Can machines sense irony? : exploring automatic irony detection on social media
Cynthia Van Hee · Ghent University Academic Bibliography (Ghent University) · 2017
The development of the social web has stimulated creative language use like irony.As a result, research in automatic irony detection has thrived in the past few years, to improve our understanding of ironic language on the one hand, and to enhance text mining applications that suffer from irony (e.g.automatic sentiment analysis) on the other.In this thesis, we present a comprehensive approach to modelling irony, including the development of a new fine-grained annotation scheme, a varied set of experiments to detect irony automatically, and an extrinsic evaluation of the irony detection system by means of a sentiment analysis use case.An important contribution of this research includes a new approach to model implicit or prototypical sentiment, which is crucial in irony detection.We assembled a gold-standard corpus of English tweets using irony-related hashtags (i.e.#irony, #sarcasm, #not), which was manually annotated according to a new annotation scheme.The scheme is grounded in irony literature and provides for a fine-grained annotation, including the identification of different forms of irony and the specific text spans that realise the irony in a tweet.This manually annotated dataset allowed us to investigate two things: the linguistic realisation of irony in online text, and the viability of our machine learning approach to irony detection.Analysis of the annotated corpus analysis revealed that one in five instances in the corpus are not ironic despite containing an irony hashtag, which confirms that manual annotations are instrumental for this task.We also observed that i As a colleague and friend would have it, "you don't write a thesis on your own".I am happy to look back at the past four years, which have been an enjoyable and challenging experience, and to thank the people who have supported me in one way or another during this period.First of all, I would like to thank my supervisor, Prof. Dr. Véronique Hoste and copromotor Prof. Dr. Els Lefever, who have been instrumental to this work.Véronique, thank you for giving me the opportunity to work in the stimulating environment LT3 is.Thank you for your trust and ambitious goals, which have made me achieve things that I would not have imagined possible a few years ago.Els, thank you for your help with many struggles, and for your ability to make the biggest obstacles seem surmountable.Your optimism and kindness are exceptional.