An ontology-driven system for detecting global health events

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 215–222, Beijing, August 2010 An ontology-driven system for detecting global health events Nigel Collier National Inst. Informatics collier@nii.ac.jp Reiko Matsuda Goodwin Fordham University reikogoodwin@gmail.com John McCrae Bielefeld University johnmccrae@gmail.com Son Doan Vanderbilt University son.doan@vanderbilt.edu Ai Kawazoe Tsuda College zoeai@tsuda.ac.jp Mike Conway University of Pittsburgh conwaym@pitt.edu Asanee Kawtrakul Kasetart University ak@ku.ac.th Koichi Takeuchi Okayama University koichi@cs.okayama-u.ac.jp Dinh Dien VietNam National University ddien66@yahoo.com Abstract Text mining for global health surveillance is an emerging technology that is gaining increased attention from public health or- ganisations and governments. The lack of multilingual resources such as Word- Nets specifically targeted at this task have so far been a major bottleneck. This pa- per reports on a major upgrade to the BioCaster Web monitoring system and its freely available multilingual ontology; improving its original design and extend- ing its coverage of diseases from 70 to 336 in 12 languages. 1 Introduction The number of countries who can sustain teams of experts for global monitoring of human/animal health is limited by scarce national budgets. Whilst some countries have advanced sensor net- works, the world remains at risk from the health impacts of infectious diseases and environmen- tal accidents. As seen by the recent A(H5N1), A(H1N1) and SARS outbreaks, a problem in one part of the world can be rapidly exported, leading to global hardship. The World Health Organization (WHO) esti- mates that in the future, between 2 to 7.4 mil- lion people could be at risk worldwide from a highly contageous avian flu virus that spreads rapidly through the international air travel net- work (WHO, 2005). Pandemics of novel pathogens have the capacity to overwhelm health- care systems, leading to widespread morbidity, mortality and socio-economic disruption (Cox et al., 2003). Furthermore, outbreaks of live- stock diseases, such as foot-and-mouth disease or equine influenza can have a devastating impact on industry, commerce and human health (Blake et al., 2003). The challenge is to enhance vigilance and control the emergence of outbreaks. Whilst human analysis remains essential to spot complex relationships, automated analysis has a key role to play in filtering the vast volume of data in real time and highlighting unusual trends using reli- able predictor indicators. BioCaster (http://born.nii.ac.jp) (Collier et al., 2008) is a Web 2.0 monitoring station for the early detection of infectious disease events. The sys- tem exploits a high-throughput semantic process- ing pipeline, converting unstructured news texts to structured records, alerting events based on time-series analysis and then sharing this informa- tion with users via geolocating maps (Fig. 1(a)), graphs (Fig. 1(b)) and alerts. Underlying the sys- tem is a publicly available multilingual applica- tion ontology. Launched in 2006 (Collier et al., 2006) the BioCaster Ontology (BCO) has been downloaded by over 70 academic and industrial groups worldwide. This paper reports on a ma- jor upgrade to the system and the ontology - ex- panding the number of languages from 6 to 12, redefining key relations and extending coverage in the number …

Content truncated to 500 words