Research On Scalable Collaborative Filtering Algorithm Based On User Profile

Throughout the development of human information delivery,the emergence of the Internet has enabled people to access more information,and the huge amount of information has caused the problem of information overload.With the development and research in academia and industry,recommendation systems have emerged,which can not only recommend personalized content for users,but also greatly reduce the time for users to search and find content,and are widely used in many fields such as travel,movies,music and e-commerce platforms,and are loved by users and suppliers alike.The core content of the recommendation system is the recommendation algorithm,the common recommendation algorithm is the memory-based collaborative filtering recommendation algorithm,the algorithm faces the problems of sparsity,cold start problem and scalability problem,this paper mainly focuses on the sparsity problem and scalability problem,do the following work:(1)Traditional similarity algorithms only use user behavior information to calculate similarity,due to the problem of data sparsity,which leads to its calculation results are not accurate enough and indirectly affects the prediction results of the algorithm,in order to alleviate the problem of data sparsity,this paper designs a user profile similarity algorithm,which firstly makes full use of multiple sources of easily accessible information to construct user profile labels;then,on this basis,calculates multiple The algorithm first makes full use of multiple sources of easily accessible information to construct user profile labels;then,based on this,multiple similarities are calculated from different perspectives;finally,they are fused into user profile similarity.The experimental results on Movie Lens dataset show that the similarity algorithm proposed in this paper predicts better results.(2)Traditional collaborative filtering algorithms usually use dimensionality reduction or clustering to improve scalability,but dimensionality reduction will lose information and lead to degradation of prediction quality;while clustering consumes extra time to construct data and does not even improve time efficiency.In order to avoid the above disadvantages,this paper proposes a scalable method that merges LANDMARKS and KMeans,which uses LANDMARKS matrix for KMeans clustering,not only does not generate extra data construction time,but also reduces the number of user similarity calculations,and the experimental results on the Movie Lens dataset show that this method,compared with the current The experimental results on the Movie Lens dataset show that this method can not only improve the scalability of the algorithm more effectively than the current commonly used scalable methods,but also achieve a prediction quality that is not weaker than the original algorithm.(3)Traditional collaborative filtering algorithms often increase the time complexity and reduce the scalability of the algorithm when alleviating the sparsity problem.In order to be able to alleviate both the data sparsity problem and the scalability problem,this paper fuses the proposed user profile similarity method and the scalability method to obtain a scalable collaborative filtering algorithm based on user profile,and the experimental results on the Movie Lens dataset show that the algorithm outperforms the current mainstream collaborative filtering algorithms in terms of time efficiency and prediction quality.