Study On Collaborative Filtering Algorithm Based On Clustering And Its Recommendation On Civil Aviation Additional Service

With the fast development of Internet, shopping online has become much more widespread. For online retailers, in order to improve their competence, they have to provide personalized electronic commence recommendation system to their customers. However, with the sharp increase of effective users and merchants, collaborative filtering recommendation technology, as the most widespread and steady technology, has been faced with great challenges and lots of problems have surged out such as sparsity, cold boot, real-time etc.To solve those problems existed in system of recommendation, this thesis presented two algorithms based on clustering technologies which are much popular in recent years. The first algorithm is the improved collaborative filtering algorithm based on mixed clustering technology. In this algorithm, we presented some improved strategies to solve the problem of low recommendation accuracy. On one hand, we added the neighbor items in the similarity calculation and introduced the balanced factor ? to adjust the rate of neighbor items. On the other hand, we presented the concept of user membership which could decrease the dimension of the Score Matrix of User-Project. By taking the public databases as the input of experiments, we got the reasonable result showing the obvious improvement of recommendation accuracy for this improved algorithm. The second algorithm is a new kind of collaborative filtering algorithm based on networking partition to solve the problems of low recommendation accuracy and poor system scalability in traditional algorithm. In this new algorithm, by taking the social network of users into consideration, we designed the users' network of relationship and presented the concept of users' similarity of relationship. Besides,this algorithm combined the Louvian community partition algorithm with collaborative filtering algorithm. We test this new algorithm using Last.fm database and the result shows that when the near number is less than 25, the recommendation accuracy of this algorithm is obviously higher than that of the traditional algorithm.Finally, we applied these two algorithms to the service of booking ticket in civil aviation system. By data analysis, we designed the travel preference model of passengers for the choice of airplane companies, ticket discount, travel time and seat choice, which can provide personalized service for every passenger.