Research On Hybrid Recommnendation Algrithm Of Collaborative Filtering And Attribute Association Rules

With the rapid development of internet technology,e-commerce has gradually entered into public view and become an indispensable part of people’s daily life.The development of e-commerce must be accompanied with the increasing number of users and items,and the number of items is far greater than the number of users.At this time,how to quickly find the satisfactory items for users in the massive data has become particularly important.Under this background,personalized recommendation technology comes into being,it provides users with personalized information service and decision support by analyzing user behavior.The user-based collaborative filtering recommendation is one of the most applied algorithms in personalized recommendation.The traditional user-based collaborative filtering algorithm has the problem of data sparsity,which leads to the poor results of the final recommendation.In this thesis,a improved collaborative filtering method based on user similarity is proposed.,this method has introduce the Bhattacharyya coefficient and the item share score factor into Pearson formula,the local similarity of the users is calculated by Bhattacharyya coefficient and the Pearson similarity,and the global similarity of the users is calculated by the proportion factor of the item common score.The improved Pearson similarity can effectively reduce the error caused by data sparsity in calculating user similarity.For traditional Apriori algorithm has the problem that it is necessary to scan the transaction database many times and if the item is not in the rules,it can not provide effective recommendation,a item attribute association rule recommendation algorithm based on matrix is proposed.This algorithm convert user historical behavior into a matrix form,and excavate the potential association between the attributes of a item to provide a recommendation for user.For the new user problem existing in the traditional user based collaborative filtering algorithm,the association rule recommendation algorithm can get the potential relationship between the items by mining the shopping habits of the platform consumers,and can effective improvement the problem that the user can not recommend the user when the user’s behavior is scarce.Therefore,a hybrid algorithm based on user based collaborative filtering and association rules is proposed in this thesis,and the two recommendation methods are mixed with Top-N algorithm to achieve the final recommendation.All in all,this thesis improves the Pearson similarity computation method,makes it adapt to the problem of data sparsity well.On the basis of traditional Apriori algorithm,a recommendation algorithm based on item attribute association rules is proposed,which effectively makes up for the problem that the number of association rules is few and is difficult to be recommended.Experiments show that these two improvements can effectively improve the recommendation performance.Finally,the two recommendation algorithms are combined to achieve the final recommendation in a mixed way,which can effectively improve the problrm of data sparsity and cold-start.