Appearance Frequency-Based Ranking Method for Improving Recommendation Diversity

Seiki Miyamoto, Takumi Zamami, Hayato Yamana

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Recommender systems are used to analyze users' preferences through their past activities and to personalize recommendations for each user based on what they might be interested in. The performance of the recommender system is most commonly measured using only recommendation accuracy. However, recommending accurate items does not mean that the generated recommendation is the best for the user because it can be biased towards some items that have a higher chance of being liked by users, such as popular items. Recommendations become repetitive and obvious with biased item selection and are less likely to be personalized. To mitigate bias and repetitiveness, recommendation diversity has been studied. However, diversity has a trade-off relationship with accuracy. Modifying the recommendation algorithm to consider diversity while learning about user preferences would not only cause loss in accuracy, but also lead to less precise reading of user preferences. Instead, using ranking methods to re-rank the priority of items predicted, the recommendation algorithm would keep the preciseness of the algorithm. In this study, a ranking method using the appearance frequency of items to restrict the items from being frequently recommended will be proposed. The experimental results showed that the proposed method consistently improved diversity in multiple diversity metrics.

Original languageEnglish
Title of host publication2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages420-425
Number of pages6
ISBN (Electronic)9781728112824
DOIs
Publication statusPublished - 2019 May 10
Event4th IEEE International Conference on Big Data Analytics, ICBDA 2019 - Suzhou, China
Duration: 2019 Mar 152019 Mar 18

Publication series

Name2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019

Conference

Conference4th IEEE International Conference on Big Data Analytics, ICBDA 2019
CountryChina
CitySuzhou
Period19/3/1519/3/18

Fingerprint

Recommendations
Ranking
Recommender systems
User Preferences
Recommender Systems
Biased
Trade-offs
Likely
Metric
Experimental Results
User preferences

Keywords

  • collaborative filtering
  • recommendation diversity
  • recommender system

ASJC Scopus subject areas

  • Artificial Intelligence
  • Information Systems
  • Information Systems and Management
  • Statistics, Probability and Uncertainty

Cite this

Miyamoto, S., Zamami, T., & Yamana, H. (2019). Appearance Frequency-Based Ranking Method for Improving Recommendation Diversity. In 2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019 (pp. 420-425). [8713185] (2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICBDA.2019.8713185

Appearance Frequency-Based Ranking Method for Improving Recommendation Diversity. / Miyamoto, Seiki; Zamami, Takumi; Yamana, Hayato.

2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019. Institute of Electrical and Electronics Engineers Inc., 2019. p. 420-425 8713185 (2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Miyamoto, S, Zamami, T & Yamana, H 2019, Appearance Frequency-Based Ranking Method for Improving Recommendation Diversity. in 2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019., 8713185, 2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019, Institute of Electrical and Electronics Engineers Inc., pp. 420-425, 4th IEEE International Conference on Big Data Analytics, ICBDA 2019, Suzhou, China, 19/3/15. https://doi.org/10.1109/ICBDA.2019.8713185
Miyamoto S, Zamami T, Yamana H. Appearance Frequency-Based Ranking Method for Improving Recommendation Diversity. In 2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019. Institute of Electrical and Electronics Engineers Inc. 2019. p. 420-425. 8713185. (2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019). https://doi.org/10.1109/ICBDA.2019.8713185
Miyamoto, Seiki ; Zamami, Takumi ; Yamana, Hayato. / Appearance Frequency-Based Ranking Method for Improving Recommendation Diversity. 2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019. Institute of Electrical and Electronics Engineers Inc., 2019. pp. 420-425 (2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019).
@inproceedings{85dc870fdb464a09863907e141d91d75,
title = "Appearance Frequency-Based Ranking Method for Improving Recommendation Diversity",
abstract = "Recommender systems are used to analyze users' preferences through their past activities and to personalize recommendations for each user based on what they might be interested in. The performance of the recommender system is most commonly measured using only recommendation accuracy. However, recommending accurate items does not mean that the generated recommendation is the best for the user because it can be biased towards some items that have a higher chance of being liked by users, such as popular items. Recommendations become repetitive and obvious with biased item selection and are less likely to be personalized. To mitigate bias and repetitiveness, recommendation diversity has been studied. However, diversity has a trade-off relationship with accuracy. Modifying the recommendation algorithm to consider diversity while learning about user preferences would not only cause loss in accuracy, but also lead to less precise reading of user preferences. Instead, using ranking methods to re-rank the priority of items predicted, the recommendation algorithm would keep the preciseness of the algorithm. In this study, a ranking method using the appearance frequency of items to restrict the items from being frequently recommended will be proposed. The experimental results showed that the proposed method consistently improved diversity in multiple diversity metrics.",
keywords = "collaborative filtering, recommendation diversity, recommender system",
author = "Seiki Miyamoto and Takumi Zamami and Hayato Yamana",
year = "2019",
month = "5",
day = "10",
doi = "10.1109/ICBDA.2019.8713185",
language = "English",
series = "2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "420--425",
booktitle = "2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019",

