Research And Implementation Of Hybrid Recommendation Algorithm In Heterogeneous Network

The growing data information in Internet has caused "information overload".So the efficient and accurate personalized recommendation in big data has become one of the hottest research.At the same time,due to the heterogeneous information network has complex object types and rich semantic information,it more likely to dig up the information which is hidden across the entire network.Therefore,in order to improve the performance of personalized recommendation,this paper focuses on the research and implementation of recommendation algorithm which is based on heterogeneous network.In algorithm,this paper proposes a hybrid recommendation algorithm based on heterogeneous networks.Due to the users with similar interests have similar user attributes,this paper uses the Canopy-k-means clustering to divide users,which can reduce the complexity of user similarity computation,improve the real-time performance and ensure the accuracy.After that,this paper focuses on the influence of time on user similarity,and puts forward a heterogeneous network similarity algorithm(Hete-DS),which is combining HeteRecom algorithm and the time factor.Hete-DS is not only considering time effect on the similarity,and improves the accuracy.At the same time,this paper introduces the expert recommendation algorithm,which is based on highly influential users in cluster.It can effectively solve the "cold start" problem.According to the above algorithms,they are used to estimate the score of projects.By analyzing the results of several sets of contrast experiments,the proposed algorithm has better accuracy.Facing the vast data,Hadoop not only makes the big data processing capabilities more cheaply,but also meets the requirement of the processing efficiency.According to the above algorithms,they have been improved to adapt to Map-Reduce parallel processing,which has good speed ratio and good extensibility.