COVID-19 Article Classification Using Word-Embedding and Extreme Learning Machine with Various Kernels

Abstract

The impact of the COVID-19 pandemic on the socially networked world cannot be understated. Entire industries need the latest information from across the globe at the earliest possible. The business world needs to cope with a very volatile market due to the pandemic. Businesses need to be swift in sensing potential profit opportunities and be updated on the changing consumer demands. Technological advances and medical procedures that successfully deal with COVID-19 can help save lives on the other side of the world. This seamless passage of crucial information, now more than ever, is only possible through the networked world. There are on average 821 articles published online on COVID-19 a day. Manually going through around 800 articles in a day is not feasible and highly time-consuming. This can prevent the industries and businesses from getting to the relevant information in time. We can optimize this task by applying machine learning techniques. In this work, six different word embedding techniques have been applied to the title and content of the articles to get an n-dimensional vector. These vectors are inputs for article classification models that employ Extreme Learning Machine (ELM) with linear, sigmoid, polynomial, and radial basis function kernels to train these models. We have also used feature selection techniques like the Analysis of Variance (ANOVA) test and Principal Component Analysis (PCA) to optimize the models. These models help to filter out relevant articles and speed up the process of getting crucial information to stay ahead of the competition and be the first to exploit new market opportunities. The experimental results highlight that the usage of word embedding techniques, feature selection techniques, and different ELM kernels help improve the accuracy of article classification.

Keywords