Research On Tourist Attractions Recommendation System Based On Hierarchical Sampling Statistics And Collabotative Filtering

Tourist attractions recommendation system is a major way for the tourism portal sites to recommend attractions to people.Presently,tourist attractions recommendation systems have obtained much progress.However,several problems still remain.First,“cold boot”is very serious.Second,the users'interesting isn't mined deeply.Last,the implicit preference of users isn't taken into account in generating recommendation results.To resolve these problems in some extent,the thesis applies the hierarchical sampling statistical model to better mine the users'interesting and it also modifies the traditional collaborative filtering algorithm.Based on the above work,the thesis designs a better tourist attractions recommendation system.The main works in the thesis are described as the following:Firstly,a new tourist attractions recommendation system based on the clustering algorithm is designed.The new recommendation system really helps to improve users'satisfaction and enhance both the influence and competitiveness of the tourism portal sites.More importantly,it helps to solve the“cold boot”problem in a certain extent.By using the combination of questionnaire survey and automatic crawling,the tourist data of users'information and users'scores are collected from the the tourism portal sites.The new dataset is called“smart tourism”.Based on the dataset,the system firstly preprocesses the users'ratings and then performs three kinds of clustering algorithms such as hierarchical clustering,FCM clustering,and K-means clustering.These clustering algorithms calculate the similarity between the target users and the clustering centers and generate a recommendation list named L_B.Finally,a new collaborative filtering based recommendation system is constructed by using these clustering algorithms.Experimental results show that the recommendation system based on both the collaborative filtering algorithm and the K-means clustering algorithm obtains the best recommendation performance among all models when the system set the clustering number k=8.Secondly,a new tourist attractions recommendation system based on hierarchical sampling statistics model and clustering algorithm is designed.The above recommendation system based on the traditional clustering algorithms only considers the similarity between users rather than users'implicit interesting.Based on the users'interesting of different user's attributes collected by the questionnaire survey,the hierarchical sampling statistics model is applied to better mine user's interesting.The hierarchical analysis method is applied in turn to set the weights of users'attributes.As a good result,a recommendation list named L_A is obtained.Then we set the clustering number k=8 and perform each clustering algorithm introduced above to complete recommendation.As a good result,a recommendation list named L_B is obtained too.Finally,L_A and L_B are mixed for hybrid recommendation.Experimental results show that:compared with the traditional method,the novel recommendation system acquires the best recommendation performance.Compared to the strongest baseline,the average precision,the average recall,and the average F1 value of the hybrid recommendation method improve about 14.3%,32.4%,27.1%respectively.The novel recommendation system gets better recommendation performance.Thirdly,a new tourist attractions recommendation system based on hierarchical sampling statistics model and SVD++algorithm is constructed.Most recommendation systems are based on the traditional clustering algorithms,which tend to fall into local minima more easily and do not consider the users'implicit preferences.As the reason,a new recommendation system based on hierarchical sampling statistics model and SVD++algorithm is proposed:based on the“smart tourism”dataset,the hierarchical sampling statistics model is applied to better mine the users interesting and a recommendation list named L_A is generated accordingly.A new collaborative filtering model based on SVD++algorithm is designed to mine the users'implicit preference and generate a recommendation list named L_C.Finally,L_A and L_C are mixed for hybrid recommendation.Experimental results show that compare to the strongest baseline,the average precision,the average recall,and the average F1 value improve about 7.5%,6.2%and 6.5%respectively.The recommendation system gets the best recommendation performance.The main innovations of the thesis:1)a new tourist attraction dataset named“smart tourism”is built for the corresponding research work.Based on the dataset,the users'interesting is mined better by the hierarchical sampling model;2)the mined users'interesting is fused into the clustering algorithm or SVD++algorithm based recommendation system to further improve the recommendation performance.