数据集:
sileod/movie_recommendation
语言:
en计算机处理:
monolingual大小:
n<1K语言创建人:
crowdsourced批注创建人:
expert-generated源数据集:
original数字对象标识符:
10.57967/hf/0257许可:
apache-2.0We showed that pretrained large language models can act as a recommender system, and compare few-shot learning results to matrix factorization baselines. This is the BIG-Bench version of our language-based movie recommendation dataset.
https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/movie_recommendation
GPT-2 has a 48.8% accuracy, chance is 25%. Human accuracy is 60.4%.
@InProceedings{sileodreclm22, author="Sileo, Damien and Vossen, Wout and Raymaekers, Robbe", editor="Hagen, Matthias and Verberne, Suzan and Macdonald, Craig and Seifert, Christin and Balog, Krisztian and N{\o}rv{\aa}g, Kjetil and Setty, Vinay", title="Zero-Shot Recommendation as Language Modeling", booktitle="Advances in Information Retrieval", year="2022", publisher="Springer International Publishing", address="Cham", pages="223--230", isbn="978-3-030-99739-7" }