MedTSS: transforming abstractive summarization of scientific articles with linguistic analysis and concept reinforcement [1]
MedTSS employs a sequence of components to process a journal article, with each component gradually contributing to information extraction for the summarizer to use. The MedTSS accepts a plain scientific journal article with its keywords and MeSH terms in either TXT or XML document. The current open-source release consists of the following steps that process the source text for effective information extraction.
• Pre-processor and sentences selector
• Multiple concepts reinforcer
• Preface entity hallucination detector and handler
• Sentence Ordering
• Lexical simplification of clinical and medical terms
We provide MedTSS working on five samples of HTSS testing dataset. HTSS reported, there top-five generated summaries, that we used to compared with MedTSS(PTMs) generated summaries. Code avaliable in 'MedTSS-Summarization.ipynb' file.
MedTSS ordered there generated summaries of k-conceptual groups based on cosine similarity computed between the sentences od the original text and generated summaries. Code avaliable in 'MedTSS-OrderedSummary.ipynb' file.
MedTSS introduced a new matric to compared the named entities similarity between the ground-truth summaries and the generated summaries using ESM. Code is avaliable in 'MedTSS-ESM.ipynb' file.
MedTSS text simplification task performed using MedTCS module. Code in 'MedTSS_Simplification.ipynb'.
[1] Saeed, Nadia, and Hammad Naveed. "MedTSS: transforming abstractive summarization of scientific articles with linguistic analysis and concept reinforcement." Knowledge and Information Systems (2024): 1-18.