Link Prediction in Relational Data

摘要: Many real-world domains are relational in nature, consisting of a set objects related to each other complex ways. This paper focuses on predicting the existence and type links between entities such domains. We apply Markov network framework Taskar et al. define joint probabilistic model over entire link graph — entity attributes links. The application RMN algorithm this task requires definition patterns subgraph structures. method two new datasets, one involving university webpages, social network. show that collective classification approach RMNs, introduction labels, provide significant improvements accuracy flat classification, which attempts predict isolation.