Text Based Emotion Detection by Using Classification and Regression Model

Abstract

In South Asia Urdu is widely spoken language. Limited amount of literature for Urdu language is exit but more is required for better implementation of NLP tasks. In text-based representation the emotions play vital role. Emotions are expression to express the feeling about any condition, but this expression is not easily understandable in text. In this research the major focus is to identify the emotions from roman Urdu. In the literature emotion-based work is reported on several different languages but limited literature is found for specifically Roman Urdu. It is needed to explore this area because Roman Urdu has also significant importance like other languages. In emotion detection the text is used instead of facial expression which quite different. Aim of study is to summarize each sentence in a single emotion tag. Each tag is a annotation for the expression which give a independent sense. Ten-fold cross validation is performed. Emotion analyses have many useful applications like product feedback, improvement in product quality and mental health. To address the issue of emotional polarity of Roman Urdu sentence, corpora of 5K is developed from different domain and annotate four emotion classes. For the validation and verification of this work few baseline algorithms are used on our corpus. An experiment is conducted by using KNN, DT, SVM and RF algorithm and better F measure score is achieved.