Sentiment Analysis of Urdu Text using Hybrid Deep Learning Techniques

Abstract

Sentiment Analysis (SA) has become the rising topic of research in data mining. There is a massive increase in the use of the Internet and business e-commerce applications. With the excess amount of text content on the internet, Sentiment Analysis has begun to attract the attention of more people. Sentiment analysis of single sentence referred as short text, is a very tricky task because single sentence does not have much contextual information. Furthermore, not much work has been done in field of sentiment analysis of Urdu text because a very few amounts of data and recourses of Urdu text are available publicly. The purpose of this research is to contribute in the field of sentiment analysis of Urdu text. Mostly in Natural Language Processing (NLP), tasks words are the main focus. This research proposed a deep learning techniques combination of Convolutional Neural Network (CNN) with Long Short-Term Memory (LSTM) for the Sentiment Analysis of Urdu Text. In this technique, word embedding is performed using fastext API which poses a high dimensionality (300 dimensions). Long Short-Term Memory helps capturing the long-term dependencies and minimizes the loss of local information. We validated these proposed techniques on the Urdu sentiment analysis dataset and compare with Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM). The results shows that the proposed technique LSTM-CNN has significantly by achiving testing accuracy 96.9% training accuracy 99.5, precision P 96.0 , precision N 98.0, Recall P 98.0, Recall N 96.0, F1-score N 97.0 , and F1-score N 97.0.The findings shows that CNN is a strong option for the given problem as it improves classification accuracy when combined with LSTM.