Abstract: The existing tourism recommendation methods are limited because of the implicit feedback and extreme sparsity of tourism data sets. To solve the problem, an improved collaborative filtering algorithm based on embedding is proposed. Firstly, every route is represented as a low dimensional vector by using Doc2vector and every tourist is represented as a low dimensional vector based on all routes he/she had taken. Secondly, the co -occurrence routes between two tourists are obtained on the similarity of routes. Then the similarities among tourists are calculated according to the set of co-occurrence routes among them. Then, the routes are recommended by using the improved collaborative filtering. Finally, the method is proved to be effective on a real tourism data set.

Key words: word vector, improved collaborative filtering, similarity, data sparsity