DEEP TEXT CLASSIFICATION AND RETRIEVAL FOR TEXT DOCUMENTS USING DEEP LEARNING TECHNIQES

Abstract

Document Retrieval (DR) is pivotal in unlocking valuable insights from the ever growing volume of medical literature. However, precise knowledge extraction from complicated clinical notes, discharge summaries, and research papers is still difficult. However, capturing these distinct dimensions of medical discourse and the complex interdependencies between entities with the currently used approaches becomes difficult. The classifications of Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), and Convolutional Recurrent Neural Networks (CRNNs) for use in document retrieval are thoroughly discussed in this review paper. The study offers a comprehensive assessment of the individual models' performance by summarizing the conclusions from over forty studies that relate to the individual models’ efficacy. It explores the potential outcomes of the group strategies in enhancing document retrieval accuracy. The yield of this research crystalizes the broad canvas encompassing the diverse features of CNNs, RNNs, and the ensemble technique CRNNs which are used to detect the complexities present in healthcare documents thereby presenting an analysis of their appropriateness included in various retrieval tasks and document types. The findings have imperative guidance for the researchers and practitioners looking to improve the retrieval system that documents have in healthcare to improve healthcare professionals' decision-making processes and promote access to vital medical information. This work solves these shortcomings by introducing a convolutional recurrent neural network (CRNN) framework that merges the advantages of CNNs and RNNs to achieve very high accuracy in medical document extraction tasks. In the CRNN model, the spatial feature extraction capacity of CNNs is integrated with the sequential learning capabilities of RNNs to successfully find medical entities and understand their complex interrelationships, which leads to high-performance document classification. As shown in the case of CORD-19, our CRNN model significantly outperforms individual CNN and RNN models on its entity recognition and relationship extraction tasks (98.93% accuracy), demonstrating that using CRNN in medical documents can yield much better results, thus increasing opportunities for informed clinical decisions, advanced drug discovery, and improved public health interventions.