Abstract
I. Introduction
II. Related Work
III. Problem Formulation and Preliminaries
IV. The SSSER Model
V. Experiment
Authors
Figures
References
Abstract
Point-of-Interest (POI) recommendation is one of the important services of location-based social networks (LBSNs), which has become an important way to help users discover interesting places and increase the potential income of related companies. Although human movement presents a sequential pattern in the LBSN. There still are the following problems: (1) when modeling the sequence data, most of the existing works assume that the check-in time depends on the location transformation in the location sequence. In particular, these works emphasize the equivalent transition probabilities between locations for all users to capture the check-in sequential pattern, whereas they ignore the spatial and temporal information of personalized context in some actual personal check-in scenarios; (2) most of the existing POI recommendation algorithms fail to utilize the social information related to modeling users to improve the final recommendation performance.To tackle the above challenges, we propose a new personalized successive POI recommendation model called Spatiotemporal Sequential and Social Embedding Rank model, named SSSER. First, we use a hybrid deep learning model based on the convolution filter and multilayer perceptron model to mine the sequence pattern among the users’ checked-in locations. Then, we use the method of metric learning to model the social relationship among users. Finally, we propose a unified framework to recommend POIs combining the users’ personal interests, the check-in sequential influence and social information simultaneously for the successive POI recommendation. And the BPR standard is used to optimize the loss function to fit the user’s partial order of POIs. The experimental results on the real datasets show that our proposed POI recommendation algorithm outperforms the other state-of-the-art POI recommendation algorithms.
Introduction
With the rapid development of Web2.0, wireless communication and location collection technology have promoted many location-based social networks (LBSNs), such as Foursquare, Yelp, Facebook and so on. Via these LBSNs, users can establish social connections with other users, explore the surrounding environment, and share their life experiences by checking in points-of-interest (POIs) such as restaurants, shopping centers, and tourist attractions. In addition to providing an interactive platform for users, LBSN contains rich data (check-in data, social relationships, comment information, etc.), which can be applied to predict users’ preferences and recommend some unvisited POIs that may be of interest to users. The recommendation, by means of LBSN, of a geographical location that a user may be interested in is referred to as the recommendation of POI. The POI recommendation, on the one hand, satisfies the individualized needs of users to explore new geographical areas and discover new POIs and at the same time alleviates the problem of information overload faced by users. On the other hand, the recommendation of POI helps LBSN service providers to play a pivotal role in realizing intelligent location services. Therefore, how to provide users with accurate POI recommendation has drawn the attention of more and more researchers in recent years [1].