use the same method to embedding all features including IDs.•The embedding and learning approach reduces accuracy and generalization ability.•Embedding ID tags independently and learning their interactions efficiently.•Optimizing prediction accuracy and convergence time of deep recommendation models.

论文关键词：Recommender system,Neural networks,Discrete feature,Feature embedding

论文评审过程：Received 20 May 2022, Revised 10 July 2022, Accepted 3 August 2022, Available online 9 August 2022, Version of Record 17 August 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.118425