NLP for knowledge discovery in scientific literature on biodegradation

Abstract

Text mining techniques provide automation of text analysis, overcoming the various limitations encountered by the volume of the data. Specifically Natural Language Processing seeks to understand and manipulate text to make computer systems comprehend natural languages to perform various tasks. As such, this research investigates the applicability of various text mining techniques in the biodegradation domain. The existing tools and methods do not focus on the domain providing an opportunity to apply o -the-shelve methods and examine their performance. In particular by applying Transfer Learning and fine-tuning the pre-trained model BioBERT with custom made data proved to outperform traditional methods. By conducting the tasks and reaching the set objectives, the research was able to demonstrate that transfer learning o ers the potential to mitigate the lack of training data on biodegradation information extraction. Finally, with the documentation of the Meta-Algorithmic Model (MAM) based on Cross Industry Process for Data Mining (CRISP-DM) provide a starting point for other researchers in customizing the process to their needs.