Research On Personalized Recommendation Algorithm Based On Collaborative Filtering

With the rapid development of the Internet and the increasing number of Internet users,the amount of data is exploding on the Internet.The phenomenon makes it increasingly difficult to find the information of interest in huge data with limited time,so the personalized recommendation system comes into being.As the core of recommendation system,recommendation algorithm directly determines whether the recommendation system can give users a good experience.The collaborative filtering recommendation algorithm is one of the more successful recommendation technologies in the field of recommendation.Based on the traditional collaborative filtering recommendation algorithm,the paper improves the algorithm according to the accuracy of recommendation results,the rationality of item ranking in recommendation list,and the sparseness of scoring data.Firstly,in the calculation of user similarity based on the jaccard coefficient,the calculation method of the jaccard coefficient is improved by taking into account the proportion of the items that users have visited together in their respective browsing items.Secondly,the list of recommendation results generated by the recommendation algorithm is reordered by using the user-based voting sorting algorithm.Factors such as project type,the time spent browsing each type of project recently and users' predicted score on the project are taken into account when users vote.Then,for the sparsity of scoring data,SVD model and Slope One model were fused by model fusion method,and the missing scoring information in scoring data was predicted and filled in by the fusion algorithm.Finally,the improved algorithm is experimentally verified on the Movielens public data set.The experimental results show that the accuracy rate,recall rate and coverage rate of the improved jaccard coefficient calculation method are improved by 0.61%,0.28% and 2.37% compared with the original algorithm.The user voting algorithm is adopted to reorder the list of recommendation results to make the recommendation results more reasonable.Compared with the best algorithm involved in fusion,the fused algorithm reduces the root mean square error by 0.68%and the mean absolute error by 0.41%.