Research On Collaborative Filtering Algorithm Based On Subjective And Objective Similarity Measure

With the development of the Internet technology and e-commerce application, personalization of e-commerce recommendation system is becoming important research content in e-commerce, and attracting more and more researchers'attention. Among them, the collaborative filtering technology is currently the most widely used and successful personalized recommendation technology, which is becoming the focus of research in the field of personalized recommendation. Thus, this study focused on the collaborative filtering technology.In the collaborative filtering technology, this study analyzed the current problems in conventional technology. With increasing of the numbers of users and goods in e-commerce system, the user-based collaborative filtering technology come up with insurmountable problems in real-time recommendations because time consumption of online users for this algorithm will increased with the increasing of number of users and number of commodities. Although some scholars proposed to improve real-time responsiveness by clustering, dimension reduction methods and so on, the time bottleneck is still not completely resolved with the cost of sacrificing the accuracy of recommendations. Therefore, the development of other item-based collaborative filtering becomes necessary. The item -based collaborative filtering method computing the similarity between the items but not the users, and the similarity of the item can be calculated offline, thus, it can fundamentally solve the real-time issues. However, as the user rating data on the project space is extremely sparse, the item-based similarity calculation formula requires a large number of users, that is, the user need to give scores for both of the participating two items. Apparently, the sparsity of the rating matrix resulting in that the user rating influences the accuracy of the recommendation to a large extent. In addition, the traditional item-based similarity measurement method does not consider the impact of characteristics of the item on similarity among the items. Thus, this method not only has problems of inaccuracy of the recommendation results, but also cannot solve problems in new items. Therefore, the study proposed a new method to calculate the similarity among items which aims to solve the existing problems in collaborative filtering recommendation techniques. This method is comprises of two aspects to calculate the similarity among the items.The first aspect is to modify the traditional algorithm which is based on the user rating to get new subjective similarity formula.The second aspect is to classify the properties of the items. By putting together several properties of items which have the largest influence to get the objective similarity formula based on the item itself, and finally integrate the two parts to get the new formula. This method expects to improve the accuracy of the items similarity calculation and to some extent solve the new item problem at the same time.Finally the experiments verify the accuracy and effectiveness of the proposed method.