Research On Recommendation Algorithm Based On Hidden Markov Model

With the rapid development of mobile Internet and the explosive growth of electronic commerce data,people faced with more and more serious "information overload" problem."Information overload" means that people can not locate the information they need from the mass of data timely,thus causing huge interference to the user's Internet behavior decisions.Recommender system is the key technology to solve this problem.Recommender system is a mining technology based on the users' historical buying and rating behavior.With the purpose to make users' visits easier,it can forecast users' preferences,and give meaningful suggestions to user for buying decisions.The main concern in this system is how to combine users'historical ratings' information with recommendation technology.According to the journals in China and other countries about personalized recommendation system,the common recommendation algorithm based on users' rating towards events has many limits,which lowers the accuracy of recommendation when users'preferences change.Therefore,for individual's rating,we develop the rates' behavior portrait by adopting association rule in data mining to gain the users groups' preference's polymerization behavior.The portrait can be used as the context of users' rating,which can decrease the system's sensitiveness towards preference changes.Based on this idea,to get a better results of user's preference' forecast,Presented Hidden Markov Model(HMM)to simulate the discrete ratings' behavior portrait in this paper.With users' rating sequence as observation symbol,user groups' triadic polymerized rating portrait as implicit state to set up HMM model,Gaussian Mixture Model(GMM)to express the emission of HMM.Since user's rates portrait and user observation symbols can be correlated.By forecasting polymerized rating portrait,we can decode user's next event with the highest possibility,so that recommendations of goods or information can be made.This model largely increases the accuracy of recommendation system,and it is strong in robustness because it can develop related recommendations when users' preferences change.The experiment shows that the recommendation algorithm based on HMM is strong in accuracy,recall ratio,and recommendation effectiveness.