Sentiment Analysis of Roman Urdu Text Using Machine Learning Techniques

Abstract

Social media has attained popularity during the last few decades due to the rapid growth of online businesses and social interaction. People can interact with one another and communicate their sentiments by expressing their ideas and points of view on social media. Businesses involved in manufacturing, sales, and marketing increasingly focus on social media to get feedback on their goods and services from people worldwide. Businesses must process and analyze this feedback in the form of sentiments to gain business insights. Every day, millions of Urdu and Roman Urdu sentences are posted on social media platforms. The critical loss of this massive amount of data results from ignoring the thoughts and opinions in language with limited resources, such as Urdu and Roman Urdu in the favor of resource-rich languages, such as English. The current study focused on sentiment analysis of Roman Urdu text. Bag of Words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF) word embedding techniques were deployed to conduct the current study. Support Vector Machine (SVM), Linear Support Vector Machine (LSVC), Logistic Regression (LR), and Random Forest (RF) classifiers were deployed. The experiments showed that SVM showed 94.74%, while RF showed 93.13% accuracy using BoW word embedding technique