Research Of Deep Collaborative Algorithm For Community Mean Model

In the era of "information overload",personalized recommendation system has become a research hotspot.It can quickly and actively recommend according to user information requirements,omits the process of screening and filtering information,and prevent users from missing important information.At the same time,it has some problems,such as the problem of data sparsity and“cold start”.Therefore,how to effectively solve these problems,accurate recommendation,has become the main research content.There are many sources of information in the recommendation system:scoring data,item information and user information.In the Internet,scoring data and item information can be easily obtained as public resources.If you extract item features and start with item information,the "cold start" problem of the item in the recommendation system can be alleviated.On the contrary,due to privacy issues,user information that can extract user characteristics is difficult to collect directly on the Internet.Therefore,this paper obtains user information from a new angel,and a symmetric model is proposed,which can train both user information and project information to extract features.On this basis,the community factor is added.Specifically,the main work are as follows:(1)after reviewing and summarizing the research background and existingproblems of the recommendation system,through depth analysis of collaborative filtering algorithm and deep learning algorithm,and based on the research of community recommendation algorithm,the direction of research is clear.Firstly,the community contains some potential information,and if the recommendation is made in the community,the accuracy of recommendation will be improved.Secondly,combining the advantages of collaborative filtering algorithm and deep learning algorithm,a deep collaborative algorithm is proposed.(2)the IBCF algorithm of community mean model is proposed.Through depth analysis of collaborative filtering algorithm and deep learning algorithm,it is found that the community information is ignored.Aim at this problem,the user vector is represented by the mean model to realize the fast division of the community.Then,the item similarity between the community and the whole user set is calculated,and the ratio of the two similarity is adjusted by using the balance factor.Finally,forecast the user ratings and complete the user recommendation.The experiments on datasets Movie Lens 100 k and Jester show that the accuracy of the recommendation is further improved that the time is not greatly improved.(3)Aiming at the problem that collaborative deep learning only uses item information to extract features,a collaborative filtering algorithm based on symmetric SDAE is proposed.In addition to the item information,the user ratings can express the user preference intuitively.By analyzing the item information and the user ratings,the user information can be extracted.Thecollaborative filtering algorithm based on symmetric SDAE can fully mine user information on the basis of utilizing item information,so that the features obtained are more accurate.Experiments show that the recommendation accuracy of this algorithm is higher than that of CDL algorithm.(4)Symmetric SDAE collaborative filtering recommendation algorithm based on community is proposed.On the one hand,the community mean model can quickly divide the communities,and the training model can capture the community information in the community.On the other hand,the collaborative filtering algorithm based on symmetric SDAE can extract user features and item features from user information and item information at the same time,which can effective mitigation of "cold start" problem.Therefore,combined with the advantages of both,this algorithm is proposed.The experiment on the data set CiteULike shows that the accuracy of the recommendation algorithm is further improved.