The Research On Collaborative Filtering-based Personalized Recommendation Technology

With the development of artificial intelligence technology,there are more and more ways for people to get information,and it also becomes more and more convenient to get information.At the same time,the amount of information generated is also increasing exponentially,and the problem of “information overload” is followed.People's personalized needs are hard to be satisfied.Personalized recommendation system is the most effective way to solve “information overload” and has been applied to relevant fields.The collaborative filtering recommendation algorithm has become one of the most successful recommendation technologies in the personalized recommendation system because it is simple and easy to implement and its cross-domain.Based on the relationship between the users and the projects,traditional collaborative filtering algorithm recommends the projects that interest users.However,the data sparseness problem and cold start problems will seriously affect the accuracy of recommendation.Therefore,it is necessary to make full use of relevant information,and many researchers solve the above problems by means of dimensionality reduction,matrix decomposition,data filling and other methods.Aiming at the problems above existing in the personalized recommendation algorithm,this article focuses on the problems about data sparseness and diversity of data to the users that traditional collaborative filtering algorithm is facing and put the project properties information into the similarity calculation.In order to improve the accuracy of recommendation,this thesis improves the traditional collaborative filtering algorithm and hybrid personalized collaborative filtering recommendation algorithm is put forward at the same time.The main work includes:First,considering that the accuracy of traditional collaborative filtering algorithm is not high when calculating the similarity calculation caused by data sparseness problem,the paper puts into user features and project information when calculating the similarity,and accurately calculate the nearest neighbor set between users and the project.Therefore,it solves the problem of low recommended accuracy caused by sparse data.Second,this paper will mix the improved traditional collaborative filtering algorithm,matrix factorization algorithm and the matrix decomposition algorithm with the linear regression model and the fusion XGBoost model.A hybrid personalized collaborative filtering recommendation algorithm is put forward.Consider the users' rating of project as the input of linear regression model and the XGBoost model and predict the unknown ratings according to the trained model.The scores of the two models are combined and the Top-N algorithm is used to generate recommendation list.Third,in the open MovieLens data set,the improved collaborative filtering algorithm and the hybrid collaborative filtering personalized recommendation algorithm are tested and verified.Comparing with other algorithms,the proposed hybrid optimal personalized collaborative filtering recommendation algorithm,followed by improved collaborative filtering algorithm,both of them are superior to the traditional collaborative filtering algorithm as to the accuracy.To a certain extent,the quality of the recommendation is improved.