Research And Application Of Recommendation Algorithm Based On Collaborative Filtering

With the advent of the information age,the degree of popularity of the Internet is rapidly increasing.While enjoying diversified network information to provide more convenience for our lives,massive data resources make it more and more difficult for users to find information of real interest.Aiming at the problem of information overload,the emergence of the recommendation system provides a solution.As one of the most widely used technologies in the system,collaborative filtering recommendation uses the user's historical evaluation information to mine the user's hidden interests and hobbies,and uses similar users or items to make personalized recommendations.Although the traditional collaborative filtering recommendation algorithm has been widely used,there are still problems such as cold start and sparsity.Aiming at the above problems,this thesis conducts in-depth research on the collaborative filtering recommendation algorithm,and proposes corresponding improvements to the deficiencies in the algorithm.The main contents of this thesis are as follows:1.Improve the calculation method of similarity between users.The traditional calculation of similarity between users is mainly based on the scores of common scoring items between different users,and the weight of each item's score is the same.However,the weight of highscoring items in the user's common evaluation set should be greater than the weight of lowscoring items,and the ratings of popular items and unpopular items should be assigned different rating weights.In response to this situation,this thesis proposes a variable weight similarity calculation method,which improves the accuracy of scoring prediction and the quality of recommendation results on the basis of improving the accuracy of finding neighbor users.2.Improve the traditional Slope One algorithm.The main idea is to introduce user relationships in the Slope One algorithm,and use the scores of the set of neighboring users as a reference.The traditional Slope One algorithm ignores the similarity between users,and the user set selected when calculating the deviation between items will introduce the interference of noise data.In this thesis,combined with the proposed calculation method for improving the similarity between users,the user similarity relationship is introduced into the Slope One algorithm,the selection method of the user set is improved,and the rating data of neighbor users is used to improve the accuracy of the algorithm's predictive rating.3.Aiming at the common sparsity problem in recommendation algorithms,this thesis proposes an improved matrix factorization algorithm.For users whose scores are less than the threshold,the algorithm fills in missing items through the improved Slope One algorithm,which mitigates the impact of sparsity to a certain extent,and reduces the interference of noise data introduced by the traditional SVD algorithm based on mean filling on the predicted score.Then combined with the idea of LFM decomposition matrix,using stochastic gradient descent method to analyze and improve the model,and improve the accuracy of the final prediction score.4.Conduct comparative experiments on the improved algorithm on the MovieLens data set to determine the optimal values of the parameters in the algorithm,and then compare the improved algorithm with the traditional collaborative filtering recommendation algorithm through commonly used evaluation indicators to verify the effectiveness of the algorithm in improving the accuracy of the prediction.5.Take the collaborative filtering algorithm proposed in this thesis based on improved user similarity calculation as the core,apply it to the actual system design.Then use the SSM framework and the MVP framework to design and implement a music recommendation system.