TakeLab at SemEval-2017 Task 6: #RankingHumorIn4Pages

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 396–400, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics TakeLab at SemEval-2017 Task 6: #RankingHumorIn4Pages Marin Kukovaˇcec, Juraj Malenica, Ivan Mrši´c, Antonio Šajatovi´c, Domagoj Alagi´c, Jan Šnajder Text Analysis and Knowledge Engineering Lab Faculty of Electrical Engineering and Computing, University of Zagreb Unska 3, 10000 Zagreb, Croatia {name.surname}@fer.hr Abstract This paper describes our system for hu- mor ranking in tweets within the SemEval 2017 Task 6: #HashtagWars (6A and 6B). For both subtasks, we use an off-the-shelf gradient boosting model built on a rich set of features, handcrafted to provide the model with the external knowledge needed to better predict the humor in the text. The features capture various cultural ref- erences and specific humor patterns. Our system ranked 2nd (officially 7th) among 10 submissions on the Subtask A and 2nd among 9 submissions on the Subtask B. 1 Introduction While extremely interesting, understanding humor expressed in text is a challenging natural language problem. Besides standard ambiguity of natu- ral language, humor is also highly subjective and lacks an universal definition (Mihalcea and Strap- parava, 2005). Moreover, humor should almost never be taken at face value, as its understanding often requires a broader context – external knowl- edge and common sense. On top of that, what is funny today might not be funny tomorrow, as hu- mor goes hand in hand with ever-changing trends of popular culture. Even though there has been some work on hu- mor generation (Petrovi´c and Matthews, 2013; Valitutti et al., 2013), most work has been con- cerned with humor detection, a task of classify- ing whether a given text snippet is humorous (Mi- halcea and Strapparava, 2005; Kiddon and Brun, 2011; Yang et al., 2015; Chen and Lee, 2017). However, this research was mostly focused on a simple binary detection of humor. In this paper, we describe a system for rank- ing humor in tweets, which we participated with in the SemEval-2017 Task 6 (Potash et al., 2017). It comprised two subtasks, one dealing with pre- dicting which tweet out of two is more humorous, and other with ranking a set of tweets by their humorousness. Even though these tasks can be both posed and tackled differently, we straightfor- wardly used the obtained pairwise classifications from the first task in coming up with ranked lists for the second. Our system uses a standard gra- dient boosting classifier (GB) based on a rich set of features and collections of external knowledge. We ranked 2nd among 10 submissions (7th offi- cially) at the Subtask 6A, and 2nd among 9 sub- missions at the Subtask 6B. 2 Task Description The dataset provided by the task organizers com- prises the tweets collected from many episodes of the Comedy Central show @midnight.1 This game show is based around contestants and view- ers providing humorous and witty tweets in re- sponse to a given topic (hashtag), which are then ranked by their humorousness. The compiled dataset …

Content truncated to 500 words