POI Recommendation Model Using Multi-Head Attention in Location-Based Social Network Big Data

Introduction

The rapid development of information technology has made the digitization of human mobile behavior and sharing with friends easier (Xu et al., 2020). Many social media platforms of location-based social networks (LBSNs), such as Twitter, Facebook, Instagram and Foursquare abroad, Public Comments, and Alibaba Koubei in China, have become increasingly popular and are now everywhere in daily life (Feng et al., 2015; Yin et al., 2017). Users on LBSNs are willing to share their experience on POI with friends and make comments and scores (Qian et al., 2019). Users generate a large amount of data on these social media platforms, including text content with spatio-temporal information. The accumulation of massive user check-in data has given birth to the research on user recommendation POI (Yang et al., 2017). This information is particularly useful for understanding the user’s behavior and preferences for POI (Chen et al., 2020; Yao et al., 2016). A POI is a specific location in which the user is interested. Mobile behavior can be used to understand and predict human movement and promote an individual’s daily life experiences such as transportation and entertainment (Zhao et al., 2020). Many historical check-in data provide valuable information for service providers to help them understand users’ preferences for the next actions (Zhang & Wang, 2016).

In recent years, with the rapid development and popularization of LBSNs, the number of their users has increased sharply (Islam et al., 2022). The acquisition and generation of POI information have exploded in geometric multiples. For example, Foursquare has more than 50 million active users, more than 8 billion POI signed in 2016, and Yelp has about 21 million users and 102 million comments with geographical coordinates (Guo et al., 2018). In this case, it leads to the “information overload” of the LBSN. For consumers, finding places of interest from massive POI will be interfered by redundant information. It would be difficult for businesses to make their locations stand out from a large amount of location information (Sun, 2021; Yang et al., 2017).