A Hybrid Solution for Summarizing Diverse Medical Texts in the Health Domain

Since medical text summarization has become essential in facilitating rapid access to vital information for healthcare professionals, this research paper introduces a novel hybrid medical text summarizer that combines both extractive and abstractive summarization techniques while incorporating domain knowledge for the extractive process. Our approach first employs a domain-specific knowledge graph to guide the identification of salient and clinically relevant content from medical literature. This is followed by the application of advanced natural language processing techniques for abstractive summarization, ensuring the generated summaries maintain coherence and readability. We detail the use of the domain-specific knowledge base containing embedder, which captures medical concepts and relationships, enabling the system to discern important information from the input text. The cosine similarity-based ranking algorithm is adapted to prioritize sentences based on their relevance to the domain and their connectivity. The hybrid medical text summarizer is evaluated on a diverse set of medical articles at the same time to value summarizing multiple documents at the same time and compare its performance with existing approaches. Results demonstrate significant improvements in summary quality, as measured by ROUGE scores and human evaluations. Furthermore, we observe that incorporating domain knowledge in the extractive process enhances the overall effectiveness of the summarization system. We achieve a range-topping ROUGE score of a bit more than 0.6 for most of the texts summarized.