Leveraging Watch-time Feedback for Short-Video Recommendations: A Causal Labeling Framework ...
With the proliferation of short video applications, the significance of short video recommendations has vastly increased. Unlike other recommendation scenarios, short video recommendation systems heavily rely on feedback from watch time. Existing approaches simply treat watch time as a direct label,...
Main Authors: | , , , , , , , , |
---|---|
Format: | Text |
Language: | unknown |
Published: |
arXiv
2023
|
Subjects: | |
Online Access: | https://dx.doi.org/10.48550/arxiv.2306.17426 https://arxiv.org/abs/2306.17426 |
id |
ftdatacite:10.48550/arxiv.2306.17426 |
---|---|
record_format |
openpolar |
spelling |
ftdatacite:10.48550/arxiv.2306.17426 2023-10-01T03:55:39+02:00 Leveraging Watch-time Feedback for Short-Video Recommendations: A Causal Labeling Framework ... Zhang, Yang Bai, Yimeng Chang, Jianxin Zang, Xiaoxue Lu, Song Lu, Jing Feng, Fuli Niu, Yanan Song, Yang 2023 https://dx.doi.org/10.48550/arxiv.2306.17426 https://arxiv.org/abs/2306.17426 unknown arXiv https://dx.doi.org/10.1145/3583780.3615483 arXiv.org perpetual, non-exclusive license http://arxiv.org/licenses/nonexclusive-distrib/1.0/ Information Retrieval cs.IR FOS Computer and information sciences H.3.3; H.3.5 ScholarlyArticle Article article-journal Text 2023 ftdatacite https://doi.org/10.48550/arxiv.2306.1742610.1145/3583780.3615483 2023-09-04T15:17:58Z With the proliferation of short video applications, the significance of short video recommendations has vastly increased. Unlike other recommendation scenarios, short video recommendation systems heavily rely on feedback from watch time. Existing approaches simply treat watch time as a direct label, failing to effectively harness its extensive semantics and introduce bias, thereby limiting the potential for modeling user interests based on watch time. To overcome this challenge, we propose a framework named Debiased Multiple-semantics-extracting Labeling(DML). DML constructs labels that encompass various semantics by utilizing quantiles derived from the distribution of watch time, prioritizing relative order rather than absolute label values. This approach facilitates easier model learning while aligning with the ranking objective of recommendations. Furthermore, we introduce a method inspired by causal adjustment to refine label definitions, thereby directly mitigating bias at the label level. We ... : 8 pages, 4 figures ... Text DML DataCite Metadata Store (German National Library of Science and Technology) |
institution |
Open Polar |
collection |
DataCite Metadata Store (German National Library of Science and Technology) |
op_collection_id |
ftdatacite |
language |
unknown |
topic |
Information Retrieval cs.IR FOS Computer and information sciences H.3.3; H.3.5 |
spellingShingle |
Information Retrieval cs.IR FOS Computer and information sciences H.3.3; H.3.5 Zhang, Yang Bai, Yimeng Chang, Jianxin Zang, Xiaoxue Lu, Song Lu, Jing Feng, Fuli Niu, Yanan Song, Yang Leveraging Watch-time Feedback for Short-Video Recommendations: A Causal Labeling Framework ... |
topic_facet |
Information Retrieval cs.IR FOS Computer and information sciences H.3.3; H.3.5 |
description |
With the proliferation of short video applications, the significance of short video recommendations has vastly increased. Unlike other recommendation scenarios, short video recommendation systems heavily rely on feedback from watch time. Existing approaches simply treat watch time as a direct label, failing to effectively harness its extensive semantics and introduce bias, thereby limiting the potential for modeling user interests based on watch time. To overcome this challenge, we propose a framework named Debiased Multiple-semantics-extracting Labeling(DML). DML constructs labels that encompass various semantics by utilizing quantiles derived from the distribution of watch time, prioritizing relative order rather than absolute label values. This approach facilitates easier model learning while aligning with the ranking objective of recommendations. Furthermore, we introduce a method inspired by causal adjustment to refine label definitions, thereby directly mitigating bias at the label level. We ... : 8 pages, 4 figures ... |
format |
Text |
author |
Zhang, Yang Bai, Yimeng Chang, Jianxin Zang, Xiaoxue Lu, Song Lu, Jing Feng, Fuli Niu, Yanan Song, Yang |
author_facet |
Zhang, Yang Bai, Yimeng Chang, Jianxin Zang, Xiaoxue Lu, Song Lu, Jing Feng, Fuli Niu, Yanan Song, Yang |
author_sort |
Zhang, Yang |
title |
Leveraging Watch-time Feedback for Short-Video Recommendations: A Causal Labeling Framework ... |
title_short |
Leveraging Watch-time Feedback for Short-Video Recommendations: A Causal Labeling Framework ... |
title_full |
Leveraging Watch-time Feedback for Short-Video Recommendations: A Causal Labeling Framework ... |
title_fullStr |
Leveraging Watch-time Feedback for Short-Video Recommendations: A Causal Labeling Framework ... |
title_full_unstemmed |
Leveraging Watch-time Feedback for Short-Video Recommendations: A Causal Labeling Framework ... |
title_sort |
leveraging watch-time feedback for short-video recommendations: a causal labeling framework ... |
publisher |
arXiv |
publishDate |
2023 |
url |
https://dx.doi.org/10.48550/arxiv.2306.17426 https://arxiv.org/abs/2306.17426 |
genre |
DML |
genre_facet |
DML |
op_relation |
https://dx.doi.org/10.1145/3583780.3615483 |
op_rights |
arXiv.org perpetual, non-exclusive license http://arxiv.org/licenses/nonexclusive-distrib/1.0/ |
op_doi |
https://doi.org/10.48550/arxiv.2306.1742610.1145/3583780.3615483 |
_version_ |
1778524267610112000 |