Research On Top-n Recommendation Based On Implicit Feedback Of Linked Data

With the rapid development of the Internet in recent years, loads of information make a huge challenge on the user’s decision. Recommended system has become an effective tool to help people to find the items that they may be interested in in a complex information space. However, in most recommend system only have ratings from users to items, limitation of information led to a ineffective recommendation. The current mainstream recommendation systems are divided into two areas, One is based on the collaborative filtering method. Using a different method to construct new features of the users and items. The others are based on matrix decomposition technique, using the method of matrix factorization to solve the problem of data sparse in recommendation systems. By adding links data to combine linked data and entity data in a relational data model, we improve the accuracy of top-n recommend. The main content of this paper is divided into the following aspects:Firstly, we implemente the user-based collaborative filtering recommend algorithm by improving the user’s choice of their neighbors and regulating scores. Then we use an intuitive three-dimensional model to show the relationship between the items and linked data. We extracte items features from the three-dimensional model and calculate the recommend list based on the content recommenda algorithm. Finally, we mix the results of previous two algorithms based on a linear model.Secondly, we construct a graph on users and items. Then we use two different similarity calculation methods to acquire the recommend list. One approach is relied on the number of edges in the bipartite graph. Another is a method that using a one dimensional model to present users and items and calculate the similarity between them. Lastly, we take the properties of bipartite graph and the similarity we previous calculate to generate the top-n recommend list.Thirdly, we make a study of the properties of bipartite graph. Making a new tripartite graph by considering the implicit feedback information of linked data. A new features of users and items can be obtrained by extracting path-based features from new graph. Then transforming the task of top-n recommend into a binary classification by using a learning to rank method. Finally, experimental results show that our algorithms improve the accuracy of the top-n recommend task.