Deducted Sentiment Analysis for Sarcastic Reviews Using LSTM Networks

Abstract

Sentiment analysis is one among various artificial intelligence domains which has become predominant in various fields of study and applications. Sentiment analysis is widely used in product reviews, movie reviews, social tweets, etc. Sentiment analysis takes a heap of reviews and predicts the sentiment collectively conveyed by them. Sarcasm is one of the complications in sentiment analysis makes the review normal syntactically but not semantically. Sarcastic reviews can be treated as normal inputs if sentiment analysis systems are capable of getting optimal sentiment from the review though it is full of sarcasm. Conventional algorithms and neural networks used for sentiment analysis are not capable to produce the optimal results for sarcastic reviews. As in advancement of deep learning, long short-term memory (LSTM) networks were introduced to give a solution to the sequence problems. In this paper, we mainly focused on sentiment analysis as sequence problem and binary classification problem. The review has been taken as input sequence to our proposed LSTM model that is capable to predict probability of two classes efficiently. The relatively optimal results have been achieved with our LSTM network.