The paper presents Ada-Retrieval, a new multi-round retrieval paradigm for recommender systems. Unlike traditional single-round models, Ada-Retrieval iteratively refines user representations to better identify potential item candidates, effectively capturing the dynamic nature of user preferences. The framework comprises two modules: the item representation adapter and the user representation adapter. It can be integrated with various models like RNNs or Transformers. Experimental results indicate significant performance enhancements with Ada-Retrieval across different datasets.
Publication date: 15 Jan 2024
Project Page: https://github.com/ll0ruc/Ada-Retrieval
Paper: https://arxiv.org/pdf/2401.06633