A COVID-19 Search Engine (CO-SE) with Transformer-based architecture.

Area of Science:

• Information Retrieval
• Computational Biology
• Medical Informatics

Background:

• The exponential growth of COVID-19 research presents challenges in accessing timely and accurate information for healthcare professionals and scientists.
• Existing methods struggle to keep pace with the volume of SARS-CoV-2 publications, hindering knowledge dissemination.
• There is a critical need for specialized tools to navigate the complex landscape of COVID-19 scientific literature.

Purpose of the Study:

• To design and develop the COVID-19 Search Engine (CO-SE), an algorithmic system for efficient information retrieval.
• To enable users to find relevant documents and obtain answers to complex questions from a large corpus of COVID-19 publications.
• To provide practitioners, front-line workers, and researchers with expert-specific methods for staying current on SARS-CoV-2 research.

Main Methods:

• Developed a two-component system: a retriever and a reader.
• The retriever utilizes a TF-IDF vectorizer to identify relevant documents from the publication corpus.
• The reader employs a Transformer-based model to extract precise answers from retrieved documents based on user queries.

Main Results:

• The CO-SE system achieved an exact match ratio score of 71.45%.
• The system demonstrated a semantic answer similarity score of 78.55%.
• The proposed approach outperformed existing models and benchmark datasets, indicating strong generalizability.

Conclusions:

• The COVID-19 Search Engine (CO-SE) offers an effective solution for navigating the extensive body of COVID-19 research.
• The system's performance in retrieving relevant information and answering complex questions surpasses previous benchmarks.
• CO-SE enhances accessibility to critical scientific knowledge, supporting researchers and practitioners in the fight against COVID-19.