Research On The Cold-start Problem Based On User Preferences And Demographics

In today’s era of data with a wide variety and large scale,whether it is e-commerce systems such as Taobao and JD.com,or searching and information apps such as Baidu and Toutiao,the recommender system has become the core part of them,as well as one of the most competitive parts.It is of great researching significance and practical value to improve the recommendation effect of the recommender system,and the problem of user cold-start is an inevitable and key concern.This paper proposes corresponding solutions for the user’s impure cold-start problem and the user’s pure cold-start problem.For user’s impure cold-start-problem,a user impure cold-start recommendation algorithm(UPMKC-UIC)based on user preference matrix and K-means clustering is proposed,which is used to reduce the negative impact of non-pure cold-start problem on the recommender system.The UPMKC-UIC algorithm first builds a UPM according to the number of user ratings and records of items,and then finds the nearest neighbors corresponding to the users who need to be recommended through two screenings,and finally forms the recommendation results with the obtained preference items of the nearest neighbors to the user.Through experiments on the Movielens-100 K dataset and Movielens-1M dataset,the UPMKC-UIC algorithm is compared with the Jaccard algorithm and the NHSM algorithm.The experiments show that the UPMKC-UIC algorithm can effectively alleviate the user’s impure cold start problem in the recommendation system.The impact of the recommendation system can improve the recommendation effect of the recommendation system.For user’s pure cold-start problem,a user pure cold-start recommendation algorithm(DJSKCUPC)based on demographic information and K-means clustering is proposed,to alleviate the impact of the user’s pure cold-start problem on the recommendation system.The DJSKC-UPC algorithm introduces demographic information to construct a joint similarity calculation method suitable for pure cold-start users.Combined with the K-means clustering idea,the joint similarity method is used in similar clusters to calculate the similarity between the target user and each user to find the nearest neighbors of the target user,and finally assign the average score of the nearest neighbor items to the target user to complete the recommendation.Through experiments on the Movielens-100 K data set,the DJSKC-UPC algorithm is compared with the UBCF,UCKF and SRCF algorithms.The experiments show that for pure cold-start users,the DJSKC-UPC algorithm still has high recommendation efficiency,which can effectively reduce the impact of the user’s pure cold-start problem.