Personalized Tourism Service Recommendation

With the development of Internet and information technology,online tourism has drawn more and more attention as a new industry.However,due to the problem of information overload,users often need to spend a lot of time choosing appropriate travel products.At present,the solutions provided by online travel agents remain in the provision of filters,sorted by price,sorted by rating without personalized recommendation.Under this background,this thesis aims at studying recommendation system in tourism area.Compared with the recommendation of books,movies,music and other objects in the traditional recommendation system,the recommendation of tourism service has its particularity,mainly in the sparseness of data,implicit feedback and the explicit attributes of the items.Based on the particularity of tourism service recommendation,this thesis firstly studies the hotel recommendation problem.By analyzing the users' historical orders,we extract the properties that has a greater impact on users' decision,and form the user profile.At the same time,in order to increase the expressiveness of the user profile in less data,this thesis introduces the data from air ticket area,and finally builds the cross-domain user profile.User's hotel selection problem is further defined as the regression problem based on user profile and hotel features.In addition,this thesis proposes another method through the combination of implicit and explicit features,which is a general method for tourism service recommendation.By combining the implicit factor analysis based on cross-domain matrix factorization and the explicit attributes of tourism products,this method can effectively reflect the potential connection between users and products,but also reflect the user's preference against the explicit attributes of tourism products.The experimental results show that the combining implicit and explicit features is more outstanding when data is sparse,and the recommendation method based on user profile shows more potential when data is sufficient.This thesis also introduces social relations in the tourism area,users that have ordered same product are considered to have strong social relations.By analogy of similarity in collaborative filtering,social relations in tourism can be used to pre-populate user-item ratings,thereby reduce the data sparsity problem.Experiments show that the introduction of social relations can lead to improvements in both methods that mentioned above.