Research Of Recommendation Algorithm Based On Grey System Theory

With the rapid development of the Internet and the improvement of modern technology,the producer of Internet content such as people or businesses,creates a large amount of information every day,which directly leads to information overloading.From the perspective of information consumers,users need accurate access to the information from a large amount of data.Therefore,the rapid growth of data and consumer demand co-spawned continuous research and development of recommended system theory in the field of data mining.Based on the user historical behavior and social relations,this dissertation utilizes the theory of uncertainty and interdisciplinary view to study the new recommendation algorithm.The thesis studies the user behavioral interests and social relationships with some models and algorithms.The main contents of this thesis are as follows:1.From the perspective of user historical behavior research,focusing on data non-uniformity and sparseness existing in recommendation schemes,this study introduces and optimizes grey system theory model in application scenarios of social network.Furthermore,a new topN recommendation algorithm is proposed.Firstly,by analyzing the rating behavior of users,factors that affect rating are discovered.To quantify the factors,time discretization method is leveraged.Secondly,in regard to the problem of time non-uniformity and data sparseness in observed sequences,the GM(1,N)of grey system is introduced to construct rating prediction model,which can mine the dominant relation between user rating and related factors.Finally,considering the timeliness of user preference,a time decay function is introduced to optimize the grey forecast model and reduce its error rate in prediction.The experimental results reveal that this improved grey forecast model can not only effectively predict the user rating behavior but also mine the explicit relation between user rating and related factors.2.From the perspective of social relationship research,aiming at the current research of poor ability to compute relevance between users,which takes user similar behavior into account while ignoring the long-term interests,this thesis introduces and improves the grey proximity correlation analysis method and designs a hidden friend recommendation algorithm based on similar behaviors and long-term interests.Firstly,the user historical behavior is analyzed and the relevant attributes including the common behavior are extracted.Secondly,extracting the label information that reflects the preferences of users,and calculating the similarity of user tags to quantify long-term interests.Finally,using the improved method of grey proximity correlation analysis,this thesis combines the grey relation degree of rating vector with similar behavior and long-term interest to mine the hidden relationship and realize friend recommendation.Experiments show that this algorithm can accurately express the relevance between users and has a great improvement in the accuracy of recommendation.