JMIR Publications

ABSTRACT

Background:

Twitter is a real time messaging platform widely used by people and organisations to share ‎information on many topics. It could potentially be useful to analyse tweets for infectious ‎disease monitoring purposes ‎ in order to reduce reporting lag time, and to provide an ‎independent complementary source of data, compared to traditional approaches. ‎However, such analysis is currently not possible in the Arabic speaking world due to lack of ‎basic building blocks for research.‎

Objective:

We collect around 4,000 Arabic tweets related to COVID-19 and Influenza. We clean and ‎label the tweets relative to the Arabic Infectious Diseases Ontology which includes non-‎standard terminology and 11 core concepts and 21 relations. The aim of this study is to ‎analyse Arabic tweets to estimate their usefulness for health surveillance, understand the ‎impact of the informal terms in the analysis, show the effect of the deep learning methods ‎in the classification process, and identify the locations where the infection is spreading.‎

Methods:

We apply multi-label classification techniques: Binary Relevance, Classifier Chains, Label ‎Powerset, Adapted Algorithm (MLKNN), NBSVM, BERT, and AraBERT to identify infected ‎people. We also use Named Entity Recognition to predict the locations affected. ‎

Results:

We achieve an F1-score up to 88% in the Influenza case study and 94% in the COVID-19 one. ‎ ‎ Adapting for non-standard terminology and informal language helps to improve ‎accuracy by as ‎much as 15% with an average improvement of 8%.‎ Deep learning methods ‎achieve around 5% on hamming loss during the classifying process. Our geo-location ‎detection algorithm can predict on average 54% accuracy for the location of the users using ‎tweet content.‎ ‎ ‎ ‎

Conclusions:

This study identifies two Arabic social media datasets for monitoring tweets related to ‎Influenza and COVID-19‎. It demonstrates the importance of including informal terms, which ‎is regularly used by social media users, in the analysis. It also proves that BERT achieves good ‎results when used with new terms in COVID-19 tweets. Finally, the tweet content may ‎contain useful information to determine the location of the disease spread.