Deep Learning-Based Sentiment Analysis for Roman Urdu Text

Abstract

Sentiment Analysis has significant attention due to its versatile approach to analysis user’s sentiments on various social networks, forums, e-marketing sites and blogs. Sentiments related data on the web has great importance and impact on customer’s, readers and business firms.Reccurent Neural Network has been widely applied to perform Natural Language Processing tasks because it is designed for modeling the sequential data efficiently.

In this paper we used Deep Neural Long-short time memory model (LSTM).It has extraordinary capability to Capture long-range information and solve gradient attenuation problem, as well as represent future contextual information, semantics of word sequence magnificently. This paper is the foundation of adapting Deep learning methods to perform Roman Urdu Sentiment Analysis. Our experimental results shows the significant accuracy of our model and surpassed accuracy of baseline Machine learning methods.