A Combined Approach to Automatic Taxonomy Extraction
Samuel Pecar, Marián Šimko · 2019
We propose a method for taxonomic relationships extraction from text based on morpho-syntactic and pattern-based approach combined with utilization of distributional vectors and application of graph algorithms. We evaluated our method on the datasets from popular SemEval Workshop series. In addition to the standard evaluation measures, we explored also taxonomic measures, which can show other aspects of hierarchy quality. We also conducted an experiment for manual evaluation, which employed participants to evaluate outputs of our method. In most observed measures we obtained better results in comparison with methods competing at the workshops. We show that utilization of vector space and additional external semantic knowledge can significantly improve overall quality of extracted taxonomic hierarchy.