Paper Key : IRJ************294
Author: Ananya Vaibhav Bhopatrao
Date Published: 22 Oct 2023
Abstract
The ability to discern and classify emotions within text data has garnered increasing attention as text-based communication continues to proliferate across diverse digital platforms. This research explores the domain of emotion detection in text, with a focus on understanding the underlying methodology and ethical considerations. The first phase of this study involves the collection and preprocessing of a heterogeneous dataset encompassing a wide array of text sources. Text data is subjected to a series of cleaning and tokenization steps, and each text sample is manually labeled with one of seven primary emotions: Joy, Fear, Anger, Sadness, Disgust, Shame, or Guilt.Feature extraction is carried out through TF-IDF vectorization and the consideration of n-grams. Subsequently, a variety of machine learning models, including Support Vector Classifiers (SVC), Linear Support Vector Classifiers (LinearSVC), Random Forest Classifiers, and Decision Tree Classifiers, are experimented with for emotion detection. The study incorporates data visualizations as an integral part of the methodology. Pie charts are utilized to visualize the distribution of emotions in the dataset, while word clouds offer insights into the most frequent words associated with each emotion. The research evaluates model performance on a separate test dataset and engages in ethical considerations by addressing potential biases and fairness in emotion detection models. Our in-depth study helps us better understand how to detect emotions in text. We look at how we collect data, choose features, and pick the right models, all while considering ethical aspects. The findings from this research can be useful for emotion detection in different areas.