Named entity recognition on bio-medical literature documents using hybrid based approach

Area of Science:

• Medical Informatics
• Natural Language Processing
• Computational Linguistics

Background:

• Technological advancements offer opportunities to extract insights from unstructured medical data.
• Medical literature contains vast amounts of valuable knowledge.
• Existing methods for information retrieval from medical texts face challenges.

Purpose of the Study:

• To propose a novel hybrid approach for named entity recognition (NER) in medical literature.
• To improve the identification of specific medical entities such as drugs, diseases, symptoms, route of administration, species, and dosage forms.
• To develop a more accurate and efficient information extraction tool for medical research.

Main Methods:

• Development of a new hybrid approach combining dictionary-based annotation and machine learning.
• Creation of a new dictionary for annotating route of administration, dosage forms, and symptoms.
• Training a Spacy machine learning model using annotated medical entities.
• Validation of the hybrid model using a dictionary and human review to calculate a confusion matrix.

Main Results:

• The proposed hybrid approach demonstrates improved entity identification compared to existing models.
• The model achieved an average F1 score of 73.79% for five key entities.
• The hybrid model shows decent accuracy in identifying named entities from medical literature.

Conclusions:

• The novel hybrid approach effectively extracts valuable information from unstructured medical literature.
• This method offers a significant improvement in identifying medical entities, aiding researchers.
• The enhanced NLP technique contributes to more efficient knowledge discovery in the medical domain.