Deep-learning based framework for sentiment analysis in Urdu language

Abstract

In recent times, Sentiment analysis has become a significant means for framing a successful business and can be very helpful in predicting customer trends to help organizations in their decision-making process. Though many software applications are available in the market for text analysis, one of the major limitations of such applications is that they are developed for rich languages like English, German, Spanish, Arabic, etc. and less popular languages like Urdu, Hindi, Roman Urdu are neglected due to lack of availability of resources. Therefore, this research project will provide an implementation of sentiment analysis in the Urdu language. Firstly, preprocessing is performed and a small-scale manual dictionary of 830 Urdu stem words is introduced for stemming. Then a deep learning-based framework of LSTM is used for Urdu sentiment analysis. Experimental results show high classification accuracy of 86.03% and 0.89 F1 Score with the use of LSTM that captures sequence information effectively to analyze sentiments than the conventional supervised machine learning approaches.