Vector Representation Based Model Considering Randomness of User Mobility for Predicting Potential Users

摘要

With increasing popularity of location-based social networks, POI recommendation has received much attenton recently. Unlike most of the current studies which provide recommendations from perspective of users, in this paper, we focus on the perspective of Point-of-Interest (POI) for predicting potential users for a given POI. We propose a novel vector representation model for the prediction. Many current matrix factorization-based methods only pay attention to combining new information and basic matrix factorization, while in our model, we improve the matrix factorization model itself by replacing dot product with cosine similarity. We also address the problem of randomness of user's check-in behavior by applying deep neural network to modeling the relationships between the user's current check-in and context information of current check-in. Extensive experiments conducted on two real-world datasets demonstrate the superior performance of our proposed model and the effectiveness of the factors incorporated in our model.