Research On Collaborative Recommendation System Based On User Clustering

Now, electronic commerce recommendation system has become an important platform for a lot of enterprises to advertise and vend goods. However, a number of information about relevant commodities make consumers difficult search for what they like accurately and rapidly. Therefore, how to design more personalized electronic commerce recommendation systems so as to serve users much better has become a significant research project for many firms. So far, almost all famous e-commerce sites, such as ebay and taobao, have adopted personalized recommendation systems to some extent. In order to recommend goods to users quickly and accurately, researchers have put forward many different recommendation techniques, such as collaborative filtering recommend technology, the bayesian network technology, clustering technology, singular value decomposition technology, the connection mining technology, etc, and the collaborative filtering recommend technology is used widely. But, there are some issues that occur with the use of it, for example, "cold start" issue, "data sparseness" issue and so on.This paper does some researches about the collaborative filtering recommendation technology to solve the problems of "cold start" and "data sparseness" existing in the collaborative filtering recommendation algorithm, and what this paper does can be as follow:1. Summing up the history and the current development of the personalized recommendation system, and pointing out the two bottleneck problems of "cold start" and "data sparseness" existing in the collaborative filtering recommendation algorithm.2. Putting the ant clustering algorithm into the collaborative filtering recommendation algorithm.3. In order to solve the "cold start" issue, this paper proposes a way to improve algorithm. The new algorithm considers not only user preference similarity among the users, but also the objective characteristic similarity among users. This can make the system recommend users who do not conduct subjective assessment.4. Adopting the ant clustering algorithm clusters new users, then, using the improved collaborative filtering recommendation algorithm searches for neighbor users to assess in order to reduce the search scope of the neighbor users and system's computation time. Meanwhile, it also can strength the system's quality to recommend and relieve the influence of the "data sparseness" problem to some extent by clustering.5. According to the theory and method proposed in this paper, a systematical imitation is conducted.