Automatic irony- and sarcasm detection in Social media
Erik Forslid, Niklas Wikén · KTH Publication Database DiVA (KTH Royal Institute of Technology) · 2015
This thesis looks at different methods that have been used for irony and sarcasm detection and also includes the design and programming of a machine learning model that classifies text as sarcastic or non-sarcastic. This is done with supervised learning. Two different data set where used, one with Amazon reviews and one from Twitter. An accuracy of 87% was obtained on the Amazon data with the Support Vector Machine. For the Twitter data was an accuracy of 71% obtained with the Adaboost classifier was used. The thesis is done in collaboration with Gavagai AB, which is company working with Big-data text with expertise in semantic analysis and opinion mining.