Research On Personalized Recommender Algorithm Based On Hidden Markov Model

The rapid development of the Internet and information technology not only provides people with convenience,but also brings information overload and other problems.In order to help people quickly search commodities they prefer,recommender system came into being.The personalized recommendation algorithm is the core of the recommender system,and it determines the performance of the recommender system.At present,the research of personalized recommendation algorithm has made many achievements.While most personalized recommendation algorithms can be effectively applied to situation in which consumers' preferences do not change over time,the real-life is often opposite,and recommendation based on static preferences may not correspond to the consumers' current expectations.This thesis researches the personalized recommendation algorithm based on consumers' dynamic preference under the condition that consumers' potential preference can be mined.The main research results of this thesis are listed as follows:1.This thesis presents a new likelihood function that combins the collaborative filtering recommendation algorithm.This thesis reaserches the existing personalized recommendation algorithms,the collaborative filtering recommendation algorithm can mine the consumers' potential preferences,and the personalized recommendation algorithm based on the probability model can effectively capture the consumers' behavior.Therefore,this thesis combines the advantages of the two algorithms to propose a new likelihood function,which solves the problem that the existing probabilistic model can not mine consumers' potential preference.2.Based on the new likelihood function mentioned above,an improved expectation maximization algorithm is proposed.The expectation maximization algorithm provides an effective solution for the parameter estimation problem when there exists hidden variables.But the expectation maximization algorithm is used to solve the problem of the maximization of the general likelihood function.In order to solve the problem of parameter estimation of the new likelihood function mentioned above,an improved expectation maximization algorithm is proposed based on the derivation process of traditional expectation maximization algorithm.3.In order to adapt to the situation where consumers' preferences change over time,a personalized recommendation algorithm based on HMM(Hidden Markov Model)is proposed.HMM can analyze the invisible factors that affect the consumers' purchasing behavior and model the behavior sequence of the consumers.But the typical HMM cannot be directly applied to the recommendation problem.Therefore,this thesis has improved the traditional HMM.In addition,in order to make the algorithm can mine the consumers' potential preferences,this thesis proposes a personalized recommendation algorithm based on the HMM and a new likelihood function mentioned above.Comparing with the traditional personalized recommendation algorithm,the algorithm proposed in this thesis performs better.