A deep learning model for detecting sarcasm in written product reviews

• Nicole Schwarz

The ability to reliably identify sarcasm and irony in a text can improve the performance of many Natural Language Processing systems, including summarization, sentiment analysis, etc. The existing sarcasm detection systems have focused on identifying sarcasm on a sentence level or for a specific phrase. However, often, it is impossible to identify a sentence containing sarcasm without knowing the context. This thesis aims at showing the possibilities of sarcasm detection in Amazon product reviews using a deep neural network. The data set for this project was acquired from the Github repository of Elena Filatova, who has already done research on collecting sarcastic Amazon reviews. As the task of irony detection yearns for corresponding algorithms, that can resolve contextual problems, the approach of using a deep learning model, containing a Convolutional Neural Network and a Long Short-Term Memory Network, was taken. Both can handle classifying texts highly depending on the context, as they have memory units to remember already learned words in a sentence. The results have shown that a deep neural network approach outperforms simpler models in their accuracy. Nevertheless, the best results have not yet been achieved due to the limit of data size.