Abstract:
"Natural Language Processing has conducted a great deal of research and development on
Named Entity Recognition. The development of domain-general NER models has advanced
significantly, but there is still a lot of room for research in domain-specific NER models for
several reasons. The lack of annotated datasets for domains, which are difficult and time-
consuming to produce, is one of the main reasons of this. To address this problem, many
researchers are working on numerous projects to produce datasets that are domain-specific so
that NER systems can be constructed for that specific domain. Every time a dataset is
developed, tests should be conducted using all available approaches to determine which one is
best for that dataset. Because not all methods will perform equally well across all datasets
Computer Science is such a domain where NER is less focused because of the non-availability
of standardized datasets. As it has not been tested on more recent transformer-based models,
the primary goal of this research is to conduct an experiment using BERT and its appropriate
variants to a novel dataset that has been created to perform NER on Computer Science related
Research Papers.
Although it has been determined by this research that BERT and its variants do not provide a
good performance for this dataset, pathways have been found on how state-of-the-art methods
can be found for this particular dataset in the future using the results of the author's critical
literature review."
