Building a Twitter Sentiment Analysis System with Recurrent Neural Networks.

Area of Science:

• Natural Language Processing
• Machine Learning
• Social Media Analytics

Background:

• Twitter sentiment analysis is crucial for understanding public opinion.
• Recurrent Neural Networks (RNNs) are effective for sequence data like text.
• Existing methods require enhancements for improved accuracy.

Purpose of the Study:

• To develop and evaluate an RNN-based sentiment analysis solution for tweets.
• To integrate an attention mechanism to boost classification performance.
• To provide a robust sentiment analysis component for the SynergyCrowds platform.

Main Methods:

• Utilized Recurrent Neural Networks (RNNs) for tweet classification.
• Implemented an attention mechanism at memory cell and network levels.
• Experimented with 20 different design approaches to optimize the model.

Main Results:

• Achieved an 80.74% accuracy rate for binary sentiment classification.
• Demonstrated the effectiveness of the attention mechanism in enhancing RNN performance.
• The developed solution serves as a core component in the SynergyCrowds platform.

Conclusions:

• The proposed RNN solution with an attention mechanism offers a high-accuracy approach to Twitter sentiment analysis.
• The two-way localization system within the attention mechanism contributes to improved classification.
• This work provides valuable insights into building effective sentiment analysis tools for social media data.