Learning and Evaluation in the Presence of Class Hierarchies: Application to Text Categorization

摘要: This paper deals with categorization tasks where categories are partially ordered to form a hierarchy. First, it introduces the notion of consistent classification which takes into account semantics class Then, presents novel global hierarchical approach that produces classification. algorithm AdaBoost as underlying learning procedure significantly outperforms corresponding “flat” approach, i.e. does not take information. In addition, proposed surpasses local top-down on many synthetic and real tasks. For evaluation purposes, we use measure has some attractive properties: is simple, requires no parameter tuning, gives credit correct discriminates errors by both distance depth in