The Research Of Cold-Start Problem In Collaborative Filtering Recommender System

With the advancement of the internet and e-commerce, internet provides an unparalleled opportunity for users to achieve great number of information services. Such a situation has induced the so-called information overload problem which leads users finding it increasingly difficult to locate the right information at the right time. Therefore, both researchers and consumers focus on providing more accurate individual information in less time to meet the personalized needs. Personalized recommendation has become a desirable requirement under this background. Currently, recommender systems have proven to be an effective technology that delivers the consumers a more intelligent and proactive information service by sifting through all the available information to find that which is most valuable to them. One of the most successful such technology is collaborative filtering (CF). However, the tremendous growth in the amount and variety of available information leads to some austere challenges to recommender systems。Cold-start problem is the most serious problem for collaborative filtering that has not been effectively addressed. As we known the fundamental assumption of CF is that CF analyzes rating matrix to recognize commonalities between users on the basis of their historical ratings, and then generates new recommendations based on liked-minded users'preferences. However, the recommender system can't provide effective recommendations for new user or new item because they have not enough ratings available. Fortunately, quite a number of personalized recommendation systems have collected content information about users and items. Inspire by this reality, we make use of the user or item content information to improve the traditional collaborative filtering.The main contributions of this dissertation are as follows:(1) We propose an algorithm framework to address the cold-start problem for collaborative filtering. The key idea of our algorithm is that we first cluster the users or items based on the existing item-user ratings, and then utilize the clustering results content information that the recommender system provides to build a partition model which can associate the novel users or items with the existing ones. When a new user or item has just entered the system, the partition model appoints the new user or item to a certain cluster. Combining the algorithm proposed with the traditional collaborative filtering technology recommendations can be achieved. Initializing the framework with the special method and contents, the algorithm framework can address both the new user and the new item problem.(2) Actualize the algorithm framework with the corresponding approaches. Considering the existing user-item ratings is quite sparse, we impute the missing values before clustering. We perform the K-means algorithm on the imputed matrix. However, k-means algorithm is sensitive to the center initialization. Aiming at dealing with the center initialization problem, we utilize an optimization strategy by taking the users or item whose rating numbers is relatively larger and the mean rating error is quite smaller as initial centers.We carry out a series of experiments to examine the superiority of our algorithms in addressing the cold-start problem. Comparing with traditional collaborative filtering algorithms and the existing algorithm which is popularly used for solving the cold-start problem, the experimental results show the availability, correctness and effectiveness of the new algorithm in tackling the cold-start problem.