Research On The Weight-Based Hybrid Recommendation Strategy

In the information highly developed today, how can people get their wanted information quickly and efficiently has become a big challenge in the field of Internet. The appearance of the recommendation system can effectively solve the problem of the current information overload.Hybrid recommendation is a kind of the recommendation system. At present the main problems of it is for their poor adaptive and personalized recommendation ability. The integration for hybrid recommendation can’t let the individual recommendation technology on certain individuals and certain items to stand out their recommendation advantage. And recommendation accuracy remains to be further improved.Based on the Mahout recommendation related API, and with the basis of the author’s design of weight-based hybrid recommendation strategy, the hybrid recommendation under the strategy was built to solve the hybrid recommendation main disadvantages and core problems, which highlight the personalized adaptive hybrid in recommending advantage. It can meet their adaptive selection of all kinds of recommendation technology requirements between different customers and objects, and improve the accuracy of the hybrid recommendation.This strategy main innovations include:first, using the idea of working backward error of prediction accuracy, and treatment of weight gain and normalized operation, adjust the impact strength of different recommendation technology in hybrid recommendation; Second, use selective weighted average method for hybrid recommendation, effectively avoid the general weight prediction results in the mixed recommended intermediate value problem; Third, in economics in the field of economic prediction mentioned combination predicting, this mix the weighting of the recommended strategy combined with the combined prediction results under this type of a kind of method, namely the weight advantage matrix method of mathematical statistics; Fourth, puts forward the predicting error results in a certain processing, with the corresponding prediction results under a single recommendation technology combined weighted end up with the final prediction results. Finally experiments show that the proposed hybrid recommendation strategy can improve the accuracy of the hybrid recommendation. Hope this thesis can work for further study on the hybrid recommendation system to provide more reference.