Lancaster at SemEval-2018 Task 3: Investigating Ironic Features in English Tweets

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 587–593 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics Lancaster at SemEval-2018 Task 3: Investigating Ironic Features in English Tweets Edward Dearden and Alistair Baron School of Computing and Communications Lancaster University Lancaster, UK, LA1 4WA initial.surname@lancaster.ac.uk Abstract This paper describes the system we submit- ted to SemEval-2018 Task 3. The aim of the system is to distinguish between irony and non-irony in English tweets. We create a tar- geted feature set and analyse how different features are useful in the task of irony detec- tion, achieving an F1-score of 0.5914. The analysis of individual features provides insight that may be useful in future attempts at detect- ing irony in tweets. 1 Introduction With so many people using social media and mi- croblogs such as Twitter, a huge amount of natural language data is available to be analysed. It is de- sirable to be able to accurately interpret what peo- ple are saying. An example of this is in the field of sentiment analysis where the aim is to determine whether the language being used by an author is positive or negative. These kinds of tasks are made more difficult by the presence of figurative language. Figura- tive language is a type of language where the con- tents of the text are not literally true, making it difficult to ascertain the true meaning of the text purely from the content. Irony is a particular type of figurative language in which the meaning is of- ten the opposite of what is literally said and is not always evident without context or existing knowl- edge. A system capable of accurately detecting irony would be a valuable addition to sentiment analysis systems, and other systems for natural language understanding which are confounded by irony. In online discourse, examples of irony are very common. Social media platforms capture natural language which often includes sarcastic sentences. An example of a use for irony detection is in the area of online product reviews (Tsur et al., 2010) which can contain large amounts of ironic lan- guage. An irony detection system could be used to prevent ironic negative reviews being misinter- preted as positive and highlighted in advertising. The system described in this paper aims to iden- tify targeted features of irony and analyse how im- portant they are in identifying ironic tweets. The annotated twitter data we use is that provided by the event organisers. The task is described in the task description paper (Van Hee et al., 2018). 2 Related Work The task of irony detection is an inherently diffi- cult one. Wallace (2015) suggests that to create a good system for irony detection, one cannot rely on lexical features such as Bag of Words, and one must consider also semantic features of the text. There have been various methods employed to detect irony. Reyes et al. (2012) created a dataset generated by searching for user-created tags and attempted to identify …

Content truncated to 500 words