Research On Cold-start Problem Of Collaborative Filtering Algorithm

With the popularity of internet, online shopping is blended into people’s daily life with the rapid development of e-commerce. Vast quantities of user information and product information in the electronic commerce system lead to the “information overload” problem, so users cannot get information required with limited energy in the limited time. Personalized recommendation system as an important tool coping with “information overload” problem has also become a focus of attention. However, collaborative filtering is the most successful recommendation technique and it has been widely used in practical applications. In spite of extensive application of collaborative filtering, it is confronted with many challenges, for example, unsatisfied quality of recommendation due to the problem of data sparsity, Scalability and cold-start.This paper does deep researches on the cold-start and data sparsity problem. Based on traditional collaborative filtering algorithm, cold-start problem is alleviated by analysis of user character similarity and item semantic similarity. The main work of this paper has following several aspects:Firstly, aiming at the shortages of similarity measurement in the traditional user-based and item-based collaborative filtering algorithm, user similarity measure method for different items and item similarity measure method for different users are put forward. It can be found the appropriate nearest neighbor for users and items.Secondly, On the basis of improved user similarity measure method, user characteristic similarity calculation model is introduced to alleviate the new user cold start problem. User characteristic similarity and improved user similarity are effectively combined by introducing dynamic balance parameter, and the ratio of the two can be adapted dynamically to the change of rating matrix. Nearest neighbors can be found to generate recommendation by characteristic similarity between users when new user entering the system.Thirdly, in order to alleviate the cold-start problem, the paper presents a collaborative filtering algorithm by combining item semantic and user characteristics. First, the method combines the similarity of user rating with user characteristics to get the user neighbors and calculate the user forecast rating. Meanwhile, it combines the similarity of item rating with item semantic to get the item neighbors and compute the item prediction rating. Then, the final recommendation is got by combining user forecast rating with item prediction rating. Since this algorithm takes user characteristics and item semantic into account simultaneously, new user cold start problem and new item cold start problem are both mitigated.Finally, the experiments are carried out by adopting dataset exposed by Movielens site. The experimental results show that the effect of modified similarity measure method is better. This algorithm can alleviate data sparsity, by way of reducing cold-start problem and increasing prediction accuracy.