Learning Points and Routes to Recommend Trajectories 

Recommending Points of Interest (POIs) and trajectories to tourists by exploring data with geographical information from online social media is an interesting and challenging task. We propose to utilize learning to rank and Markov Chain to capture the properties of POIs as well as the transition pattern between POIs, transition probabilities between POIs were factorized according to several POI features to deal with data sparsity. Furthermore, a probabilistic model and a structured support vector machine were leveraged to combine the results of learning to rank as well as the factorised Markov Chain to recommend trajectories. Experimental results on five trajectory datasets extracted from Flickr photos show performance improvements of our approach over the state-of-the-art and reveal many interesting properties of trajectories in different datasets.