Analysis and Comparative Study of Recurrent Neural Networks for Improved and Accurate Classification of Medical Paper Abstracts

This paper investigates the utilization of various recurrent neural network (RNN) architectures, namely Simple RNN, LSTM, GRU, and Bidirectional RNN, for the purpose of classifying consecutive sentences in medical abstracts. The study assesses the efficacy of various models in categorizing phrases into predetermined classes: background, objective, method, result, and conclusion. The results emphasize the versatility of RNN architectures in capturing the sequential relationships within medical text, demonstrating their effectiveness in identifying various sentence functions. Notably, the best performance in terms of accuracy was achieved by the LSTM model, with an impressive 84.32%. This project aims to solve the demand for effective literature skimming, especially in information-rich sectors such as medicine, by implementing automated sentence classification inside abstracts. The motivation behind choosing this specific area of study stems from the critical need to streamline information retrieval in healthcare research, addressing the challenges posed by the vast amount of medical literature. The study highlights the necessity to utilize sophisticated algorithms to enhance text classification and facilitate the efficient skimming of literature.