Deep Transfer Learning for Successive POI Recommendation

Abstract

Personalized POI recommendation attracts more and more attention from both industrial and research fields. Due to data collection mechanism, it is common to see data collection with the unbalanced spatial distribution. For example, some cities may release check-ins for multiple years while others only release a few days of data. In this paper, we tackle the problem of successive POI recommendation for the cities with only a short period of data samples. We aim to leverage the transfer learning technique to utilize the knowledge from long-period data to enhance the POI recommendation process in target cities. Different from existing methods that transfer knowledge from one single city to a target city, we utilize the knowledge from multiple source cities to increase the stability of transfer. Specifically, our proposed model is designed as the spatio-temporal attentive recurrent neural network, MetaGRU, with a meta-learning paradigm. The spatio-temporal attentive mechanism leverage an external memory layer to store processed user historical preference information, and use spatio-temporal attention to capture the different correlations between user current status with past check-in behaviors. In addition, the meta-learning paradigm learns a well-generalized initialization of the spatio-temporal neural network, which can be effectively adapted to target cities. Extensive experiments show that our method is able to outperform state-of-the-art successive POI recommendation models in multiple tasks.

Supported by the National Natural Science Foundation of China under Grant No.: 62077044, 61702470, 62002343.