Automatic Sarcasm Detection in Arabic Text: A Supervised Classification Approach

Mohammed M. Abuteir, Eltyeb S. A. Elsamani

Sarcasm, Automatic sarcasm detection, Sentiment analysis and opinion mining, Natural language processing, Machine learning, Arabic language.

Sarcasm is a form of communication that is intended to mock or harass someone by using words with the opposite of their literal meaning. It has an implied negative sentiment, but a positive surface sentiment. However, detection of Sarcasm is somewhat difficult due to the gap between its literal and intended meaning and the different ways in which sarcasm may be expressed especially for the Arabic language which has a rich nature and very complex morphology. Detection of sarcasm is of great importance to Sentiment Analysis (SA) and it can improve the performance of many Natural Language Processing (NLP) applications. In this paper, we propose an approach for automatic detection of sarcasm in the Arabic text of Twitter data by building the first Arabic corpus to study sarcasm in text and applying different supervised classification techniques such as Naïve Bayes (NB), Logistic Regression (LogR), and Random Forest (RF) based on a set of lexical and structural features. The experimental results obtained are promising. The classifiers have a comparable classification performance on the sarcasm identification task with a slight difference. The best results for all feature sets achieved using the NB classifier with the overall accuracy equal to 89.17%, which these results are quite high especially regarding Arabic text.