Deep Learning for Structuring Epidemiological Reports and Temporal Event Extraction: A TBNN Model Study (Preprint)

BACKGROUND Background: The ongoing global health emergency has brought to the fore the need for quick and precise analysis of epidemiological investigations, as traditional methods have proven to be inadequate and limited in their ability to extract crucial information. To address this challenge, this study introduces an innovative deep learning approach: a Time-Based Neural Network (TBNN), which aims to efficiently extract and organize important data from epidemiological investigation reports. OBJECTIVE Objectives: To streamline and enhance the process of extracting essential information from epidemiological investigation reports, this study focuses on patient event data. The proposed TBNN model integrates temporal-aware deep learning algorithms to improve the efficiency of information processing. The anticipated outcome is a significant enhancement in the efficacy of epidemic response mechanisms. METHODS Methods: The Time-Based Neural Network (TBNN) method employs a synergistic approach by integrating a pre-trained Bidirectional Encoder Representations from Transformers (BERT) language model with a Bi-directional Long Short-Term Memory (Bi-LSTM) network for information extraction. This technique involves the generation of time-enhanced textual representations, targeted extraction of information, and the assimilation of temporal dynamics, offering a holistic strategy for processing unstructured data in epidemiological investigation reports. To ensure robustness, the model training and validation are conducted using the ECR-COVID-19 dataset. RESULTS Results: Experimental results demonstrate that TBNN outperforms existing models in event extraction, specifically in terms of precision, recall, and F1 scores. The model's operational efficiency is highlighted by its capability to process reports efficiently on a commonly available 4-core Intel i5-class CPU, underscoring its practical feasibility for deployment in real-world epidemiological investigations. CONCLUSIONS Conclusions: The TBNN model offers a promising solution for refining data extraction and structuring within epidemiological investigation reports, providing a more efficient and accurate method for information extraction. The performance and efficiency of the model underscore its potential use in improving the overall capacity of epidemic responses during large-scale outbreaks.