Improved Slope One Algorithm Based On Item Category And K-means Cluster

Collaborative filtering recommendation algorithm is one of the most widely used and mature algorithms in real world.The Slope One algorithm is a item-based collaborative filtering recommendation algorithm,which is highly praised by many scholars because of its simple and efficient characteristics.But the algorithm considers all users and items to be equally important,and the algorithm's prediction accuracy and scalability will face serious challenges when the dataset is too sparse.Aiming at these problems,this paper puts forward the corresponding improvement measures.Firstly,the article has introducing the background and current status of the recommendation system and describing the relevant knowledge of the collaborative filtering recommendation algorithm.Then the paper focuses on the content of Slope One algorithm,and summarizes the advantages and disadvantages of the algorithm,and proposes the corresponding improvement measures.Secondly,as the number of users and items grows dramatically,the amount of computation also increases,which can lead to poor scalability of the algorithm.Therefore,this paper considers using the item category information to divide the rating matrix,and then the correlation calculation of the Slope One algorithm is carried out in the category matrix where the target item is located.To distinguish users,this paper introduces category experts,that is,average deviations between pairs of relevant items are generated on the basis of ratings from category experts;To distinguish items,this paper introduces the dynamic k-nearest-neighborhood about items,that is,average deviations and prediction ratings are generated on the basis of ratings from k-nearest-neighborhood items.This improvement measure not only reduces the amount of computation,but also filters out the impact of unrelated users or items on the forecast results and improves the prediction accuracy.Then,when datasets don't contain items' attribute information,the K-means clustering method is used to cluster users and items respectively by the rating information.For the method of user-based clustering,it needs to find the k-nearest-neighborhood of the user in the cluster where the target user is located,and then calculates the average deviations according to the ratings of the neighbor users.For the method of item-based clustering,it needs to find the k-nearest-neighborhood of the item in the cluster where the target item is located,and then calculates the average deviation and the forecast rating based on the ratings from the neighbor items.This scheme also can reduce the computational complexity and improve the prediction accuracy.Finally,the experimental results show that the proposed algorithm is better than the original algorithm in predictive accuracy.