Destination prediction considering both tweet contents and location transition hitstory

We propose a method of predicting destinations by using Twitter posts with location information. The proposed method chooses base tweets, which is close to the current user´s tweet, and then predict destination using the next set of tweets of base tweet. The base tweets are selected based on not only location closeness but also similarity of tweet content. We evaluate the proposed method by the error range of the distance between predicted destination and golden answer. We used three months of Twitter data with location information (almost 40 mil.) as the test set tweets. The experimental result demonstrates that the prediction accuracy of the proposed method is superior to the baseline, which only consider the location similarity.