|
|
|
| Package-based Hybrid Naive Bayesian Model |
| ZENG Xi, HAN Hua, LI Qiuhui, LI Qiaoli
|
| School of Science, Wuhan University of Technology, Wuhan 430070, China |
|
|
|
|
Abstract Hidden Naive Bayesian Model (HNB) and Tree Augmented Naive Bayesian Model (TAN) alleviate the strong independence assumption of Local Naive Bayesian Model (LNB) by mining the intrinsic associations between co-neighboring nodes, but ignore that there are both closely correlated nodes and relatively independent nodes in the real network. On this basis, a package criterion is designed, which divides the co-neighboring nodes into correlated co-neighboring nodes and independent co-neighboring nodes according to the degree of association. Then, packaging HNB and TAN respectively, so that the packaged-based hybrid naive Bayesian models are obtained. On FWFW networks with high average number of co-neighbors, the AUC values of the HNB and TAN models after packaging are increased by 12% and 11.6%, respectively. The experimental results show that the proposed method can effectively improve the link prediction performance and has good robustness.
|
|
Received: 21 February 2022
Published: 21 July 2023
|
|
|
|
|
|
|
|
|
