Large language models make sample-efficient recommender systems

This letter investigates the sample efficiency property of recommender systems enhanced by large language models. We propose a simple yet effective framework (i.e., Laser) to validate the core viewpoint - large language models make sample-efficient recommender systems - from two aspects: (1) LLMs themselves are sample-efficient recommenders; and (2) LLMs make conventional recommender systems more sample-efficient. For future work, we aim to improve the sample efficiency of LLM-based recommender systems from the following two aspects: (1) exploring effective strategy to select the few-shot training samples instead of uniformly sampling, and (2) applying Laser for downstream applications like code snippet recommendation.

The Shanghai Jiao Tong University team was partially supported by the National Natural Science Foundation of China (Grant No. 62177033). Jianghao Lin is supported by the Wu Wen Jun Honorary Doctoral Scholarship. The work was sponsored by Huawei Innovation Research Program. We thank MindSpore [3] for the partial support of this work, which is a new deep learning computing framework.

Authors and Affiliations

1. Computer Science and Technology, Shanghai Jiao Tong University, Shanghai, 200240, China

   Jianghao Lin, Rong Shan, Yong Yu & Weinan Zhang

2. Huawei Noah’s Ark Lab, Shenzhen, 518129, China

   Xinyi Dai, Bo Chen & Ruiming Tang

Corresponding author

Correspondence to Weinan Zhang.

Competing interests The authors declare that they have no competing interests or financial conflicts to disclose.

Reprints and permissions