Research And Analyse On E-commerce Personalized Recommendation Based On Collaborative Filtering

With the growing popularity of the Internet and e-commerce, the amount of data also will be increased exponentially. Against the problem of “information explosion”, the researchers propose the personalized recommendation algorithms, which focus respectively. But among these algorithms, the most successful one is collaborative filtering recommendation algorithm. The scoring information is provided by the users, which is used for data mining, selecting the nearest neignbors which have high similarity with the target user(or item). It predicts scores and generates the recommendations from the set of the nearest neignbors. However, due to high dimensional data, sparse of the scoring information and inaccurate calculation of users’ interests, which result in the deterioration of the accuracy of recommendation, the decrease of real-time performance and the cold start problem of new items. To solve the above problems, it suggests the users’ interests integration based on user-item double clustering collaborative filtering algorithm(UIIDCCF).To solve the question of the decrease of real-time performance, it adopts the general architecture, which is based on double clustering of user-item and makes optimization of user clustering: it improves the algorithms of k-means partitional clustering which is based on user. Not only is the modified way of choosing the initial cluster centers, but also it improves the formula of distance resemblance. On one hand, generate of acnodes is reduced. On the other hand, the efficiency of cluster is promoted, which moves users who like items of similar types of item into a same class cluster.To solve the question of inaccurate calculation of users’ interests and bad effect of recommendation under the circumstances of sparse data, it ameliorates calculation of similarity, which is based on user clustering and items clustering from the perspective of types of items respectively. For the former, to get the favor similarity which is based on types of items between users, there are two views, which are the amount of score and the value of score. For the sake of new harmonic similarity, it joints above similarity and similarity of items being scored commonly between users. For the latter, it proposes a new method of data filled pretreatment and weight coefficient which is based on types of items that can find the nearest neignbors more precisely.To solve the question of the cold start of new items, it relies on the mechanism of recommendation, which is based on the architecture of type of items and double clustering to ease the cold start problem, which is caused by a little amount of score on new items through it can utilize various aspects of factors comprehensively.Finally, using the dataset ml-100 k, which is belong to classical MovieLens, it validates the users’ interests integration based on user-item double clustering collaborative filtering algorithm(UIIDCCF) is effective at six different aspects. The result of experiment shows that the new algorithm has the advantages of significantly higher inquiry efficiency of the nearest neighbors and the precision of recommendation to relieve the problem of data sparse and cold start of new items compared with the improved algorithm and classical algorithms which were used by the previous.