TaxE: a Testbed for Hierarchical Document Classifiers

In the last decade the interest in the hierarchical organization of documents is increased. New challenges arise as hierarchical document classification, both unsupervised and supervised. A recognition of the most recent literature on these topics shows that none of the published works refer to the same dataset to enable the experimental phase. Moreover the papers don`t provide enough details to reproduce the same datasets starting from the same information sources. The drawback is twofold: from one hand the waste of time to preprocess suitable datasets, to the other hand the lack of a common testbed to compare alternative solutions. In this paper we propose a dataset extracted from Google and LookSmart web directories to support the experimentation effort in the field of hierarchical document classification. For such a task we aim to provide a kind of reference corpus in analogy with the role that Reuters plays in the scientific community. The paper illustrates the proces! s performed to generate a well defined dataset. This dataset is freely distributed over the we