User-experience oriented POI recommendation with pseudo rating

Point-of-interest (POI) recommendation faces challenges in managing data sparsity and cold-start problems, particularly in terms of personalized recommendation. Firstly, most existing users only visit a small number of POIs, which results in severe data sparsity and makes most conventional methods unsuitable because they require adequate historical data from each user to learn the model. In addition, new users have little or no historical record. Existing ask-to-rate methods for cold start user scenarios treat user answers as historical data, which is inappropriate and causes difficulty in capturing user preferences. Therefore, in this paper, to tackle the data sparsity problem, we propose a user-experience model (UEM) to uncover user behaviors from several aspects. The proposed UEM involves four factors: Who (does it), What (the user does), Where (the user does it), and When (the user does it). In contrast to other methods that model user behavior from individual records, our model introduces a latent variable of user experience to extract similarity among users. This feature makes our model suitable for mitigating the severe sparse-data issue in POI recommendations. Furthermore, to handle the cold-start problem for new users, we propose a pseudo-rating mechanism (PRM) to capture new users’ preferences and recommend POIs. For new users, photos are provided for rating, and these ratings are then used to capture their preferences and estimate the probability of their visits. Extensive experimental results demonstrate that the proposed method achieves better recommendation performance and fairness than the state-of-the-art methods. Moreover, user study results validates that the users’ preferences captured via PRM are satisfactory, and UEM provides satisfactory POI recommendations.