Videos Recommend System Based-on Collaborative Filtering Technology

The development of the internet brings people from the real world to a virtual world. It has brought great wealth to the human beings, meanwhile it has also created a lot of useless information. The net citizens around the world create a large amount of information in their own ways and this leads to the rapid increase of the internet information, which means the human beings have stepped into the information explosion society. In such kind of situation, the cost will be higher and higher if people want to choose the information they want. How to get the information people want in a low cost has been a very important subject in the field of internet research for a very long time. By learning and collecting the user’s information, the personalized recommender system provides the users with precise information recommendation service and it can also cost less for the people to get the information which they want. Hence, the individual recommender system is a very important information dissemination tool.In the internet information dissemination, the video dssemination is a very important way. Since Tudou was founded in2005, until now, the number of the videos which have been uploaded into this website amounts to more than50,000,000, and it is still increasing by more than five hundred thousand each day. Among so many videos, how to find the users’favorite ones in a convenient way is really of great importance for the users during the internet surfing. Based on the exsiting data, this thesis aims to design a set of video recommender system named TDRecsys for the Tudou users. By using the frequently used collaborative filtering technology, the designed TDRecsys recommender system filter the input data, i.e. the video data of the whole website and the users’behavior data, so as to get rid of the data which may disturb the recommended results. Based on the User-base and Item-base model, it will use the similar formulas to calculate the similarity of the videos and of the users. As the original data of the recommeder system, the users’ watch records will be input into the system. According to these watch records, the recommender system will calculate the users’ possible favorite videos and then recommend to the users.The main tasks of this thesis are as follows: First, to analyse the data characteristic of Tudou and propose the solution to the common problems, such as the cold start, data sparsity, diversity and accuracy which will appear during the implementation procedure of the recommender system.Second, to design the quality control and VideoRank sorting algorithm and select the method which can promote the system recommender accuracy, such as the noise filter of the data, the selection of the recommened video collection and the mixed recommendation.Third, to design and realize the online and offline module of the TDRecsys system, and improve the evaluation method of the recommender system.The TDRecsys system developed in this thesis has been applicated in the "recommended to me" and the play page in Tudou’s home page. Having been proved by actual practice, the TDRecsys system can effectively improve the corresponding window’s click-through rate, that make users watch Tudou’s video longer so as to develop Tudou’s attraction and users’ satisfaction.