A Unified Contextual Bandit Framework for Long- and Short-Term Recommendations

Maryam Tavakol; Ulf Brefeld

doi:10.1007/978-3-319-71246-8_17

A Unified Contextual Bandit Framework for Long- and Short-Term Recommendations

Professorship of Information Systems, in particular Machine Learning

Research output: Contributions to collected editions/works › Article in conference proceedings › Research › peer-review

4 Citations (Scopus)

Abstract

We present a unified contextual bandit framework for recommendation problems that is able to capture long- and short-term interests of users. The model is devised in dual space and the derivation is consequentially carried out using Fenchel-Legrende conjugates and thus leverages to a wide range of tasks and settings.We detail two instantiations for regression and classification scenarios and obtain well-known algorithms for these special cases. The resulting general and unified framework allows for quickly adapting contextual bandits to different applications at-hand. The empirical study demonstrates that the proposed long- and short-term framework outperforms both, short-term and long-term models on data. Moreover, a tweak of the combined model proves beneficial in cold start problems.

Original language	English
Title of host publication	Machine Learning and Knowledge Discovery in Databases : European Conference, ECML PKDD 2017 Skopje, Macedonia, September 18 – 22, 2017; Proceedings, Part II
Editors	Michelangelo Ceci, Jaakko Hollmen, Ljupco Todorovski, Celine Vens, Saso Dzeroski
Number of pages	16
Volume	2
Publisher	Springer Verlag
Publication date	30.12.2017
Pages	269-284
ISBN (Print)	978-3-319-71245-1
ISBN (Electronic)	978-3-319-71246-8
DOIs	https://doi.org/10.1007/978-3-319-71246-8_17
Publication status	Published - 30.12.2017
Event	ECML-PKDD 2017 : European Conference on Machine Learning and Principles and Practice of Knowledge Discovery - ECML PKDD 2017 - Skopje, Macedonia, Skopje, Macedonia, The Former Yugoslav Republic of Duration: 19.09.2017 → 22.09.2017 Conference number: 27 http://ecmlpkdd2017.org (Official Event Website) http://ecmlpkdd2017.ijs.si/

Research areas and keywords

Informatics
Machine learning
Recommender systems
Contextual bandits
Recommendation
Dual optimization
Personalization

Access to Document

10.1007/978-3-319-71246-8_17

Cite this

Tavakol, M., & Brefeld, U. (2017). A Unified Contextual Bandit Framework for Long- and Short-Term Recommendations. In M. Ceci, J. Hollmen, L. Todorovski, C. Vens, & S. Dzeroski (Eds.), Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2017 Skopje, Macedonia, September 18 – 22, 2017; Proceedings, Part II (Vol. 2, pp. 269-284). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10535 LNAI). Springer Verlag. https://doi.org/10.1007/978-3-319-71246-8_17

Tavakol, Maryam ; Brefeld, Ulf. / A Unified Contextual Bandit Framework for Long- and Short-Term Recommendations. Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2017 Skopje, Macedonia, September 18 – 22, 2017; Proceedings, Part II. editor / Michelangelo Ceci ; Jaakko Hollmen ; Ljupco Todorovski ; Celine Vens ; Saso Dzeroski. Vol. 2 Springer Verlag, 2017. pp. 269-284 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inbook{0f45dbe652f64731ab63d83ca20dbbef,

title = "A Unified Contextual Bandit Framework for Long- and Short-Term Recommendations",

abstract = "We present a unified contextual bandit framework for recommendation problems that is able to capture long- and short-term interests of users. The model is devised in dual space and the derivation is consequentially carried out using Fenchel-Legrende conjugates and thus leverages to a wide range of tasks and settings.We detail two instantiations for regression and classification scenarios and obtain well-known algorithms for these special cases. The resulting general and unified framework allows for quickly adapting contextual bandits to different applications at-hand. The empirical study demonstrates that the proposed long- and short-term framework outperforms both, short-term and long-term models on data. Moreover, a tweak of the combined model proves beneficial in cold start problems.",

keywords = "Informatics, Machine learning, Recommender systems, Contextual bandits, Recommendation, Dual optimization, Personalization",

author = "Maryam Tavakol and Ulf Brefeld",

year = "2017",

month = dec,

day = "30",

doi = "10.1007/978-3-319-71246-8\_17",

language = "English",

isbn = "978-3-319-71245-1",

volume = "2",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Verlag",

pages = "269--284",

editor = "Michelangelo Ceci and Jaakko Hollmen and Ljupco Todorovski and Celine Vens and Saso Dzeroski",

