Answers are grounded in this paper and cite their sources.
Biomedical Named Entity Recognition (BNER) is the task of classifying biomedical instances such as genes, proteins, diseases, chemical compounds and others. Several approaches have been proposed for BNER specifically supervised machine learning techniques. Most of these techniques demonstrated reasonable performance. However, there is still a gap that lies on the multi-word BNEs such as the chemical compound ‘Tri-acetyl-glucal-galactono-lactone’ in which the classifier could not recognize these instances due to the complex characters used to separate the words. Complex characters could be punctuation (e.g. – *_^) or numeric characters. Therefore, this paper aims to propose a Back-propagation Neural Network (BPNN) for identifying BNEs. BPNN has the ability to encode the characters which facilitate the identification of multi-word BNEs. For this purpose, this study proposed multiple features for the encoding task including digits, special characters, affixes and capitalization. Experiments have been conducted using two benchmark datasets including SCAI and GENIA. SCAI is a corpus that contains chemical compounds, whereas GENIA is a corpus that contains multiple biomedical instances such as genes, proteins, DNA and RNA. Using 80% training and 20% testing, BPNN has shown 90% f-measure for the SCAI corpus and 82% f-measure for the GENIA corpus. Such results emphasize an enhancement of f-measure when compared to other related work. This implies that BPNN is effective in classifying BNEs.
No takes yet. Share an insight, caveat, or question.
Hashim et al. (2016) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: