Research On The Graph-based Recommendation Algorithm Of Mobile Applications

With the current rapid development of the Internet, the number of APP is rapidly growing. Facing with massive APP, how to recommend the suitable APP for users becomes a major problem. Graph-based recommendation is one of the recommended technology for rapid development in recent years, which compared with the traditional recommendation technology(Collaborative filtering and Content-based recommendation).Graph-based recommendation can make full use of context information of items, which can improve the recommendation effect.Firstly, based on experimental data, we use a web crawler tool to gather the information of android market, and then preprocess the data to conform to the test requirements. Secondly, based on the graph-based recommendation technology on research at home and abroad, we propose a graph-based recommendation algorithm for APP, whose framework can be divided into four modules, constructed graph model module, node similarity calculation module, the objective function module and the recommended list module.In this paper, based on MMC, we propose a new learning sheme to automatically determine the influence weights between different types of entities, which need not to manually set the influence weights between different types of entities. To enhance the effect of recommendation, this paper will use the scoring matrix for the graph model.We design the following AUC(Area Under the ROC Curve) objective to model the objective function, based on whose convergence to analysis the influence weights between different types of entities. Then, the recommendation produces the corresponding APP list for every user by transfer parameters between different types of nodes.In practical experiments, we construct our algorithm with CF and CBF based on the precision and recall, which prove that our algorithm outperforms better.