TY - GEN
T1 - GN-GCN
T2 - 24th International Conference on Information Integration and Web Intelligence, iiWAS 2022, held in conjunction with the 20th International Conference on Advances in Mobile Computing and Multimedia Intelligence, MoMM 2022
AU - Mo, Fan
AU - Yamana, Hayato
N1 - Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - Point-of-interest (POI) recommendation helps users filter information and discover their interests. In recent years, graph convolution network (GCN)–based methods have become state-of-the-art algorithms for improving recommendation performance. Especially integrating GCN with multiple information, such as geographical information, is a promising way to achieve better performance; however, it tends to increase the number of trainable parameters, resulting in the difficulty of model training to reduce the performance. In this study, we mine users’ active areas and extend the definition of neighbors in GCN, called active area neighbors. Our study is the first attempt to integrate geographic information into a GCN POI recommendation system without increasing the number of trainable parameters and maintaining the ease of training. The experimental evaluation confirms that compared with the state-of-the-art lightweight GCN models, our method improves Recall@ 10 from 0.0562 to 0.0590 (4.98%) on Yelp dataset and from 0.0865 to 0.0898 (3.82%) on Gowalla dataset.
AB - Point-of-interest (POI) recommendation helps users filter information and discover their interests. In recent years, graph convolution network (GCN)–based methods have become state-of-the-art algorithms for improving recommendation performance. Especially integrating GCN with multiple information, such as geographical information, is a promising way to achieve better performance; however, it tends to increase the number of trainable parameters, resulting in the difficulty of model training to reduce the performance. In this study, we mine users’ active areas and extend the definition of neighbors in GCN, called active area neighbors. Our study is the first attempt to integrate geographic information into a GCN POI recommendation system without increasing the number of trainable parameters and maintaining the ease of training. The experimental evaluation confirms that compared with the state-of-the-art lightweight GCN models, our method improves Recall@ 10 from 0.0562 to 0.0590 (4.98%) on Yelp dataset and from 0.0865 to 0.0898 (3.82%) on Gowalla dataset.
KW - Geographical information
KW - Graph convolution network
KW - POI recommendation
KW - Trainable parameter number-holding
UR - http://www.scopus.com/inward/record.url?scp=85145006703&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85145006703&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-21047-1_15
DO - 10.1007/978-3-031-21047-1_15
M3 - Conference contribution
AN - SCOPUS:85145006703
SN - 9783031210464
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 153
EP - 165
BT - Information Integration and Web Intelligence - 24th International Conference, iiWAS 2022, Proceedings
A2 - Pardede, Eric
A2 - Delir Haghighi, Pari
A2 - Khalil, Ismail
A2 - Kotsis, Gabriele
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 28 November 2022 through 30 November 2022
ER -