AutoAssign: Automatic Shared Embedding Assignment in Streaming Recommendation

Fengyi Song; Bo Chen; Xiangyu Zhao; Huifeng Guo; Ruiming Tang

doi:10.1109/ICDM54844.2022.00056

AutoAssign: Automatic Shared Embedding Assignment in Streaming Recommendation

Fengyi Song, Bo Chen, Xiangyu Zhao^*, Huifeng Guo, Ruiming Tang

^*Corresponding author for this work

Department of Data Science

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

9 Citations (Scopus)

Abstract

In streaming recommender systems, the traditional approach for handling new user IDs or item IDs is to assign randomly initialized ID embedding, leading to two practical issues: (i) Items or users with insufficient interactive data can result in suboptimal prediction performance; and (ii) The embedding of new IDs or low-frequency IDs will consistently increase the size of the embedding table, thereby consuming unnecessary memory. To this end, we propose a reinforcement learning-based Automatic Shared Embedding Assignment framework, AutoAssign. To be specific, an Identity Agent serves to (i) field-wisely represent low-frequency IDs by utilizing a small number of shared embeddings, so as to enhance the embedding initialization; and (ii) dynamically identify the ID features that need to be retained or eliminated in the embedding table. We conduct extensive experiments on three public benchmark datasets and observe that AutoAssign can significantly improve the recommendation performance by alleviating the cold-start problem. Besides, AutoAssign reduces the memory space by 20-30 %, which demonstrates the effectiveness and efficiency of our framework in practical streaming recommender systems. © 2022 IEEE.

Original language	English
Title of host publication	Proceedings - 22nd IEEE International Conference on Data Mining
Subtitle of host publication	ICDM 2022
Editors	Xingquan Zhu, Sanjay Ranka, My T. Thai, Takashi Washio, Xindong Wu
Publisher	IEEE
Pages	458-467
ISBN (Electronic)	9781665450997
ISBN (Print)	978-1-6654-5100-0
DOIs	https://doi.org/10.1109/ICDM54844.2022.00056
Publication status	Published - Nov 2022
Event	22nd IEEE International Conference on Data Mining (ICDM 2022) - Hilton Orlando, Orlando, United States Duration: 28 Nov 2022 → 1 Dec 2022 https://icdm22.cse.usf.edu/

Publication series

Name	Proceedings - IEEE International Conference on Data Mining, ICDM
ISSN (Print)	1550-4786
ISSN (Electronic)	2374-8486

Conference

Conference	22nd IEEE International Conference on Data Mining (ICDM 2022)
Country/Territory	United States
City	Orlando
Period	28/11/22 → 1/12/22
Internet address	https://icdm22.cse.usf.edu/

Bibliographical note

Full text of this publication does not contain sufficient affiliation information. With consent from the author(s) concerned, the Research Unit(s) information for this record is based on the existing academic department affiliation of the author(s).

Research Keywords

Embedding
Reinforcement Learning
Streaming Recommendations

Access Output

10.1109/ICDM54844.2022.00056

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

Song, F., Chen, B., Zhao, X., Guo, H., & Tang, R. (2022). AutoAssign: Automatic Shared Embedding Assignment in Streaming Recommendation. In X. Zhu, S. Ranka, M. T. Thai, T. Washio, & X. Wu (Eds.), Proceedings - 22nd IEEE International Conference on Data Mining: ICDM 2022 (pp. 458-467). (Proceedings - IEEE International Conference on Data Mining, ICDM). IEEE. https://doi.org/10.1109/ICDM54844.2022.00056

Song, Fengyi ; Chen, Bo ; Zhao, Xiangyu et al. / AutoAssign : Automatic Shared Embedding Assignment in Streaming Recommendation. Proceedings - 22nd IEEE International Conference on Data Mining: ICDM 2022. editor / Xingquan Zhu ; Sanjay Ranka ; My T. Thai ; Takashi Washio ; Xindong Wu. IEEE, 2022. pp. 458-467 (Proceedings - IEEE International Conference on Data Mining, ICDM).

@inproceedings{50b5d582fc584337a7f8848cf30bb7f1,

title = "AutoAssign: Automatic Shared Embedding Assignment in Streaming Recommendation",

abstract = "In streaming recommender systems, the traditional approach for handling new user IDs or item IDs is to assign randomly initialized ID embedding, leading to two practical issues: (i) Items or users with insufficient interactive data can result in suboptimal prediction performance; and (ii) The embedding of new IDs or low-frequency IDs will consistently increase the size of the embedding table, thereby consuming unnecessary memory. To this end, we propose a reinforcement learning-based Automatic Shared Embedding Assignment framework, AutoAssign. To be specific, an Identity Agent serves to (i) field-wisely represent low-frequency IDs by utilizing a small number of shared embeddings, so as to enhance the embedding initialization; and (ii) dynamically identify the ID features that need to be retained or eliminated in the embedding table. We conduct extensive experiments on three public benchmark datasets and observe that AutoAssign can significantly improve the recommendation performance by alleviating the cold-start problem. Besides, AutoAssign reduces the memory space by 20-30 %, which demonstrates the effectiveness and efficiency of our framework in practical streaming recommender systems. {\textcopyright} 2022 IEEE.",

keywords = "Embedding, Reinforcement Learning, Streaming Recommendations",

author = "Fengyi Song and Bo Chen and Xiangyu Zhao and Huifeng Guo and Ruiming Tang",

note = "Full text of this publication does not contain sufficient affiliation information. With consent from the author(s) concerned, the Research Unit(s) information for this record is based on the existing academic department affiliation of the author(s).; 22nd IEEE International Conference on Data Mining (ICDM 2022) ; Conference date: 28-11-2022 Through 01-12-2022",

