Project of Text Analytics course, Data Science and Business Informatics - University of Pisa
A. Carnevale, G. Marcoccio, S. Telera
Philosophers and rhetoricians have been interested in irony and sarcasm for over 2500 years. In recent times, they have been popular issues discussed; in this project will be taken an Irony and Sarcasm detection on Tweets provided by an Evalita Challenge, called IronITA (Irony Detection in Italian Tweets)
- Binary classification ironic / not ironic
- Multi Class Classification: not ironic / ironic but not categorized as sarcastic / sarcastic
For the development of these tasks initially we have undertaken a Data Understanding analysis trying to figure out our data, later through a Data Preprocessing we manipulated the raw text we had, in order to obtain different types of it. Finally we did some analysis like distribution of words, using a WordCloud to have a representative visualization. Following, a classification analysis with simple classifiers has carried out (SVM, Naive Bayes and Decision Tree) both binary and multi class and we have tried to improve the classification results by integrating sentiment and emotions features. Finally, we developed LSTM and BERT to have an exhaustive overview on the classification task with models specifically suited for NLP.