A Study Of Services Recommendation Models Of Mobile Internet Based On Data Mining

In recent years, mobile internet and e-commerce have been growing fast. Facing such enormous information of products and services, people are facing a serious problem of information overload. Under this circumstance, a good ecommerce recommendation system is particularly important. Recommendation systems’goal is to provide customers with what they want or need without requiring them to ask for it explicitly. To achieve this goal, it’s important to analyze and understand customer’s behaviors. By using recommendation systems, e-commerce websites can obviously improve the cross-selling ability and customer’s loyalty. In short, recommendation systems can not only help users overcome information overload problems but also find new business opportunities and bring potential customers to enterprise.Personalized recommendation has always being a focal point of researches among the internet and telecommunication fields. Reviewing most recommendation systems, there are two major branches: content-based recommendation and collaborative filtering. As single recommendation algorithm has limitations such as real-time problem, scalability, sparsity, cold start and so on, some researchers started to use hybrid recommendation. The hybrid recommendation approach is a combination of two or more of approaches to emphasize the strengths of these approaches and to achieve the peak performance of a recommendation system. Moreover, based on CF or CB, researchers also applied data mining and machine learning technologies such as decision tree, clustering, association rule, regression model in their systems to improve recommendation accuracy and efficiency.This study proposes a hybrid recommendation system framework. Firstly, we provide customer segmentations based on customer value using DFM model, a adaptive RFM model that fits telecom customers. Then, for the high value customers generated by DFM model, we apply association rules to provide recommendation; for low value customers, we compute the strength of social ties between customers to generate recommendation using their friends’preference information. For the experiments, we use china mobile customer data to perform the evaluation. The proposed hybrid methods are compared with the single association rules recommendation and the single social ties without customer segmentation. The results showed that the performance of the hybrid method exceed those of the single association method and social ties method.