A comparison of collective classification techniques on network data
Ozge Ataseven, Yusuf Yaslan · 2017
Collective Classification techniques aim to improve the classification performance of linked data by utilizing unknown nodes in the network that are classified by using known nodes and network structure. In this paper, we consider both single and multi-labeled linked data classification problem using local and global classification algorithms. Initially, single-labeled linked data classification problem is evaluated using ICA-KNN, ICA-Naïve Bayes, LBP and MF algorithms on bibliographic datasets. Then we extend our experiments on terrorism relation multi-labeled linked dataset by using ML-LBP, ML-MF global classification algorithms. The experimental results show that for single-labeled linked data the best classification accuracy is obtained by MF global classification algorithm. For multi-labeled data both ML-LBP and ML-MF algorithms perform similarly.