Research On Link Prediction Algorithms For Complex Network

The network is the form that runs through everything,and with the rapid development of network science,it has greatly facilitated human life.As one of the important research directions in the field of network science,the main task of link prediction is to predict the possible future connected edge relationships between nodes in a network based on the existing network topology relationships.The research of link prediction is not only at the theoretical level,but also has a wide range of practical applications in the fields of scientific cooperation networks,e-commerce networks and biological networks.How to improve the accuracy of link prediction algorithms is therefore the primary problem to be solved.Traditional similarity methods based on network topology are simple in principle,easy to implement,and fast in computation,but do not consider the impact of network data noise on its prediction accuracy,which is often inevitably accompanied by noise in the data of complex networks.Most of the machine learning based methods use Deep Walk graph embedding methods combined with simple binary classifiers,which cannot perform vector representation of network nodes as well as link prediction well.To address the above problems,this paper carries out research under the condition of vectorless powerless networks,and the specific research work and results obtained mainly include the following three parts:(1)In order to improve the prediction accuracy of traditional network topologybased similarity algorithms,a link prediction method based on k-shell decomposition with neighbor node degree denoising(KSDNN)is proposed.The KSDNN algorithm uses the k-shell decomposition with neighbor node degree index to make a comprehensive evaluation of the importance of network nodes at both global and local levels,and then performs link prediction after removing the network data judged as noisy by setting the noise conditions.The experimental results show that the prediction accuracy of the KSDNN algorithm is superior to the traditional similarity algorithm based on network topology among the six selected networks of different sizes.(2)The Node2vec-RF algorithm combines the Node2 vec algorithm,which has better network representation and classification,with the random forest(RF)algorithm.The Node2vec-RF algorithm combines the Node2 vec algorithm with the random forest(RF)algorithm,which learns the representation of the network nodes through a biased random walk,and then uses the RF algorithm to binary classify the possible connected edges and the impossible connected edges in the network.The experimental results show that the Node2vec-RF algorithm has better accuracy than the previous Deep Walk algorithm combined with a general binary classifier for the six selected networks of different sizes.(3)Based on the Node2vec-RF-based link prediction method,a friend recommendation system is designed and implemented to help users make more likeminded friends and help improve their overall ability.After inputting a user’s information in the system,the Node2vec-RF prediction model can be called to recommend 10 friends with similar information to the input user,and visualize them to the user.