Research On One-class Collaborative Filtering In Scientific Social Networks

The rapid development of information technology,especially the introduction of Web2.0 technology,enables users to change from simple information recipients to information producers and publishers.Various social platforms continue to emerge,breaking the communication barriers of time and space between users.With the vigorous development of social networks,there has also been a social network specializing in serving researchers in the field of scientific research,namely scientific social network.As a social platform that takes into account both the social convenience and the special needs of researchers,the scientific social network has attracted a large number of researchers in recent years.However,the large number of user-generated content generated by the burgeoning users also poses a serious problem of information overload.Faced with a large number of papers,groups,etc.,it is difficult for users to choose the most interesting information.As an efficient information filtering method,the recommendation system is gradually becoming a common hot spot in academia and industry.When researching how to apply the recommendation system to scientific social networks,the data available contains only a small number of positive examples that clearly reflect user preferences,leaving unlabeled data in which a large number of negative examples are mixed with potential positive examples.Therefore,it natural belongs to One-Class Collaborating Filtering(OCCF).However,as far as we know,there have been few scholars studying the recommendation issues in scientific social networks from the perspective of OCCF,not to mention basing on the widely existing but not yet fully used social information to solve the low recommended accuracy caused by the extreme imbalance and sparsity problems of data.To this end,this thesis constructed an improved one-class collaborative filtering approach that considered the scientific social network characteristics and social information.Based on the in-depth analysis of the basic theory and the current research status both at home and abroad related to scientific social network as well as OCCF,this thesis from the perspective of OCCF,respectively from two aspects of individual recommendation for papers and groups,built a hybrid approach of Social and Content aware One-class Recommendation of Papers(SCORP),and a Group characteristics and Social information based One-class Recommendation of Groups in scientific social networks(GSORG).Finally,the experiments were conducted on a real scientific social network dataset.The results show that the new methods proposed in this thesis has achieved better experimental results which proved the effectiveness of the two improved one-class collaborative filtering methods.Through this thesis,on the one hand,the relevant theories and research status of scientific social networks and one-class collaborative filtering methods were analyzed in detail.And on the basis of existing researches,this thesis took the lead from the perspective of OCCF to study the recommendation issues in scientific social networks,enriched and expanded the theoretical research system of recommendation system for scientific social networks.On the other hand,in view of the extreme data imbalance and data sparsity problems inherent in OCCF,this thesis has considered the scientific social networking characteristics and social information at the same time,which provided a new idea and approach for integrating other effective additional information in the one-class collaborative filtering method.