A Study On SVD++-Based Of Collaborative Filtering Group Recommendation Algorithm

Now,we are in the era of big data,with the emergence of recommendation systems,it has been successfully applied to many areas of daily life.This not only alleviates the problem of information overload to some extent,it brings convenience to people's life,but also promotes the development of network economy and society.Recommendation system is a hot research topic in recent years,according to the different objects of recommendation service,it can be divided into:personalized recommendation and group recommendation.Group recommendation is to provide recommendation services to groups,which contains 2 or more than 2 users.It is developed on the basis of personalized recommendation,but which more complex than personalized recommendation,for more factors need to be considered in the process of processing.Group recommendation is of great commercial application prospects and great social value because of offering recommendation services for group's user in many area of daily life.This article revolves around the group recommendation related theory,the algorithm and the existence question launches the research,the content organization and the main research work are as follows:Firstly,in terms of theoretical research,due to the research on group recommendation is not hot to personalized recommendation research at home and abroad,there are relatively few literatures on group recommendation,especially the number of Chinese literature could be count.Therefore,through the research and analysis of the current status of group recommendation research at home and abroad,combing,refining and integrating related literature so that forms the theoretical Knowledge content of group recommendation.It mainly includes:the concept of group recommendation,the recommended generation process,the key technology of group recommendation,the group recommendation Method classification,the experiment data set and the evaluation index are used in the group recommended experiment,etc.I hope it is helpful to study and research group recommended for beginners.Secondly,in the field of algorithm research,the aggregation method used in most of the current group recommendation algorithms(score aggregation or recommendation results aggregation,and generally use only explicit scoring information and ignore implicit feedback information),it is easy to cause some members' preference information loss in group and data sparsity in group recommendation.This paper makes use of the recommendation principle and process similarity between group recommendation and personalized recommendation,drawing on the measures adopted in personalized recommendation and combining the characteristics of SVD++ model and the characteristics of group recommendation,so this paper proposes SVD++ based of collaborative filtering group recommendation algorithm.Meanwhile,presents a strategy of feature factor aggregation,which is different from the traditional group recommendation.In this paper,an item based collaborative filtering method is used to predict and fill a partial score missing item in the scoring matrix to alleviate the data sparsity problem in the group recommendation.Then,the stochastic gradient descent method is used to decompose the scoring matrix to get user characteristic factor vector,item characteristic factor vector and latent feedback factor vector.After that,the feature factor aggregation strategy is used to aggregate the user related feature vector in the group so as to get the characteristic factor vector of the group.Finally,the SVD++model is used to predict group scores,and achieve the recommendation by ranking the scores.This method could help to reduce the loss of member preference information in the group,so that the accuracy of recommendation is improved.Finally,in the algorithm simulation experiment and system design and realization aspect,the results of this method in different scale groups and the effect of alleviating data sparsity are tested by simulation experiments.Moreover,the collaborative filtering group recommendation method based on SVD++ is compared with the current group recommendation method which once obtained good results.The experimental results show that the proposed method is superior to the related group recommendation method in terms of recommended accuracy and recall rate.In addition,the paper further designs and implements the group recommendation system prototype,which based on SVD++ and collaborative filtering.