Research And Implementation Of Personalized Hybrid Video Recommendation System Based On Feature Vectors

With the rapid development of Internet technology, more and more users choose to get video resources from the Internet service rather than to buy CDs from the video store or to wait for the broadcast of a program from television Station in a fixed period of time. Along with text information, e-commerce and other fields to obtain information, users are gradually aware of the current problem of information overload. To solve the problem, recommendation systems with data mining methods appear in many aspects of life. In the process of development of video recommendation system services, recommendation systems based on different recommended methods are gradually developing from experiment to commercialization. The statistics published from some early commercial video recommendation system show that the video websites with recommendation system can get better user data such as user residence time and user clicks. To provide a more accurate recommendation system service can bring higher benefits to video website, and it mainly depends on the reasonable recommended method in a recommendation system.In order to satisfy the requirement of personalized video recommendation system service from users in campus network, it mainly studies to provide a solution of a hybrid personalized video recommendation system which can be adapted to the campus network in this paper. On the one hand, it explains the main ideas of various recommended technology and analyzes its advantages and disadvantages; on the other hand, it summarizes the current research progress and the usual architectures of recommender systems. Then it puts forward the design scheme of a hybrid video recommendation system based on matrix transformation and association rules (we call MAR-HVRS in this paper) in campus network, and describes the implementation process of MAR-HVRS in detail in this paper. MAR-HVRS finds similarity or frequent patterns between users and video information based on the method with feature vectors and the association rules. Then, it gets the weighted combination from the predicted results of different recommended methods. This paper also puts forward a data preprocessing method CS-DP (Data Preprocessing based on Customer Segmentation) based on the customer segmentation thought, and this method is used in this recommendation system. It shows in the experimental results that compared with the traditional systems, the recommended results of the MAR-HVRS system with CS-DP have more kinds of video types, and the recall rate can increase by half maintaining the stability of accuracy.At present, many kinds of evaluation index can only reflect a few characteristics of a recommendation system, especially the evaluation of online recommended system. The system in this paper still needs some further research to improve the effect of the recommended results, for example, to adjust parameters in recommendation algorithms. There are some other aspects in this hybrid recommendation system we can continue to study, including modification or increase of the effective characteristics in the matrix model, the standardized measures for encryption and decryption in the data transmission process to improve security of the system, etc.