KEP-1.0: An Automatic Pipeline to Assist a Rapid Learning of COVID-19 Publications

Toward efficient learning of massive publications during the COVID-19 pandemic, we propose a pipeline, Knowledge Extraction for COVID-19 Publications (KEP), that aims at automatic extraction and representation of key knowledge from user-interested publications. The first version, KEP-1.0, has been developed and published on the Python Package Index (PyPI) (URL: https://pypi.org/project/KEP/). In this first release, knowledge about key topics, disease discussions, and location mentions for each publication is provided. KEP-1.0 not only extracts relevant knowledge but, more importantly, emphasizes the top discussed entities and presents visualizable plots, including bar graphs and word clouds. This allows a rapid preliminary understanding of the main discussions in the publication from these three aspects. Moreover, an enhanced TF-IDF algorithm, the weighted TF-IDF, targeting the publication topic identification purpose, has been proposed and evaluated. The pipeline is fully open-sourced and customizable. KEP-1.0 is ready for use in its current form or to be embedded into existing literature platforms. This pipeline is designed for COVID-related publications, but it has the potential to benefit similar knowledge extraction tasks for other topics of interest with a rapidly increasing number of publications.