Identifying personal health experience tweets with deep neural networks

摘要: Twitter, as a social media platform, has become an increasingly useful data source for health surveillance studies, and personal experiences shared on Twitter provide valuable information to the surveillance. are known their irregular usages of languages informal short texts due 140 character limit, noisiness such that majority posts irrelevant any particular These factors pose challenges in identifying experience tweets from data. In this study, we designed deep neural networks with 3 different architectural configurations, after training them corpus 8,770 annotated tweets, used predict set 821 annotate tweets. Our results demonstrated significant amount improvement predicting by over conventional classifiers: 37.5% accuracy, 31.1% precision, 53.6% recall. We believe our method can be utilized various studies using source.