Integrated Framework for Improving Large-Scale Hierarchical Classification

摘要: Hierarchical classification (HC) approaches leverage the hierarchical structure during training and testing of models which helps in improvement performance prediction efficiency. However, their comparison to flat may be poor if hierarchy used for contains: (i) Inconsistent relationships such as parent-child or siblings (ii) Less cohesive overlapping classes. Moreover, dynamic changes data characteristics over time requires new classes (orphan nodes) identified improve generalization learned models. In addition, we also need deal with imbalance large-scale problem where large number have a few examples training, posing statistical challenges. this paper, propose an integrated framework address aforementioned issues improving HC. Our experimental evaluations on various image text datasets shows improved our proposed framework.