Use of semantic expansion in vector representation of documents for hierarchical classification

This master’s thesis deals with the topic of text classification more precisely with hierarchical text classification. Readers of this thesis will be introduced to a few selected methods of text document representation and classification algorithms. It describes fundamental aspects of hierarchical classification and ways how to solve the problem of classification into a hierarchy of classes. This work presents a Python implementation of a hierarchical classifier that uses semantic expansion to extend the vector representation of a document. The goal of this thesis is to prove whether the use of semantic expansion leads to improved results of hierarchical expansion. The semantic expansion method in question lies in the aggregation of the vector representation of a classified document with the vector representations of similar documents. In this thesis, several aggregation functions and classifier configurations are tested. Experimental results show that the use of semantic expansion and certain aggregation functions leads to a statistically significant improvement in the classification results.