Cold Start Problem In Recommendation System For New User

In Internet era,users can easily obtain many kinds of information,but they must spend much time to find the most useful part,so personalized recommendation algorithms were born.Among them,collaborative filtering recommendation algorithm is used widely.It relies on user historical behavior data to discover user preferences,then divides users into different groups with similar preferences and finds neighbor user set,and finally recommends items for target users with the help of neighbor users.However,when new users are firstly added to the system,because there is no historical record,so the system cannot make recommendations for them,which causes the cold start problem.This problem reduces system accuracy and user experience,so how to solve this problem is of great significance.To solve cold start problem,a user-based recommendation system is designed and implemented.This system first preprocesses user behavior data,then uses K-means method to cluster users with similar preferences and factorizes the user-item rating matrix to reduce the complexity of the similarity calculation by the principal component analysis method,and finally establishes a user comprehensive similarity calculation model to calculate the similarity between users,so the system can produce a candidate list for users.Experiments are conducted on real user datasets.The results prove that new system solves the cold start problem for new users and achieves higher accuracy.Compared with the traditional system,the mean absolute error is reduced by about 10%.