Segmentation of Natural Language Documents using Term Distance as Discourse Coherency Measure

摘要

Machine learning (ML) methods are used to extract new knowledge from existing datasets. Ensemble methods (EM) were introduced to improve the ML performance. While EM's offer performance improvements, they limit the amount of control and general understanding of how the final result was achieved. We are researching the possibility to verify ML results with formalization of research results accessible in natural language. We propose to use the extracted information from scientific papers for the verification of ML results. In order to be able to extract the information, relevant to the ML results, an approach to extract specific information rich segments from scientific papers is needed. We present an approach to automated selection of relevant segments from large repositories.