Multi-Round Recommendations for Stable Groups
Heiska, Ilmo (2021)
2021
Tietojenkäsittelyopin maisteriohjelma - Master's Programme in Computer Science
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2021-04-16
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202104142960
https://urn.fi/URN:NBN:fi:tuni-202104142960
Tiivistelmä
Recommender systems have been used for suggesting the most suitable products and services for users in diverse scenarios. More recently, the need for making recommendations for groups of users has become increasingly relevant. In addition, there are applications in which recommendations are required in a consecutive sequence. Group recommendations present a challenge for recommender systems: how to balance the preferences of the individual members of a group. On the other hand, when making recommendations for a group for multiple consecutive rounds, a recommender system has a possibility to dynamically try to balance the preference differences between the group members.
This thesis suggests two novel group recommendation methods for multi-round group recommendation scenarios: adjusted average aggregation method and average-min-disagreement aggregation method. Both of the novel methods aim to provide a group with highly relevant results for the group while remaining fair for all group members. An experimental evaluation is designed and implemented as a 15-round recommendation sequence in order to assess the performance of the two novel methods. The experiment includes several types of groups with different degree of similarity between group members, to check the performance of the methods in various differ- ent scenarios. A recently introduced recommendation method, sequential hybrid aggregation method, is used as a baseline method for multi-round group recommendation performance.
The experimental results show that the two novel methods exceed the performance of the baseline method in all scenarios. Of the two novel methods, adjusted average method outperforms average-min-disagreement method in the early stages of the multi-round recommendation sequence. Average-min-disagreement method, on the other hand, achieves better overall results in the later stages of the multi-round recommendation sequence. Additionally, the average-min- disagreement method achieves the most fair results for all group members in all scenarios.
This thesis suggests two novel group recommendation methods for multi-round group recommendation scenarios: adjusted average aggregation method and average-min-disagreement aggregation method. Both of the novel methods aim to provide a group with highly relevant results for the group while remaining fair for all group members. An experimental evaluation is designed and implemented as a 15-round recommendation sequence in order to assess the performance of the two novel methods. The experiment includes several types of groups with different degree of similarity between group members, to check the performance of the methods in various differ- ent scenarios. A recently introduced recommendation method, sequential hybrid aggregation method, is used as a baseline method for multi-round group recommendation performance.
The experimental results show that the two novel methods exceed the performance of the baseline method in all scenarios. Of the two novel methods, adjusted average method outperforms average-min-disagreement method in the early stages of the multi-round recommendation sequence. Average-min-disagreement method, on the other hand, achieves better overall results in the later stages of the multi-round recommendation sequence. Additionally, the average-min- disagreement method achieves the most fair results for all group members in all scenarios.