| Summary: | 碩士 === 國立政治大學 === 資訊科學系 === 106 === With the development of globalization and the popularity of social media, self-guided tour has become a trend. It takes much time for a user to plan a tour, including searching for attractions. Therefore, the demand for automatic attractions recommendation for self-guided tourist is rising.
Compared with the traditional recommendation mechanism, tourist attractions recommendation needs to overcome no user rating scores for tourist attractions and data sparsity, because most users usually visit only few attractions. In the previous research, most work only considers the similarity between users to improve the data sparsity problem based on collaborative filtering. Less consideration is paid to the correlation of visited attractions between tourist destinations. In this thesis, we collected user travel records from a large number of geo-tagged photos, and utilized Latent Dirichlet Allocation (LDA) to discover the preference distribution of each tourist for each destination. Then, Partial Least Square Regression (PLSR) is employed to find the correlation relationship of preference distributions between tourist destinations. The attractions are personally recommended based on the user’s preferences and the discovered correlation relationships. The experiment shows our proposed method is better than other approaches.
|
|---|