}

TY - GEN

T1 - Appearance Frequency-Based Ranking Method for Improving Recommendation Diversity

AU - Miyamoto, Seiki

AU - Zamami, Takumi

AU - Yamana, Hayato

PY - 2019/5/10

Y1 - 2019/5/10

N2 - Recommender systems are used to analyze users' preferences through their past activities and to personalize recommendations for each user based on what they might be interested in. The performance of the recommender system is most commonly measured using only recommendation accuracy. However, recommending accurate items does not mean that the generated recommendation is the best for the user because it can be biased towards some items that have a higher chance of being liked by users, such as popular items. Recommendations become repetitive and obvious with biased item selection and are less likely to be personalized. To mitigate bias and repetitiveness, recommendation diversity has been studied. However, diversity has a trade-off relationship with accuracy. Modifying the recommendation algorithm to consider diversity while learning about user preferences would not only cause loss in accuracy, but also lead to less precise reading of user preferences. Instead, using ranking methods to re-rank the priority of items predicted, the recommendation algorithm would keep the preciseness of the algorithm. In this study, a ranking method using the appearance frequency of items to restrict the items from being frequently recommended will be proposed. The experimental results showed that the proposed method consistently improved diversity in multiple diversity metrics.

AB - Recommender systems are used to analyze users' preferences through their past activities and to personalize recommendations for each user based on what they might be interested in. The performance of the recommender system is most commonly measured using only recommendation accuracy. However, recommending accurate items does not mean that the generated recommendation is the best for the user because it can be biased towards some items that have a higher chance of being liked by users, such as popular items. Recommendations become repetitive and obvious with biased item selection and are less likely to be personalized. To mitigate bias and repetitiveness, recommendation diversity has been studied. However, diversity has a trade-off relationship with accuracy. Modifying the recommendation algorithm to consider diversity while learning about user preferences would not only cause loss in accuracy, but also lead to less precise reading of user preferences. Instead, using ranking methods to re-rank the priority of items predicted, the recommendation algorithm would keep the preciseness of the algorithm. In this study, a ranking method using the appearance frequency of items to restrict the items from being frequently recommended will be proposed. The experimental results showed that the proposed method consistently improved diversity in multiple diversity metrics.

KW - collaborative filtering

KW - recommendation diversity

KW - recommender system

UR - http://www.scopus.com/inward/record.url?scp=85066620659&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85066620659&partnerID=8YFLogxK

U2 - 10.1109/ICBDA.2019.8713185

DO - 10.1109/ICBDA.2019.8713185

M3 - Conference contribution

AN - SCOPUS:85066620659

T3 - 2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019

SP - 420

EP - 425

BT - 2019 4th IEEE International Conference on Big Data Analytics, ICBDA 2019

PB - Institute of Electrical and Electronics Engineers Inc.

ER -