| With the rapid improvement of science and technology,academic information is growing rapidly,and quite a number of academic websites are emerging.This kind of academic website mainly focuses on the research field of scholars and the cooperative relationship between scholars,laying the foundation for personnel cooperation and scientific research cooperation.As the content of academic websites is mainly academic papers,how to extract attributes from papers and make reasoning analysis is the key problem to be solved.According to the characteristics of heterogeneous academic networks,we build a cooperative relationship prediction model based on machine learning.The whole model includes the following parts.Firstly,according to the rich semantic information characteristics of heterogeneous networks,we design a structural feature extraction module based on meta path,and use the similarity measurement function to calculate the structural similarity between nodes.Secondly,the existing methods take the similarity of network structure as the main starting point and ignore the attribute information of nodes.To solve this problem,we design three kinds of attribute features,including digital features,key information text features and paper information abstract text features.Then we use attribute information in academic network to calculate attribute similarity between nodes.Finally,we use a machine learning classifier to combine the structure similarity with attribute similarity to learn each feature and complete the cooperative relationship prediction task.The experimental results show that the AUC(Area Under the Curve)value of this method is increased by about 3-10% compared with the benchmark method in the AMiner academic data set.The model has higher prediction accuracy.This method can interpret the semantic information of the cooperative relationship from different levels by using three meta path features,and has better interpretability. |
PDF Full Text Request