Article Preview
TopIntroduction
In location-based social networks (LBSNs), people can share their position and activities to other users through check-ins. Generally, a check-in includes user information, time, location and some reviews. Different from traditional ratings for items, user check-ins implicitly demonstrate user preference. Mass check-in data provides an opportunity to mining user preference and enables lots of location-aware services as shown in Figure 1. Point of interest (POI) recommendation is one of such services, which is aiming to recommend new places which users have not visited before when they explore their surroundings. POI recommendation not only helps users identify favorite locations, but also help POI owner acquire more target customers (Zhang & Chow, 2016; Liu, Tuan-Anh, Gao, & Yuan, 2017).
Figure 1. The location based social network
POI recommendation has attracted more and more researchers’ attention (Zeng, Li, Wen, & Zhou, 2017). Collaborative filtering (CF) is widely adopted to POI recommendation. Ye, Yin, Lee, and Lee (2011) compared user-based and item-based method to make POI recommendation. Their experimental results show user-based approach performs better than item-based approach. In this case, there are much work about user-based CF (Liu, Ma, Chen, & Xiong, 2013; Ye, Yin, & Lee, 2010; Bao, Zheng, & Mokbel, 2012; Yuan, Cong, Ma, Sun, & Thalmann, 2013; Ye et al., 2011) and model-based CF (Gao, Tang, Hu, & Liu, 2015; Lian et al., 2014; Liu, Fu, Yao, & Xiong, 2013; Gao, Tang, Hu, & Liu, 2013; Hu, Sun, & Liu, 2014; Yuan, Cong, & Sun, 2014; Tang, Hu, Gao, & Liu, 2013; Cho, Myers, & Leskovec, 2011). For example, Ye et al. (2011) thought social relationship are incorporated into recommendation. Ye et al. (2010) proposed Friend-based CF (FCF), which directly recommends POIs for users from those locations their friends visited before. Compared with traditional user-based CF (user-CF), this method has lower computational overhead. However, the method is limited because the tastes of a user’s friends may vary greatly. Cheng, Yang, King, and Lyu (2012) found the average check-ins overlap between friends is about 9.6%. This indicates fiends may has different interest. Some matrix factorization (MF) based methods can easily fuse factors, such as geographical distance and social relation, but the optimal result is greatly affected by all kinds of parameters.
In this paper, we propose a novel method which fused geographical influence and social reputation to improve POI recommendation. The main contributions are as follows: