Attention-Emotion-Enhanced Convolutional LSTM for Sentiment Analysis.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Machine Learning

Background:

• Deep learning models, including Long Short-Term Memory (LSTM) networks, are prevalent in text sentiment analysis.
• Current models often overlook the impact of emotion on feature extraction and utilize word- or sentence-level attention, potentially limiting performance.
• Higher-level abstractions are crucial for effective sentiment feature learning.

Purpose of the Study:

• To propose a novel model, AEC-LSTM, for improved text sentiment detection.
• To enhance LSTM networks by integrating emotional intelligence (EI) and attention mechanisms.
• To address the limitations of existing models in capturing complex sentiment features.

Main Methods:

• Developed an emotion-enhanced LSTM (ELSTM) using EI for improved feature learning via an emotion modulator and estimator.
• Integrated ELSTM with convolution, pooling, and concatenation to capture diverse text sequence patterns.
• Introduced a topic-level attention mechanism to adaptively weight text hidden representations.

Main Results:

• The proposed AEC-LSTM model effectively utilizes sentiment semantic information from text topics and context.
• Experiments demonstrate significant improvements in sentiment classification performance.
• The approach outperforms state-of-the-art deep learning-based methods on real-world datasets.

Conclusions:

• AEC-LSTM offers a more effective approach to sentiment representation and classification.
• Integrating EI and topic-level attention enhances the capability of deep learning models for sentiment analysis.
• The proposed model represents a significant advancement in text sentiment detection research.