An Optimized Approach for Emotion Detection in Real- Time for Twitter Sentiment Analysis with Natural Language Processing

Abstract

The automated method of identifying and deciphering the emotions expressed in written text is called sentiment analysis (SA). In the last ten years, SA has become incredibly popular in the NLP (Natural Language Processing) domain. Web-based social networking websites have become a powerful tool for influencing user perceptions and how businesses are marketed. People's opinions are very important when analyzing the effects of information propagation on people's lives in a large-scale network such as Twitter. The polarity and predisposition of a large population towards a particular topic, item, or entity can be determined by data analysis of the tweets. These days, it's easy to see how such analysis is applied in a variety of contexts, including public elections, movie marketing, brand endorsements, and many more. We will build a program that examines the content of tweets on a specific subject in this project. The main goal is to present an approach for polarity score analysis in Twitter streams with noise.

We suggest an emotive categorization of a large number of tweets in this paper. Here, we categorize an expression's sentiments into positive and negative emotions using deep learning approaches. Motivation, fun, happiness, affection, neutral, relief, and surprise are other categories for positive feelings, while anger, boredom, emptiness, hatred, sadness, and worry are categories for negative emotions. We demonstrated how to attain high emotion classification accuracy by experimenting with and evaluating the approach using recurrent neural networks and long-term short-term memory on three distinct datasets. Based on a comprehensive evaluation, the system achieves 88.47% accuracy for positive/negative classifications and 89.3% and 93.3% accuracy for both positive and negative subclasses, respectively, for emotion prediction using the LSTM model.