year = "2022",

month = nov,

doi = "10.1109/ICDM54844.2022.00056",

language = "English",

isbn = "978-1-6654-5100-0",

series = "Proceedings - IEEE International Conference on Data Mining, ICDM",

publisher = "IEEE",

pages = "458--467",

editor = "Xingquan Zhu and Sanjay Ranka and Thai, {My T.} and Takashi Washio and Xindong Wu",

booktitle = "Proceedings - 22nd IEEE International Conference on Data Mining",

address = "United States",

url = "https://icdm22.cse.usf.edu/",

}

Song, F, Chen, B, Zhao, X, Guo, H & Tang, R 2022, AutoAssign: Automatic Shared Embedding Assignment in Streaming Recommendation. in X Zhu, S Ranka, MT Thai, T Washio & X Wu (eds), Proceedings - 22nd IEEE International Conference on Data Mining: ICDM 2022. Proceedings - IEEE International Conference on Data Mining, ICDM, IEEE, pp. 458-467, 22nd IEEE International Conference on Data Mining (ICDM 2022), Orlando, United States, 28/11/22. https://doi.org/10.1109/ICDM54844.2022.00056

AutoAssign: Automatic Shared Embedding Assignment in Streaming Recommendation. / Song, Fengyi; Chen, Bo; Zhao, Xiangyu et al.
Proceedings - 22nd IEEE International Conference on Data Mining: ICDM 2022. ed. / Xingquan Zhu; Sanjay Ranka; My T. Thai; Takashi Washio; Xindong Wu. IEEE, 2022. p. 458-467 (Proceedings - IEEE International Conference on Data Mining, ICDM).

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

TY - GEN

T1 - AutoAssign

T2 - 22nd IEEE International Conference on Data Mining (ICDM 2022)

AU - Song, Fengyi

AU - Chen, Bo

AU - Zhao, Xiangyu

AU - Guo, Huifeng

AU - Tang, Ruiming

N1 - Full text of this publication does not contain sufficient affiliation information. With consent from the author(s) concerned, the Research Unit(s) information for this record is based on the existing academic department affiliation of the author(s).

PY - 2022/11

Y1 - 2022/11

N2 - In streaming recommender systems, the traditional approach for handling new user IDs or item IDs is to assign randomly initialized ID embedding, leading to two practical issues: (i) Items or users with insufficient interactive data can result in suboptimal prediction performance; and (ii) The embedding of new IDs or low-frequency IDs will consistently increase the size of the embedding table, thereby consuming unnecessary memory. To this end, we propose a reinforcement learning-based Automatic Shared Embedding Assignment framework, AutoAssign. To be specific, an Identity Agent serves to (i) field-wisely represent low-frequency IDs by utilizing a small number of shared embeddings, so as to enhance the embedding initialization; and (ii) dynamically identify the ID features that need to be retained or eliminated in the embedding table. We conduct extensive experiments on three public benchmark datasets and observe that AutoAssign can significantly improve the recommendation performance by alleviating the cold-start problem. Besides, AutoAssign reduces the memory space by 20-30 %, which demonstrates the effectiveness and efficiency of our framework in practical streaming recommender systems. © 2022 IEEE.

AB - In streaming recommender systems, the traditional approach for handling new user IDs or item IDs is to assign randomly initialized ID embedding, leading to two practical issues: (i) Items or users with insufficient interactive data can result in suboptimal prediction performance; and (ii) The embedding of new IDs or low-frequency IDs will consistently increase the size of the embedding table, thereby consuming unnecessary memory. To this end, we propose a reinforcement learning-based Automatic Shared Embedding Assignment framework, AutoAssign. To be specific, an Identity Agent serves to (i) field-wisely represent low-frequency IDs by utilizing a small number of shared embeddings, so as to enhance the embedding initialization; and (ii) dynamically identify the ID features that need to be retained or eliminated in the embedding table. We conduct extensive experiments on three public benchmark datasets and observe that AutoAssign can significantly improve the recommendation performance by alleviating the cold-start problem. Besides, AutoAssign reduces the memory space by 20-30 %, which demonstrates the effectiveness and efficiency of our framework in practical streaming recommender systems. © 2022 IEEE.

KW - Embedding

KW - Reinforcement Learning

KW - Streaming Recommendations

UR - http://www.scopus.com/inward/record.url?scp=85147734378&partnerID=8YFLogxK

UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-85147734378&origin=recordpage

U2 - 10.1109/ICDM54844.2022.00056

DO - 10.1109/ICDM54844.2022.00056

M3 - RGC 32 - Refereed conference paper (with host publication)

SN - 978-1-6654-5100-0

T3 - Proceedings - IEEE International Conference on Data Mining, ICDM

SP - 458

EP - 467

BT - Proceedings - 22nd IEEE International Conference on Data Mining

A2 - Zhu, Xingquan

A2 - Ranka, Sanjay

A2 - Thai, My T.

A2 - Washio, Takashi

A2 - Wu, Xindong

PB - IEEE

Y2 - 28 November 2022 through 1 December 2022

ER -

Song F, Chen B, Zhao X, Guo H, Tang R. AutoAssign: Automatic Shared Embedding Assignment in Streaming Recommendation. In Zhu X, Ranka S, Thai MT, Washio T, Wu X, editors, Proceedings - 22nd IEEE International Conference on Data Mining: ICDM 2022. IEEE. 2022. p. 458-467. (Proceedings - IEEE International Conference on Data Mining, ICDM). doi: 10.1109/ICDM54844.2022.00056

AutoAssign: Automatic Shared Embedding Assignment in Streaming Recommendation

Abstract

Publication series

Conference

Bibliographical note

Research Keywords

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

Cite this