Automatic identification of public health related Turkish tweets
EE Küçük, Kürşad Yapar, Doğan Küçük, Doğan Küçük · European Journal of Public Health · 2016
Background People started to instantly share news/opinions through social media platforms like Twitter. Public health related concerns/news are also shared, especially on Twitter, so there is an increasing research attention on tweet analysis to track public health events. This study presents a software application that automatically detects public health related Turkish tweets, together with its performance evaluation result on a genuine tweet set. Methods First, we built a public health ontology in Turkish so that the ontology concepts (terms) can be used by our application The ontology is built semi-automatically where related Wikipedia articles are automatically processed to extract some concepts which are later revised by the authors and the remaining concepts are manually included. The concepts are from four classes: (1) generic public health terms (2) diseases (3) generic medications (4) adverse drug reactions. We implemented an application in Java, which utilizes the existence of the ontology terms in a given tweet to determine whether it is related to public health or not. Results The application was evaluated on a randomly sampled set of 1,000,000 tweets published between February, 25 and March, 16, 2015. The application identified 1,052 tweets (∼0.1%) as related to public health. To evaluate this outcome, the authors labeled these tweets and it was found that 819 tweets are related to public health while 233 tweets are not, hence, the precision of the application is 78%, which is a promising result. 386 of the correctly identified tweets (∼47%) report upper respiratory tract infections like common cold, flu, laryngitis, swine flu, and sinusitis, as expected due to the period of our set. Conclusions Social media texts stand as a valuable source of instant information about public health events. Our ontology and application, with future enhancements, could be beneficial to public health experts who need to track public health events in a timely manner. Key messages: Social media analysis can help track public health events Related software applications can be developed to be used by the experts for this purpose