Research On Cold-start Problem Of Collaborative Filtering Algorithm

The rapid development of Internet technology has made the society enter the era of big data and caused information explosion problem.It is a research problem to obtain valuable information from massive information data quickly.Personalized recommendation can be based on the user's behavior data to mine the user's preference types,thereby effectively reducing the information overload problem and improving the use of information.Collaborative Filtering recommendation technology is the most successfully and widely used algorithm in personalized recommendation.The algorithm is based on data mining.By calculating the similarity of user's rating to items,the neighbor of the target user or target item is found,then the recommendation process can be completed.In this paper,two improved collaborative filtering algorithms are proposed for the cold start problem,the recommendation for new users and new items is completed without historical behavior records.(1)Collaborative Filtering Algorithm Combined with User's Attributes and Back Propagation Network,UA-BP-CF algorithm analyzes the similarity of user's attributes as the input neurons of BP network,then the neurons are weighted and summed to obtain the value of the output neurons,which is used as a basis for measuring the similarity of new users,thereby solving the cold start problem of new users and completing the recommendation.(2)Collaborative Filtering Algorithm Combined with Item's Contents and Hierarchical Clustering,IC-HC-BP algorithm analyzes the correlation between project contents,calculates the Euclidean distance between objects based on the content information.Through the hierarchical clustering method,the algorithm finds the neighbor of the target item,calculates the predicted score,and recommends for the new project.This paper selects the open source data set--Movie Lens to verify the improved algorithm,sets up multiple basic experiments to find the best value of parameters,and compares thedifferences between the improved algorithm and the traditional one.The results of the experiment tell us that the new algorithm has higher recommendation accuracy for new users and new items.It can alleviate the cold start problem of the system and improve the recommendation quality of the algorithm.