Reserach On Tourist Attractions Recommendation System Based On Deep Collaborative Filtering And Multimodal Analysis

Statistics shows that about three-quarters tourists will go to the mainstream travel websites to check users' reviews or ratings,which helps them to better choose tourist attractions and make good travel plan.The research on tourist attraction recommendation system has made some progress,but some problems still remain such as data sparseness,users' hidden preference information wasn't considered,and the latent semantic information of attractions' images wasn't mined.In order to achieve better recommendations performance,the stratified sampling statistical model is used to obtain users' preferences,the traditional Bayesian Personalized Ranking(BPR)and Visual Bayesian Personalized Ranking(VBPR)models are modified to optimize the recommendation system.The main tasks are described as follows:First,the tourist attraction recommendation system based on stratified sampling statistics and BPR model is proposed.A questionnaire survey is used to obtain users' preference.Users' ratings and attractions' images are captured in turn.Then the "Wisdom Tourism" data set is established.Based on the data set,a novel tourist attraction recommendation system based on stratified sampling statistics and BPR model is designed.The experimental results show that the average accuracy,the average recall rate,and the F1 of the hybrid recommendation system,are improved about 3.6%,5.1%,and 5.0% compared to the best baseline model respectively,which alleviates the data sparseness problem to some extent.Second,the tourist attraction recommendation system based on stratified sampling statistics and improved VBPR model is proposed.As is described above,the tourist attraction recommendation system based on the traditional BPR model only considers users' ratings rather than the latent semantic information of attractions' images.Therefore,a novel tourist attraction recommendation system based on the stratified sampling statistics model and an improved VBPR model is proposed.The experimental results show that the average accuracy,the average recall rate,and the average F1 of the hybrid recommendation system,are improved about 3.4%,7.2%,and 6.4% compared to the best baseline model respectively.The data sparseness problem is further alleviated to some extent.Third,the tourist attraction recommendation system based on multi-modal visual bayesian personalized ranking model(MM-VBPR)model is proposed.The recommended performance of the modified VBPR model is obvious,but it doesn't fully exploit the multi-modal semantic correlation between different image features,which will affect the finally recommendation result.To address the problem,a novel tourist attraction recommendation system based on multi-modal visual bayesian personalized ranking model is proposed.Multi-modal analysis is carried out from two perspectives: the semantic correlation between different image features is deeply explored based on the start-of-the-art latent semantic analysis model.The image features are fused on basis of the semantic correlation mentioned above and they are imported into the modified VBPR model to complete the multi-modal analysis between attractions' images and users' ratings.The experimental results show that the average accuracy,the average recall rate,and the average F1 of the hybrid recommendation system,are improved about 1.2%,3.7%,and 3.5% compared to the best baseline model respectively.The proposed MM-VBPR model beats any other baseline and the data sparseness problem is effectively alleviated,which means the proposed recommendation system has very high practical value.