Xi'an tourism destination image analysis via deep learning

Existing methods focus on destination image construction by textual description or visual content separately. However, descriptions and images are closely related since they are taken from the same reviews and represent tourists impression of the city. It's questionable to study them separately. In this paper, we used both images and descriptions from the reviews to construct Xi'an tourism destination image. More concretely, scene recognition, landmark recognition and food image recognition are utilized to obtain visual image. Lexical analysis is applied to obtain semantic image. We further compared the differences between visual image and semantic image then we proposed the fusion image. Finally, the top 300 key words and differences of the photo contents between the adjacent 2 years are selected to discovering new changes of the destination image. Results show that the visual image and semantic image are significant different from each other and the new changes of semantic image are closely related to the events or things that happened in that year and changes of visual image are not significant.