Research And Application Of Collaborative Filtering Recommendation Algorithm Based On Clustering And Time Factor

Today,it becomes extremely difficult for people to get the data they need out of massive amounts of data,a phenomenon known as "information overload".Personalized recommendation is another effective solution to the problem of "information overload" after keyword-based search engine information retrieval technology.It taps out users’ interest preferences based on existing data and provides users with personalized recommendation services.The traditional collaborative filtering algorithm is the most representative algorithm recommended by individualization,and its principle is easy to understand.But there is still much trouble,such as high computational cost,ignoring time factor affecting user interest,and sparsity of scoring matrix.Therefore,aiming at the above problems,this article is to improve the traditional collaborative filtering recommendation algorithm and apply the research results.The specific work is as follows:A dynamic collaborative filtering algorithm based on popularity clustering and time decay,named PTDCF is designed.Through the analysis of the traditional collaborative filtering algorithm,it is found that the difference of score will lead to higher computing cost when selecting the nearest neighbor of the target project.Aiming at this problem,PTDCF algorithm references the idea of popular recommendation and clustering,selects the cluster center according to the scored times,and uses cosine similarity formula to calculate the similarity between each item and each cluster center.After clustering,the nearest neighbor is searched in the clustering,which reduces the number of iterations and search candidates,and further improves the search efficiency.The traditional collaborative filtering recommendation algorithm ignores the influence of time factor on user interest,which leads to the problem that the recommendation accuracy needs to be improved.PTDCF algorithm chooses Ebbinghaus function as time attenuation function,and introduces time attenuation factor to calculate item similarity and predict user score,so as to reflect the change of user interest and improve the accuracy of recommendation.Considering that in the clustering of the PTDCF algorithm,different cluster centers maybe belong to the same type of items,and due to the sparsity of the scoring matrix,some items may not be recommended because they cannot be selected as the nearest neighbor.This thesis further designs a collaborative filtering algorithm based on item attribute information and time factor is designed,named IATCF.First,it uses the PTDCF algorithm to form clusters,and calculates the similarity between each cluster center item according to the number of the same attribute values.The clusters whose similarity is higher than the threshold is merged,and the cluster center with more evaluation times is used as the merged cluster center.Second,when calculating item similarity in clustering,time decay function similarity and similarity based on item attributes are combined to form a comprehensive similarity.Compared with the PTDCF algorithm,the IATCF algorithm makes the recommended items more comprehensive through merging clusters,and makes the recommended items more targeted through using the comprehensive similarity.In order to check the performance of the designed algorithms,Movie Lens dataset and Delicious dataset are used for the experimental verification of each algorithm.The experimental results show that algorithms designed in this thesis can effectively improve the accuracy and diversity of recommendation.In order to verify the usability of the research results,a simple book recommendation system is developed,and IATCF algorithm is used to generate recommendation information in the system.The application results show that the research results are practical and effective.