A Content-Aware POI Recommendation Method in Location-Based Social Networks Based on Deep CNN and Multi-Objective Immune Optimization

Abstract

Aiming at the problem of sparse data and multi-attribute data in location-based social networks (LBSNs), a content-aware point-of-interest (POI) recommendation method based on deep convolution neural network (CNN) and multi-objective immune optimization is proposed. Firstly, three types of content information are modeled: Geographic information is modeled by location weighting strategy; Emotional information from users’ comment texts is modeled by CNN; And user preferences are modeled by interaction matrix between comment content features and user potential features. Then, the three types of content information are inputted into a CNN based POI recommendation framework. To avoid adjusting too many weight coefficients at the same time, geographic information, user emotional information and user preferences are respectively optimized in three optimization objective functions. Finally, the non-dominated neighbor immune algorithm (NNIA) is used to solve the multi-objective optimization problem. Without adjusting any weight coefficients, a variety of POI lists can be respectively recommended for each user. In Foursquare and Brightkite datasets, the check-in records and comment texts data from New York (NY), Los Angeles (LA) and Austen were selected for experimental analysis. It can be seen from the experimental results that compared with other methods, the proposed method can ensure high recommendation accuracy under cold start and can achieve the accuracy and diversity of POI recommendation under different recommendation list length.