#WarTeam at SemEval-2017 Task 6: Using Neural Networks for Discovering Humorous Tweets

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 407–410, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics #WarTeam at SemEval-2017 Task 6: Using Neural Networks for Discovering Humorous Tweets Iuliana Alexandra Fleșcan-Lovin-Arseni, Ramona Andreea Turcu, Cristina Sîrbu, Larisa Alexa, Sandra Maria Amarandei, Nichita Herciu, Constantin Scutaru, Diana Trandabăț, Adrian Iftene University Alexandru Ioan Cuza of Iași, Romania {alexandra.flescan, ramona.turcu, cristina.sirbu, larisa.alexa, sandra.amarandei, nichita.herciu, dtrandabat, adiftene}@info.uaic.ro Abstract This paper presents the participation of #WarTeam in Task 6 of SemEval2017 with a system classifying humor by comparing and ranking tweets. The training data con- sists of annotated tweets from the @midnight TV show. #WarTeam’s system uses a neural network (TensorFlow) hav- ing inputs from a Naïve Bayes humor clas- sifier and a sentiment analyzer. 1 Introduction One of the most recent direction in Artificial In- telligence is related to humor and, in recent years, comedy based computing such as Manatee (Gustin, 2014), the joke writing computer, STANDUP - System to Augment Non-Speakers’ Dialogue Using Puns (Waller et al., 2009); SASI the sarcasm-detector (Davidov et al., 2010), or DeviaNT (Kiddon and Brun, 2011) were devel- oped, with more or less success (Leybovich 2017). If the well-hidden structure of humor, from which are derived all uncertainties, would be uncovered, it would have great applicability in social networks and human computer interactive systems. In time, research has been made and progress is undeniable. However, most recent studies are concerned with a binary perspective over humor where two main features are ignored: its continuous nature and subjectivity. Our objectives in Task 6 of SemEval 2017 (Potash et al., 2017) were: (1) to build an applica- tion able to score the degree of humor in tweets from the Midnight TV show, the Hashtag War section and (2) to discover ways to automatically determine amusement and how to quantify it. The paper is structured in 5 sections: Section 2 discusses existing approaches to humor detection and Section 3 presents the methodology of our system. Section 4 briefly analyses the obtained results, before Section 5 drafting some conclu- sions and further work. 2 State of the Art In the area of identifying, describing and evaluat- ing humor, the majority of studies succeeded only to describe if something is funny or not. The actu- al tendency is to move forward to something more specific, namely to the value or the degree of hu- mor. Currently, studies are mainly concerned with the binary evaluation of humor, whether it is fun- ny or not. Their object of study is different as some of them focused on evaluating humor in videos and images, while others in texts expressed in natural language. As for the studies related to identifying humor in pictures (Chandrasekaran et al., 2016), theories in this area suggest that humor’s key components are qualities such as unexpectedness, incongruity, pain, as observed by analyzing a database of 6,400 funny and not funny images. The linguistic side of this computational ap- proach identifies the mechanisms for …

Content truncated to 500 words