Roman Urdu Sentiment Analysis of Songs‘ Reviews

Abstract

To process Natural Language reviews using Machine Learning techniques is known as Sentiment Analysis. It is a way to categorize people's opinions, sentiments, and attitudes towards a specific entity. Due to easy access to the internet and smart devices, people are becoming habitual in posting reviews about any specific entity/product, they use. These reviews are very helpful for all types of users in decision-making. In the past, most of the work in Sentiment Analysis was carried out on resource-rich language but very little literature is witnessed on resource-poor languages. Very few efforts have been made to build language resources to process the Roman Urdu language. This research targets to perform Sentiment Analysis on Urdu (i.e. source-poor language) in Roman script. To perform sentiment analysis of roman urdu reviews on songs, the dataset is generated from the comments on songs. Three songs from the Sub-continent music industry opt from YouTube. After pre-processing the reviews, Roman Urdu reviews are analysed using Naïve Bayes, KNN, Decision Tree (ID3) and ANN. Naïve Bayes outperforms the other classifiers and achieved 82.41% results in terms of accuracy.