Developing a Model for the Automated Identification and Extraction of Agricultural Terms from Unstructured Text
Hercules Panoutsopoulos, Christopher Brewster, Borja Espejo-García
Abstract
Text is the prevalent medium for conveying research findings and developments within and beyond the domain of agriculture. Mining information from text is important for the (research) community to keep track of the most recent developments and identify solutions to major agriculturerelated challenges. The task of Named Entity Recognition (NER) can be a first step in such a context. The work presented in this paper relates to a custom NER model for the automated identification and extraction of agricultural terms from text, built on Python's spaCy library. The model has been trained on a manually annotated text corpus taken from the AGRIS database, and its performance depending on different model configurations is presented. We note that due to the domain ambiguity, inter-annotator agreement and model performance can be improved.