Blocking Distribution Based Hierarchical Reconstruction for Text Categorization

As one of the important techniques in large-scale data organizing, text categorization has been widely investigated. But the existing hierarchical classification methods often suffer from inter-level error transmission, namely blocking. In this paper, blocking distribution based topology reconstruction method was proposed for hierarchical text categorization problem. Firstly, blocking distribution recognition technique is put forward to mining out the serious high-level misclassification class. Subsequently, original hierarchical structure are reconstructed using blocking direction information obtained ahead, which increasing the path for the blocking instance to the correct subclass. Experimental studies on Chinese text classification benchmark Tan Corp, demonstrate that the proposed algorithm performs better than the traditional hierarchical and state-of-the-art flat classification strategies.