TY - JOUR

T1 - Hybrid POI group recommender system based on group type in LBSN

AU - Bahari Sojahrood, Zahra

AU - Taleai, Mohammad

AU - Cheng, Hao

PY - 2023/6/1

Y1 - 2023/6/1

N2 - Point of interest (POI) group recommender systems (GRSs) aim to suggest places for a group of users. Compared to recommender systems for individual users, GPRs are more challenging due to the complexity of groups formed by members with conflicting interests and preferences. To cope with this challenge, the current POI GRSs utilize strategies-based approaches, e.g., group history, aggregation of group members’ history, and further improved aggregation with the consideration of group members’ interactions. However, these GRSs still suffer from low accuracy and the cold-start problem for groups with insufficient historical information. Moreover, rare studies have attempted to compare/combine the different strategies for GRSs. In this paper, to achieve the collective advantages over existing GRSs, a hybrid method performs switching among different recommendation strategies based on the criteria of group type (persistent and ephemeral regarding group history) and homogeneity (similarity of group members’ interests and preferences) is proposed. A dataset from the city of Ankara that contains user and group check-ins, was extracted from the Foursquare Swarm application to evaluate the performance of the proposed method. This hybrid method outperforms the single strategies using various algorithms by improving at least 30% in precision@5 and 25% in recall@5 across all the test groups. The empirical results demonstrate that the introduced hybrid solution, which automatically identifies group type and homogeneity, can quickly and effectively go beyond the restrictions of the individual strategies for GRSs.

AB - Point of interest (POI) group recommender systems (GRSs) aim to suggest places for a group of users. Compared to recommender systems for individual users, GPRs are more challenging due to the complexity of groups formed by members with conflicting interests and preferences. To cope with this challenge, the current POI GRSs utilize strategies-based approaches, e.g., group history, aggregation of group members’ history, and further improved aggregation with the consideration of group members’ interactions. However, these GRSs still suffer from low accuracy and the cold-start problem for groups with insufficient historical information. Moreover, rare studies have attempted to compare/combine the different strategies for GRSs. In this paper, to achieve the collective advantages over existing GRSs, a hybrid method performs switching among different recommendation strategies based on the criteria of group type (persistent and ephemeral regarding group history) and homogeneity (similarity of group members’ interests and preferences) is proposed. A dataset from the city of Ankara that contains user and group check-ins, was extracted from the Foursquare Swarm application to evaluate the performance of the proposed method. This hybrid method outperforms the single strategies using various algorithms by improving at least 30% in precision@5 and 25% in recall@5 across all the test groups. The empirical results demonstrate that the introduced hybrid solution, which automatically identifies group type and homogeneity, can quickly and effectively go beyond the restrictions of the individual strategies for GRSs.

KW - Aggregation-based strategy

KW - Cold start

KW - Group-based methods

KW - Probabilistic Matrix Factorization (PMF)

UR - http://www.scopus.com/inward/record.url?scp=85150340476&partnerID=8YFLogxK

U2 - 10.1016/j.eswa.2023.119681

DO - 10.1016/j.eswa.2023.119681

M3 - Article

AN - SCOPUS:85150340476

VL - 219

JO - Expert systems with applications

JF - Expert systems with applications

SN - 0957-4174

M1 - 119681

ER -