Recommendation Algorithm Research And Application

The Internet is playing a key role in every corner of our life,with the development of the Internet information technology has changed rapidly in recent years.The large amount of information is a double-edged sword.which brings the large scale information on the one hand,on the other hand much of them are useless information.Recommendation system research has become a hot topic.And personalized recommendation is one of the areas analyzed by the links between user and item.Because of the superiority of personalized recommendation,it has been widely used in e-commerce,social networking sites,news sites,and many other fields.Collaborative filtering is the most widely used algorithm in recommendation system.For sparse matrix problems,matrix factorization technique has become a new choice for many researchers.Major contributions of this thesis include these works.First,this thesis introduces the reasearch background and the principle of matrix factorization algorithm.Second,this thesis has improved the timeSVD++model which can refer to Yehuda Koren,and the new model is named as timeSVDi++ model.On the one hand the new model timeSVDi++ has incorporates implicit feedback factor,on the other hand it has manifested the time effects in data sets.It has been considered from three aspects,namely user bias,item bias and user features bias.We not only considered the cycle characteristics of the dataset,but also considered the emotional changes of users.Emotion changes have two aspects.One is transient changes which are irregular,and the other is regular part which can be called periodic social emotions.The third part discusses the implementation of matrix factorization algorithm in single-threaded version.And with the amount of data grows,we discuss the steps to implement it under a distributed framework.Fourth,experiment on Netflix Dataset.Based on matrix factorization algorithm,this thesis proposes the timeSVDi++ model,which incorporates implicit feedback factors and manifested the time effects in data sets.And implement this new model under a distributed framework.It has comparatively higher theoretical and practical value.