Link prediction of directed knowledge network based on patent co-citation

At present, the link prediction for knowledge network is mainly based on literature data, and rarely considers patent data, leading to the omission of key information and other problems. Therefore, this paper explores the method of link prediction for knowledge network based on patent co-citation. By comparing the collection of e-commerce patent data from different patent databases, 50 groups of co-citation data were selected to establish a co-citation network and its eigenvalues were visualized. Four machine learning methods, random forest, Knn, XGBoost and support vector machine, were used to predict the link of the network. AUC means are used to compare their performance on Adamic/Adar(AA) values and Common Neighbours(CN) values. The results show that XGBoost algorithm has the best performance, and the average AUC of AA and CN values are 0.804 and 0.791, respectively. Random forest is second; KNN algorithm and support vector machine (SVM) algorithm are unstable on this data set.