Multilingual Epidemiological Text Classification: A Comparative Study

Proceedings of the 28th International Conference on Computational Linguistics, pages 6172–6183 Barcelona, Spain (Online), December 8-13, 2020 6172 Multilingual Epidemiological Text Classification: A Comparative Study Stephen Mutuvi Multimedia University Kenya smutuvi@mmu.ac.ke Emanuela Boros University of La Rochelle France emanuela.boros@univ-lr.fr Antoine Doucet University of La Rochelle France antoine.doucet@univ-lr.fr Ga¨el Lejeune Sorbonne University France gael.lejeune@sorbonne-universite.fr Adam Jatowt Kyoto University Japan jatowt@gmail.com Moses Odeo Multimedia University Kenya modeo@mmu.ac.ke Abstract In this paper, we approach the multilingual text classification task in the context of the epidemi- ological field. Multilingual text classification models tend to perform differently across differ- ent languages (low- or high-resource), more particularly when the dataset is highly imbalanced, which is the case for epidemiological datasets. We conduct a comparative study of different machine and deep learning text classification models using a dataset comprising news articles related to epidemic outbreaks from six languages, four low-resourced and two high-resourced, in order to analyze the influence of the nature of the language, the structure of the document, and the size of the data. Our findings indicate that the performance of the models based on fine-tuned language models exceeds by more than 50% the chosen baseline models that include a specialized epidemiological news surveillance system and several machine learning models. Also, low-resource languages are highly influenced not only by the typology of the languages on which the models have been pre-trained or/and fine-tuned but also by their size. Furthermore, we discover that the beginning and the end of documents provide the most salient features for this task and, as expected, the performance of the models was proportionate to the training data size. 1 Introduction Monitoring and containment of infectious disease outbreaks have been an ongoing challenge globally. Whether previously with Ebola or today with the Covid-19 pandemic, surveillance has remained a key component of public health strategy to contain the diseases. The ability to detect disease outbreaks in an accurate and timely manner is critical in the deployment of efficient intervention measures. For instance, Ebola cases and outbreaks need to be immediately detected in order to be contained and stopped. A delayed response can have a significant economic and social impact, in addition to increased morbidity and mortality rates. Thus, the detection needs to be done as soon as the first reports appear, and, naturally, as such reports may not be in English, there is a need for effective multilingual surveillance systems. In recent years, there has been a rapid increase in data generated as a result of the progressive evolution of the Internet. The proliferation of digital data sources provide an avenue for data-driven surveillance, referred to as Epidemic Intelligence. Epidemic intelligence involves the collection, analysis, and dis- semination of key information related to disease outbreaks, with the objective of detecting outbreaks and providing early warning to public health stakeholders (World Health Organization, 2014). Natural Language Processing (NLP) techniques have made it possible to analyze data from web sources, such as social media, search queries, blogs, and online news articles for health-related incidents and/or …

Content truncated to 500 words