booktitle = "Machine Learning and Knowledge Discovery in Databases",

address = "Germany",

note = "ECML-PKDD 2017 : European Conference on Machine Learning and Principles and Practice of Knowledge Discovery - ECML PKDD 2017, ECML PKDD 2017 ; Conference date: 19-09-2017 Through 22-09-2017",

url = "http://ecmlpkdd2017.org, http://ecmlpkdd2017.ijs.si/",

}

Tavakol, M & Brefeld, U 2017, A Unified Contextual Bandit Framework for Long- and Short-Term Recommendations. in M Ceci, J Hollmen, L Todorovski, C Vens & S Dzeroski (eds), Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2017 Skopje, Macedonia, September 18 – 22, 2017; Proceedings, Part II. vol. 2, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10535 LNAI, Springer Verlag, pp. 269-284, ECML-PKDD 2017 : European Conference on Machine Learning and Principles and Practice of Knowledge Discovery - ECML PKDD 2017, Skopje, Macedonia, The Former Yugoslav Republic of, 19.09.17. https://doi.org/10.1007/978-3-319-71246-8_17

A Unified Contextual Bandit Framework for Long- and Short-Term Recommendations. / Tavakol, Maryam ; Brefeld, Ulf.
Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2017 Skopje, Macedonia, September 18 – 22, 2017; Proceedings, Part II. ed. / Michelangelo Ceci; Jaakko Hollmen; Ljupco Todorovski; Celine Vens; Saso Dzeroski. Vol. 2 Springer Verlag, 2017. p. 269-284 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10535 LNAI).

Research output: Contributions to collected editions/works › Article in conference proceedings › Research › peer-review

TY - CHAP

T1 - A Unified Contextual Bandit Framework for Long- and Short-Term Recommendations

AU - Tavakol, Maryam

AU - Brefeld, Ulf

N1 - Conference code: 27

PY - 2017/12/30

Y1 - 2017/12/30

N2 - We present a unified contextual bandit framework for recommendation problems that is able to capture long- and short-term interests of users. The model is devised in dual space and the derivation is consequentially carried out using Fenchel-Legrende conjugates and thus leverages to a wide range of tasks and settings.We detail two instantiations for regression and classification scenarios and obtain well-known algorithms for these special cases. The resulting general and unified framework allows for quickly adapting contextual bandits to different applications at-hand. The empirical study demonstrates that the proposed long- and short-term framework outperforms both, short-term and long-term models on data. Moreover, a tweak of the combined model proves beneficial in cold start problems.

AB - We present a unified contextual bandit framework for recommendation problems that is able to capture long- and short-term interests of users. The model is devised in dual space and the derivation is consequentially carried out using Fenchel-Legrende conjugates and thus leverages to a wide range of tasks and settings.We detail two instantiations for regression and classification scenarios and obtain well-known algorithms for these special cases. The resulting general and unified framework allows for quickly adapting contextual bandits to different applications at-hand. The empirical study demonstrates that the proposed long- and short-term framework outperforms both, short-term and long-term models on data. Moreover, a tweak of the combined model proves beneficial in cold start problems.

KW - Informatics

KW - Machine learning

KW - Recommender systems

KW - Contextual bandits

KW - Recommendation

KW - Dual optimization

KW - Personalization

UR - http://ecmlpkdd2017.ijs.si/papers/paperID239.pdf

UR - https://www.scopus.com/pages/publications/85040222425

U2 - 10.1007/978-3-319-71246-8_17

DO - 10.1007/978-3-319-71246-8_17

M3 - Article in conference proceedings

SN - 978-3-319-71245-1

VL - 2

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 269

EP - 284

BT - Machine Learning and Knowledge Discovery in Databases

A2 - Ceci, Michelangelo

A2 - Hollmen, Jaakko

A2 - Todorovski, Ljupco

A2 - Vens, Celine

A2 - Dzeroski, Saso

PB - Springer Verlag

T2 - ECML-PKDD 2017 : European Conference on Machine Learning and Principles and Practice of Knowledge Discovery - ECML PKDD 2017

Y2 - 19 September 2017 through 22 September 2017

ER -

Tavakol M , Brefeld U. A Unified Contextual Bandit Framework for Long- and Short-Term Recommendations. In Ceci M, Hollmen J, Todorovski L, Vens C, Dzeroski S, editors, Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2017 Skopje, Macedonia, September 18 – 22, 2017; Proceedings, Part II. Vol. 2. Springer Verlag. 2017. p. 269-284. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-319-71246-8_17

A Unified Contextual Bandit Framework for Long- and Short-Term Recommendations

Abstract

Research areas and keywords

Access to Document

Fingerprint

Cite this