Research On Collaborative Filtering Recommendation Algorithm And Its Application In The Virtual Tour Of Han Chang’an City

In recent years,Internet and virtual reality technologies have been widely used in the field of cultural heritage protection and virtual tours,such as virtual tours in the Louvre Museum in France,the Notre Dame Cathedral,the Palace Museum in Beijing,the Terracotta Army in Xi’an and so on.However,how to recommend relevant cultural heritages that users are interested in from vast cultural heritage information is a challenge for personalized cultural heritage virtual tours.Based on the application scenario of the virtual tour of Han Chang’an City,the personalized recommendation method for cultural heritage virtual tour is studied in this thesis.Considering that the amount of data accessed by related users in different scenarios or different categories of Han Chang’an City varies greatly,no user data,data sparsity,data dense and other situations exist at the same time.This thesis focuses on the collaborative filtering personalized recommendation methods for data sparsity,cold start,and data dense problems,and applies them to the Han Chang’an virtual tour system.The main research work and innovations of this thesis are:For the problem of data sparsity in the virtual tour system,a user similarity model based on modified co-rated items is proposed in this thesis.Compared with existing collaborative recommendation methods,this thesis extends the dimension of the co-rated items’ vector space from the intersection items rated by two users to the union items.In the case of data sparsity,the data filling method is used to increase the number of corated items on the expansion space to ensure that co-rated items between the two users are always exist.The proposed model improves the utilization rate of historical ratings without increasing the time complexity,takes into account the impact of users’ global rating preferences at the same time,and solves the problem of co-rated items in traditional similarity calculation methods.Experimental results show that the model has low computational complexity,high scalability,good recommendation performance,meets online real-time recommendation,and is suitable for personalized recommendation for users on a relatively sparse dataset.For the cold start problem in the virtual tour system,the cold start problem is divided into an incomplete cold start problem and a complete cold start problem in this thesis,and the corresponding two recommendation methods are proposed.One is based on two heterogeneous feedback data and the other is based on multiple implicit feedback data.Compared with existing methods,these two methods aim to improve the recommendation quality of the virtual tour system by extracting common latent features of users and items,which effectively suppress the problem of excessive noise introduced by the existing methods in the process of extracting user and item features.For the problem of incomplete cold start,this thesis maps explicit rating data and implicit feedback data to the same model,and a Bayesian Personalized Ranking model based on historical rating data is proposed.This model also incorporates Bayesian Personalized Ranking model and Probabilistic Matrix Factorization model,realize the sharing of user feature space and item feature space.For the problem of complete cold start,this thesis fuses implicit feedback data together and a Bayesian Personalized Ranking model based on multiple feedback data is proposed,aiming to extract more user preferences and item characteristics from implicit feedback data.The experimental results show that the proposed models in this thesis can alleviate the cold start problem to a certain extent.For the virtual tour system with dense rating data,a new collaborative filtering recommendation approach is proposed in this thesis,which based on the fusion of explicit feedback and implicit feedback information.Compared with existing recommendation algorithms,this thesis based on the assumption that users prefer items with higher ratings,the definition of explicit rating data is added to the Bayesian Personalized Ranking model,and an improved Bayesian Personalized Ranking model is proposed to extract the implicit feedback features of users and items,the model makes full use of the existing feedback data and introduces negative feedback data,which alleviates the problem of lack of negative feedback in the implicit feedback data.This thesis further adopts a matrix factorization model to extract explicit feedback features of users and items,and the predicted rating data are used as a supplement of the ranking rating data to improve the ranking performance of the proposed model.Experimental results show that the proposed algorithm can significantly improve the recommendation accuracy of the system in a low sparsity dataset,and reduce the impact of noise introduced when training unrated items as negative feedback data to a certain extent.Combined with the virtual reality technology,the virtual reconstruction of the Han Chang’an city ruins was realized,and the Han Chang’an city virtual tour system based on the above collaborative filtering recommendation approaches was also realized in this thesis.The system has been published on the Internet,and users can experience it in the virtual Han Chang’an City through the system’s website link.The system can provide users with personalized recommendation services to achieve resource sharing.