Sentiment Analysis in Roman Urdu at the Sentence Level through Advanced Deep Learning Technique

Sentiment analysis (SA) helps in expressing whether a test or textual review leans towards positivity, negativity, or neutrality. In this research study, SA has been conducted on Roman Urdu reviews SA. We have collected 35,139 reviews from seven different domains for this research, and these reviews have been categorized into five classes: "very positive," "very negative," "positive," "negative," and "neutral". To build Roman Urdu (RU) SA model, we have applied deep learning (DL) algorithms, including Recurrent Neural Network-Long Short-Term Memory (RNN-LSTM), Recurrent Neural Network-bidirectional long short-term memory (RNN-BiLSTM), Gated Recurrent Unit (GRU), Bi-Gated Recurrent Unit (BiGRU), and Recurrent Convolutional Neural Network (R-CNN). To achieve better results with these algorithms, we have incorporated six hidden layers within each classifier to maximize accuracy. In our experimental study, we found that using 64 hidden layers resulted in good accuracy for all classifiers except for the R-CNN, which achieved good accuracy with only 16 hidden layers.