Self-revise Hierarchical Classifier Based on SMO Algorithm for Chinese Question Classification

For hierarchical classifier, a mainstream approach for question classification, a fatal pitfall exists that the accuracy of fine categories relies on that of coarse categories and stays on a limited level accordingly. This paper proposes a novel self-revise hierarchical classifier based on sequential minimal optimization (SMO) algorithm. The information of scores calculated by SMO is utilized to make a classifier of question category decision, which helps to correct the misjudged coarse label for each question and eventually revise the fine label. The precision for coarse and fine classes on HIT-IR Chinese question dataset are 88.54% and 72.46% respectively, which reaches the peak in the current research.