Research Of Link Prediction Based On Generalized H-index

There are many kinds of complex systems in the world.In order to have a more clear and comprehensive understanding of these complex systems,we abstract the research objects into nodes and the complex relationship between them into edges,and then recombine them thus form various networks.By studying these abstract networks,we can find the underlying law between those research objects.Such works will bring great benefits to people's actual life.The field of link prediction is one of the results of long-term research on abstract networks,which integrated lots of methods and technologies in similarity analysis,network dynamics,Bayesian model,machine learning,etc.This field is devoted to solving the problem of how to predict the possibility of connection between two nodes in the network that have not yet have edge.The research results of link prediction have been applied to many actual problems,such as social network analysis,biological network analysis,network reconstruction,personalized recommendation,etc.Link prediction problem has great significance in both practical application and theoretical research.Therefore,improving the accuracy of link prediction is bound to have a huge promotion for these two aspects.The research results in link prediction field have been very fruitful,there are many link prediction models with good prediction accuracy and performance evaluation indexes with excellent validation.However,on the one hand,with the continuous development of network science theory,lots of new and more effective measurement indicators are constantly proposed.Some indicators used in the link prediction algorithm model cannot grasp the essential characteristics of the network well.To improve the accuracy of the link prediction algorithm is our constant pursuit,which requires to making further efforts to improve existing link prediction algorithm models.On the other hand,the applications of using the new link prediction algorithm model to make relevant recommendations is still very few,and the application scenarios based on the latest theories in the field of link prediction are still not enough.In view of this,the research results of this thesis mainly include the following aspects:(1)Studying the traditional link prediction algorithm based on similarity analysis in detail.On this basis,combined with the h-index generalization theory of nodes,proposing that the intermediate process of h-index generalization of nodes is also an important attribute of nodes,and then use the intermediate process of h-index generalization of nodes to improve the link prediction algorithm based on random walk similarity analysis,thus forming the new random walk similarity index.Meanwhile,testing the link prediction algorithm based on the new random walk similarity index on six different types of real networks,so as to prove the validity and accuracy of the generalization intermediate process of h-index as an important attribute of nodes,and further improves the accuracy of link prediction algorithm.(2)Using the popular open-source asynchronous web application framework Django based on Python and the classic relational database management system mysql,and based on the new link prediction algorithm model of random walk similarity index,proposing a new collaborative filtering recommendation algorithm which combining bipartite graph projection and sorting,and take it as the core of the recommendation algorithm to develop a personalized movie recommendation system,thus make the link prediction theory become a reality.This system is the concrete implementation and application of the research results of this paper.In order to ensure the accuracy and rationality of the recommendation results,the new collaborative filtering algorithm based on the generalized h-index is compared with the classical collaborative filtering recommendation algorithm on this thesis,which shows that the new collaborative filtering recommendation algorithm can greatly improve the accuracy of the recommendation.