Automatic summarization of academic studies published in the health field using natural language processing and deep learning methods

For researchers working in the health field, it is crucial to read and analyze scientific papers quickly. In this regard, readers generally prefer to read summary sections rather than the entire text. Automatic text summarization systems provide researchers with the opportunity to access information in a shorter time. For researchers contributing to this field, the concise summaries that select the most important parts of the topic and present extensive publications in a brief format are of vital importance as they allow accessing a greater amount of information in a shorter period. This thesis focuses on the automatic summarization of academic papers in the health field. Scientific papers already contain manually crafted abstract sections. The aim of this study is to generate expanded summaries using algorithms employed in automatic summarization systems and to compare existing algorithms using a dataset. As there was no available dataset on this subject, an original dataset was created. The study involved the application of seven different algorithms, including extractive summarization methods such as Term Frequency, Term Frequency-Inverse Document Frequency, TextRank, Latent Semantic Analysis, and Latent Dirichlet Allocation, and abstractive summarization methods like the T5 algorithm and Generative Pre-trained Transformer. The generated summaries demonstrated a similarity of 76% to the original summaries, indicating promising results close to human-generated summarizations. It is believed that this study will contribute to Turkish works in the field of natural language processing in terms of dataset and application.