A semantic Similarity-Based approach to extract respiratory disease-symptom relations from biomedical literature

In the biomedical domain, the surge in article volume means valuable insights on diseases and symptoms are oftenhidden in academic literature. Leveraging natural language processing and text mining to sift through biomedicaltexts is vital for advancing early diagnosis, enhancing clinical decision support systems, and refining ontologies.Particularly for respiratory diseases, which share symptoms like fever, cough, and breathlessness, differentiatingbetween diseases based on symptoms is crucial for early and accurate diagnosis. This study introduces a methodfor extracting disease-symptom relationships, aiming to identify rare symptoms not mentioned in health resources but potentially related to diseases, and to ascertain the association strength between diseases and symptoms.Initially, a hybrid entity recognition approach was proposed for identifying diseases and symptoms in medicaltexts. Then, the diseases and symptoms were normalized, and their associations ranked by semantic similarityscores. Evaluated on a dataset of respiratory diseases, including academic article abstracts on asthma, bronchitis,pulmonary embolism, and COVID-19, the study uncovered rare symptoms in addition to characteristic ones. Thedot product similarity method proved more effective, achieving an average similarity score of 0.66, in establishingthe associations between diseases and symptoms, revealing the significance of literature validation in identifying rare symptom-disease relations