Application Of The Improved Collaborative Filtering Technology In Video Recommendation

With the development of Internet technology, nowadays the Internet has become such an important channel that makes all sorts of information and digital resources available for people all over the world. However, in spite of offering users nearly all kinds of information, the ever-increasing resources on most large websites force them to spend massive time on searching for what they really need. Therefore, how to help clients obtain what they want swiftly has become one of the key requirements for large resources websites.The proposal of the personalization recommendation technology is to bring about an effective solution addressing the above-mentioned requirement. By studying the hobbies and interests of each individual user, this technology can actively provide a kind of much more suitable recommendation service for the individual in due course. As a result, it serves as a mechanism assisting people in locating online resources quickly. The collaborative filtering recommendation technology, currently, is one of the most successful and most widely used personalization recommendation technologies in the world. This technology makes the recommendation of resources based on the similarity of interests between different users without considering the representation of various resources. Nevertheless, the traditional collaborative filtering recommendation technology still has some drawbacks in both accuracy and efficiency that need to be improved.Aiming at the features of online video resources, this thesis proposes some improvement methods on both accuracy and efficiency aspects of the collaborative filtering recommendation technology. Firstly, based on users watching behavior, this thesis proposes an implicit user rating collection strategy. This strategy can effectively solve several problems that have negatively influenced the recommendation accuracy of the traditional collaborative filtering technology to a large extent, including the sparsity of the rating matrix and the unreliable of the subjective ratings. Secondly, since the introduction of the time factor into the traditional recommendation model can reflect the changing pattern of user interests and item values along with the time respectively and further enhance its recommendation accuracy, this thesis proposes two reduction functions, namely the rating effectiveness reduction function based on the interest deviation and the video value reduction function based on video timeliness. Finally, this thesis proposes a context-based distributed collaborative filtering technology which includes a series of processing methods such as the rating matrix distributed storage method, the offline context information collection method and the online distributed recommendation algorithm. As a result, it can effectively solve the efficiency problem of the traditional collaborative filtering recommendation while securing its accuracy when it is applied to large and dynamic settings.The experimental results show that all the approaches proposed in the thesis can be more effectively applied to the personalization recommendation of online video by enhancing both recommendation accuracy and efficiency in comparison to the traditional collaborative filtering recommendation. In addition, these approaches can also be applied to other various online resources directly or after being partly modified. Thus, it has a wide range of application in practice.