Link Prediction Based on Three-Type Heterogeneity Punishment

Link prediction based on node similarity has become a popular research topic in complex networks. Many existing models focus on endpoint influence on two-order paths, and rarely consider endpoint influence on one-order, two-order and three-order paths simultaneously. The one-order, two order and three-order paths between two endpoint pairs are called three-type paths. Endpoint influence is usually expressed by node degree. We consider the endpoint heterogeneity on the three-type of paths on different feature networks. The heterogeneity of endpoints plays a crucial role in attracting each other and forming stable links. Endpoints with greater heterogeneity have a higher likelihood of generating stable links. This paper proposes a link prediction model that enhances prediction performance by utilizing heterogenous path penalization. To compare our model with mainstream models, we will conduct experiments on 12 different data sets and analyze the results theoretically.