MGSAN: A Multi-granularity Self-attention Network for Next POI Recommendation

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Next Point-of-Interest (POI) recommendation has become a vital research trend, helping people find interesting and attractive locations. Existing methods usually exploit the individual-level POI sequences but failed to utilize the information of collective-level POI sequences. Since collective-level POIs, like shopping malls or plazas, are common in the real world, we argue that only the individual-level POI sequences cannot represent more semantic features and cannot express complete transition patterns. To this end, we propose a novel Multi-Granularity Self-Attention Network (MGSAN) for next POI recommendation, which utilizes the multi-granularity representation and the self-attention mechanism to capture the transition patterns of individual-level and collective-level POI sequences on two different levels of granularities. Specifically, individual-level and collective-level POI sequences are first constructed and embeddings of each check-in tuple are normalized. Then, MGSAN incorporates spatio-temporal features by introducing two temporal-aware encoders and two spatial-aware encoders and learns sequential patterns with the self-attention network for two granularities. Finally, we recommended a user’s next POI with the help of two sub-tasks, i.e., the activity task to predict the next category and the auxiliary task to predict the next POI type. Extensive experiments on three real-world datasets show that MGSAN outperforms state-of-the-art methods consistently.

Original languageEnglish
Title of host publicationWeb Information Systems Engineering - WISE 2021 - 22nd International Conference on Web Information Systems Engineering, WISE 2021, Proceedings
EditorsWenjie Zhang, Lei Zou, Zakaria Maamar, Lu Chen
PublisherSpringer Science and Business Media Deutschland GmbH
Pages193-208
Number of pages16
ISBN (Print)9783030915599
DOIs
StatePublished - 2021
Event22nd International Conference on Web Information Systems Engineering, WISE 2021 - Melbourne, Australia
Duration: Oct 26 2021Oct 29 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13081 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference22nd International Conference on Web Information Systems Engineering, WISE 2021
Country/TerritoryAustralia
CityMelbourne
Period10/26/2110/29/21

Fingerprint

Dive into the research topics of 'MGSAN: A Multi-granularity Self-attention Network for Next POI Recommendation'. Together they form a unique fingerprint.

Cite this