A BERT Framework to Sentiment Analysis of Tweets.

Area of Science:

• Natural Language Processing
• Machine Learning
• Computational Linguistics

Background:

• Sentiment analysis is crucial for understanding opinions on microblogging platforms like Twitter.
• Existing methods struggle with context due to varied text lengths and ambiguous emotional cues.
• User context is vital for accurate sentiment interpretation.

Purpose of the Study:

• To improve sentiment analysis accuracy by incorporating user context.
• To evaluate the effectiveness of Bidirectional Encoder Representations from Transformers (BERT) for sentiment classification.
• To compare BERT-based models with traditional methods like Word2vec.

Main Methods:

• Utilized Bidirectional Encoder Representations from Transformers (BERT) for text classification.
• Integrated BERT with Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Bidirectional Long Short-Term Memory (BiLSTM).
• Compared performance against BERT with Word2vec and BERT without variants.

Main Results:

• BERT combined with CNN, RNN, and BiLSTM demonstrated superior performance.
• These hybrid models achieved higher accuracy, precision, recall, and F1-scores.
• The proposed BERT-based approaches outperformed Word2vec and baseline models.

Conclusions:

• BERT-based models significantly enhance sentiment analysis by capturing user context.
• Hybrid architectures (BERT-CNN, BERT-RNN, BERT-BiLSTM) are effective for microblogging sentiment classification.
• The study validates the superiority of advanced transformer models for nuanced sentiment understanding.