CTSys at SemEval-2018 Task 3: Irony in Tweets

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 576–580 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics  CTSys at SemEval-2018 Task 3: Irony in Tweets Myan Sherif Faculty of Engineering Alexandria University sherif.myan@gmail.com Sherine Mamdouh Faculty of Engineering Alexandria University sherym51@gmail.com Wegdan Ghazi Faculty of Engineering Alexandria University wegdan.ghazi@gmail.com Abstract The objective of this paper is to provide a description for a system built as our participation in SemEval-2018 Task 3 on Irony detection in English tweets. This system classifies a tweet as either ironic or non-ironic through a supervised learning approach. Our approach is to implement three feature models, and to then improve the performance of the supervised learning classification of tweets by combining many data features and using a voting system on four different classifiers. We describe the process of pre-processing data, extracting features, and running different types of classifiers against our feature set. In the competition, our system achieved an F1- score of 0.4675, ranking 35th in subtask A, and an F1-score score of 0.3014 ranking 22th in subtask B. 1 Introduction Irony detection in text has extended to different data forms (tweets, reviews, TV series dialogues), our domain of data in this task is a Twitter corpus provided by SemEval2018 organizers. Here, irony detection refers to computational approaches to predict if a given text is sarcastic. This problem is hard because of the nuanced ways in which irony may be expressed. The most difficult part of the problem mentioned is the process of feature engineering, because it defines the parameters and the relationships and dependencies between semantic meanings, and gives us the numerical model that the classifier would proceed to work on, thus being crucial to the soundness and efficiency of the system. This led us to dive into deeper questions, such as the nature of tweets, and how we are dealing with a version of the English language that is not directly workable. We need to perform pre- processing to deal with annotations and hashtags. Another question is how to analyze irony in English language and derive a rule-based approach that can be implemented to better understand the semantics of ironic text. The problem, as described by the SemEval- 2018 task organizers, addresses both the binary distinction between irony and non-irony, as well as different types of irony. 1.1 Task Description The SemEval-2018 Task 3 is divided into two subtasks: Subtask A is a binary classification problem where we are asked to classify a tweet as ironic or not ironic, based on a given training set of labeled tweets (0 for non-ironic and 1 for ironic). Subtask B is a multi-classification problem where we classify the tweets to which type they belong as either situational irony or verbal irony or other irony or not ironic. Each tweet in the training set is labeled as follows: (0 for non-ironic, 1 for situational irony, 2 for verbal irony, and 3 for other forms of irony). 1.2 The Dataset The used …

Content truncated to 500 words