カンファレンス (国際) Improving Taxonomies for Large-Scale Hierarchical Classifiers of Web Documents
Proceedings of the 19th ACM international conference on Information and knowledge management
We focused on taxonomy modification algorithms for grad- ually improving the relevance performances of large-scale hierarchical classifiers of web documents. Considering the research results of Tang et al. [5, 4], who took the same ap- proach, we investigated and implemented two heuristic tax- onomy modification algorithms for performing practical clas- sification processes for large-scale taxonomies. Although a taxonomy modification algorithm continuously improves the relevance performances of hierarchical classifiers, it increases the computational costs of those classifiers for training and predicting processes. We developed an improved taxonomy modification algorithm for reducing computational costs by preventing child node concentration. Although the relevance performances of the algorithm-modified taxonomy classifiers improved without increasing computational costs until the fourth generation by spreading the set of predicted classes, their relevance performances and behaviors went in opposite directions from the fifth generation.