A sui generis QA approach using RoBERTa for adverse drug event identification

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Pharmacovigilance

Background:

• Monitoring drug safety through adverse drug event (ADE) extraction from biomedical literature is crucial.
• Existing entity-relation extraction methods, often using Bi-LSTM, lack optimal feature representation.

Purpose of the Study:

• To develop an improved framework for extracting ADEs from biomedical text.
• To overcome limitations of existing methods by leveraging advanced NLP techniques.

Main Methods:

• A question answering (QA) framework utilizing RoBERTa (Robustly Optimized BERT Pretraining Approach).
• Domain adaptation techniques applied to RoBERTa.
• An end-to-end pipeline integrating QA and transformer architecture.

Main Results:

• The proposed QA framework with RoBERTa achieved a 9.53% higher F1-Score compared to prior work.
• Demonstrated superior performance in entity-relation extraction for ADEs.

Conclusions:

• An end-to-end pipeline combining transformer architecture and QA significantly enhances biomedical entity-relation extraction.
• This approach is valuable for identifying potential adverse drug reactions in various therapeutic contexts.