The prediction of next location for users in location-based social networks has become an increasing significant requirement since it can benefit both users and business. However, existing methods lack an integrated analysis of sequence context, input contexts and user preferences in a unified way, and result in an unsatisfactory prediction. Moreover, the interaction between different kinds of input contexts has not been investigated. In this paper, we propose a multi-context integrated deep neural network model (MCIDNN) to improve the accuracy of the next location prediction. In this model, we integrate sequence context, input contexts and user preferences into a cohesive framework. Firstly, we model sequence context and interaction of different kinds of input contexts jointly by extending the recurrent neural network to capture the semantic pattern of user behaviors from check-in records. After that, we design a feedforward neural network to capture high-level user preferences from check-in data and incorporate that into MCI-DNN. To deal with different kinds of input contexts in the form of multi-field categorical, we adopt embedding representation technology to automatically learn dense feature representations of input contexts. Experimental results on two typical real-world datasets show that the proposed model outperforms the current state-of-the-art approaches by about 57.12% for Foursquare and 76.4% for Gowalla on average regarding F1- score@5.

INTRODUCTION

ith the rapid development of wireless communication technologies and the popularization of mobile devices, the emergence of location-based social networks (LBSNs), e.g., Foursquare, Gowalla, and Yelp, has bridged the gap between cyberspace and the physical world. In LBSNs, users can post their physical locations in the form of “check-ins”. They can also share their life experiences in the physical world, resulting in new opportunities to extract further insights into user preferences and behaviors [1]. Predicting location in LBSNs accurately is crucial for helping users find interesting places and services [2], for contributing to the connection of next hop in high-speed Internet of Things (IoT) [3] and for facilitating business owners to launch mobile advertisements to target users [4]. To gain significant benefit for both users and business, the prediction of next locations for users has recently attracted much academic attention [5], [6]. Predicting the next location is not just confined to estimating user preferences, which a general location prediction focuses on [7], [8]. It also includes the modeling of sequence transition from check-ins to predict user’s future location. This is relevant because human movement exhibits strong sequence dependency [9], [10]. Current studies on the modeling sequence pattern are mainly based on first-order Markov Chain (MC) model such as Hidden Markov Model (HMM) [5], and Factorizing Personalized Markov Chain Model (FPMC) [9]. However, those methods are used to predict the possibility of visiting location based only on the latest check-in due to the higher computational complexity, and the influences of short-term and long-term sequence context (i.e., a set of locations visited before) have been ignored. Recently, deep neural networks have proved to be useful in modeling those sequence contexts in next location prediction. For example, by an analogous user’s check-in trajectory to a sentence, Liu et al. [10] and Zhao et al. [11] employed the word2vec framework to learn the hidden representation of locations by capturing the influence of shortterm sequence context. Liu et al. [13] and Yang et al. [14] leverage recurrent neural network (RNN) to capture the influence of long-term sequence context on next location decision. Cui et al. [15] propose a Hierarchical Contextual Attention-based GRU (HCA-GRU) network to capture longterm dependency and short-term interest. Their results show better performance in predicting precision than MC-based approaches.