SciBERT Optimisation for Named Entity Recognition on NCBI Disease Corpus with Hyperparameter Tuning

Named Entity Recognition (NER) in the biomedical domain faces complex challenges due to the variety of medical terms and their context of use. Transformer-based models, such as SciBERT, have proven to be effective in natural language processing (NLP) tasks in scientific domains. However, the performance of these models is highly dependent on proper hyperparameter selection. Therefore, the aim of this study is to analyse the impact of hyperparameter tuning on the performance of SciBERT in NER tasks on the NCBI Disease Corpus dataset. The methods used in this study include training the baseline SciBERT model without tuning, followed by hyperparameter optimisation using grid search, random search, and bayesian optimisation methods. Model evaluation is done with precision, recall, and F1-score metrics. The experimental results showed that of the three methods grid search and random search produced the best performance with a precision, recall and F1-score of 0.82, improving from the baseline which only achieved a precision and recall of 0.72 and F1-score of 0.68. This study confirms that proper hyperparameter tuning can improve model accuracy and efficiency in medical entity extraction tasks. These results contribute to the development of optimisation methods in biomedical text processing, particularly in improving the effectiveness of the SciBERT Transformer model for NER.