A Tag-based Double Direction Collaborative Filtering Model

With the exponential growth of the network information, it has long been the lack of the required information. How to get the information people needed precisely from the flood of information has become a common problem. So a lot of new technology appears, such as information extraction, information index, information filtering. YouTube and Youku, contains tens of thousands of video, they allow ordinary Internet users to upload their own interest videos, which not only mobilize the enthusiasm of Internet users, but also improve the site’s visibility and effectiveness. But how can users find the information they need from the vast amounts of video information is also a problem. The users often in this situation:if they want to identify something using a search engine such as Google or Baidu or some other way, the returned information can often reach hundreds of pages. There are still a lot of circumstances being submerged in the ocean of the vast information. The user has to pay the time and effort, but to find a few (or even no) wanted. Personalized recommendation system is produced in the context of the knowledge economy era of rapid development of the Internet and the explosion of information, which has become an indispensable online support system in people’s daily life. It originated from the actual life. As the saying goes, feather flock together, people in groups. People tend to access to information and transfer it through their own trust or friends have common taste.This article briefly describes the necessity of the personalized recommendation system research, and then introduces the status quo, including both domestic and abroad. The development of recommendation system abroad has make Google, Amazon industry giants. The late start of personalized recommendation system in our county, but has showed a booming trend. Personalized recommendation technology has made a lot of achievements, but also far into the mature stage, which is still facing many challenges.The recommendation system is a black box, external users do not know its process. Recommendation system select some of the potential products of the ranked users loved (list, pictures, etc.) presented to the user through certain recommendation algorithm.Main recommendation algorithms are collaborative filtering algorithm, content-based filtering algorithm, algorithm based on the recommendation of the bipartite graph structure and hybrid recommendation algorithm. Each recommendation system has its own application scenarios and areas of expertise. This paper proposed a tag-based double direction collaborative model, DCMu. The items user rated could be divided into two classes though social tags. We could extract positive and negative patterns though digging both positive and negative to the tag information, then the initial user model Mu is formed. And reciprocal between the users is calculated on the basis of the user’s initial model similar correlation. Model DMu is formed by enrich model Mu though user’s similar neighbors’models. Model DCMu is the combination of Mu and DMu. Finally, we compare the algorithm with two baseline algorithms in recommended accuracy and the precision of the sort of recommended list. DCMu algorithm is better than two benchmark algorithms in both standards.