A Comparative Study of Increasing Automation in the Integration of Multilingual Social Media Information

Multilingual support in global applications that integrate and filter social media data is a significant challenge due to the cost of manually developing such social media filters for each language. Using LITMUS landslide information system as an experimental platform, we compared six design alternatives with varied combinations of manually developed filters and automatically translated filters for integrating and filtering social media data. Our experiments on Japanese tweets show that automatically translated filtering produces comparable or better results than manually created filters, achieving similar result quality: in false positives, false negatives, and F-1 scores. Compared to manual development, our results suggest automated translation may be a faster and cheaper approach to the integration of non-English languages, without sacrificing data quality. These results are encouraging for reducing the development cost of localized applications that integrate social media, by using automated translation tools.