Abstract:
This work aims to classify tweets into multiple topics and predict emotions and sentiments from tweets by using techniques like Natural Language Processing (NLP), Transformers, Deep Learning, and Machine Learning based on the context and meaning of tweets. This approach helps in the easy prediction and analysis of tweets that might be perceived as hate speech, misinformation, or other issues. Furthermore, this work seeks to improve the comprehension of the emotions conveyed in tweets, hence offering insightful solutions to problems with online communication. The proposed methodology showcases the potential of combining NLP and deep learning effectively to decode sentiments and contribute to the creation of a safer and more informed digital environment. The proposed system involves three main steps. Firstly, data pre-processing is done using Tokenization, Stemming, and Lemmization to get noiseless data. Then, feature engineering using LDA, BERTopic, and GloVe embeddings. Feature extraction is done using BERT and XLNet Transformers. In order to achieve this, we utilized a variety of Gated Recurrent Units (GRU), Bidirectional GRU (Bi-GRU), Long Short-Term Memory (LSTM) networks, and Bidirectional Encoder Representations from Transformers (BERT) on the sentiment analysis dataset. LSTM networks are useful for understanding the context of tweets because of their capacity to extract relationships over time from sequential input. Because they process sequences efficiently and achieve a compromise between computational expenses and results, GRU and bi-GRU models are used. The implementation of BERT, a modern transformer model, comes from its unique ability to understand the complexity and contextual meanings of words within tweets. The proposed system was assessed on the Emotions dataset and achieved an accuracy of 98.16%. The experimental results show that the proposed method outperforms the existing approach in sentiment analysis using tweet data.
