M. Mei, O. Bembom, и A. Ehmann. Proceedings of the 32nd ACM Conference on User Modeling, Adaptation and Personalization, стр. 195–200. New York, NY, USA, Association for Computing Machinery, (22.06.2024)
DOI: 10.1145/3627043.3659553
Аннотация
Next-item recommender systems are often trained using only positive feedback with randomly-sampled negative feedback. We show the benefits of using real negative feedback both as inputs into the user sequence and also as negative targets for training a next-song recommender system for internet radio. In particular, using explicit negative samples during training helps reduce training time by ∼ 60% while also improving test accuracy by 6%; adding user skips as additional inputs also can considerably increase user coverage alongside improving accuracy. We test the impact of using a large number of random negative samples to capture a ‘harder’ one and find that the test accuracy increases with more randomly-sampled negatives, but only to a point. Too many random negatives leads to false negatives that limits the lift, which is still lower than if using true negative feedback. We also find that the test accuracy is fairly robust with respect to the proportion of different feedback types, and compare the learned embeddings for different feedback types.
Описание
Negative Feedback for Music Personalization | Proceedings of the 32nd ACM Conference on User Modeling, Adaptation and Personalization
%0 Conference Paper
%1 Mei2024
%A Mei, M. Jeffrey
%A Bembom, Oliver
%A Ehmann, Andreas F.
%B Proceedings of the 32nd ACM Conference on User Modeling, Adaptation and Personalization
%C New York, NY, USA
%D 2024
%I Association for Computing Machinery
%K music-recommendation negative-feedback pandora umap2024
%P 195–200
%R 10.1145/3627043.3659553
%T Negative Feedback for Music Personalization
%U https://doi.org/10.1145/3627043.3659553
%X Next-item recommender systems are often trained using only positive feedback with randomly-sampled negative feedback. We show the benefits of using real negative feedback both as inputs into the user sequence and also as negative targets for training a next-song recommender system for internet radio. In particular, using explicit negative samples during training helps reduce training time by ∼ 60% while also improving test accuracy by 6%; adding user skips as additional inputs also can considerably increase user coverage alongside improving accuracy. We test the impact of using a large number of random negative samples to capture a ‘harder’ one and find that the test accuracy increases with more randomly-sampled negatives, but only to a point. Too many random negatives leads to false negatives that limits the lift, which is still lower than if using true negative feedback. We also find that the test accuracy is fairly robust with respect to the proportion of different feedback types, and compare the learned embeddings for different feedback types.
%@ 9798400704338
@inproceedings{Mei2024,
abstract = {Next-item recommender systems are often trained using only positive feedback with randomly-sampled negative feedback. We show the benefits of using real negative feedback both as inputs into the user sequence and also as negative targets for training a next-song recommender system for internet radio. In particular, using explicit negative samples during training helps reduce training time by ∼ 60% while also improving test accuracy by 6%; adding user skips as additional inputs also can considerably increase user coverage alongside improving accuracy. We test the impact of using a large number of random negative samples to capture a ‘harder’ one and find that the test accuracy increases with more randomly-sampled negatives, but only to a point. Too many random negatives leads to false negatives that limits the lift, which is still lower than if using true negative feedback. We also find that the test accuracy is fairly robust with respect to the proportion of different feedback types, and compare the learned embeddings for different feedback types.},
added-at = {2024-07-01T16:57:10.000+0200},
address = {New York, NY, USA},
author = {Mei, M. Jeffrey and Bembom, Oliver and Ehmann, Andreas F.},
biburl = {https://www.bibsonomy.org/bibtex/24ca47949bf510aa4a42b9ea4d4e0072f/brusilovsky},
booktitle = {Proceedings of the 32nd ACM Conference on User Modeling, Adaptation and Personalization},
day = 22,
description = {Negative Feedback for Music Personalization | Proceedings of the 32nd ACM Conference on User Modeling, Adaptation and Personalization},
doi = {10.1145/3627043.3659553},
interhash = {db4fe32141e6c778d99d1c73ce18f213},
intrahash = {4ca47949bf510aa4a42b9ea4d4e0072f},
isbn = {9798400704338},
keywords = {music-recommendation negative-feedback pandora umap2024},
location = {Cagliari, Italy},
month = {6},
pages = {195–200},
publisher = {Association for Computing Machinery},
series = {UMAP '24},
timestamp = {2024-07-01T16:57:10.000+0200},
title = {Negative Feedback for Music Personalization},
url = {https://doi.org/10.1145/3627043.3659553},
year = 2024
}