Text Sentiment Analysis Based on a New Hybrid Network Model.

Area of Science:

• Natural Language Processing
• Deep Learning
• Artificial Intelligence

Background:

• Current deep learning models for text sentiment analysis often struggle with semantic information loss.
• Improving the accuracy of sentiment analysis requires better understanding of nuanced semantic details.

Purpose of the Study:

• To propose a novel hybrid sentiment analysis model that minimizes semantic information loss.
• To enhance prediction accuracy in text sentiment analysis through an integrated deep learning approach.

Main Methods:

• A hybrid model combining doc2vec, Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and an attention mechanism was developed.
• Doc2vec pre-trained paragraph vectors were used to capture overall semantic information, reducing information loss.
• CNN extracted local text features, while BiLSTM handled context interaction, and the attention mechanism prioritized important information.

Main Results:

• The proposed doc2vec+CNN+BiLSTM+Attention model achieved 91.3% accuracy on the IMDB dataset and 93.3% on the DailyDialog dataset.
• The model demonstrated a loss rate of 22.1% on IMDB and 19.9% on DailyDialog.
• Accuracy was 1.0% and 0.5% higher than existing CNN-BiLSTM-Attention and ATT-MCNN-BGRUM models on the IMDB dataset.

Conclusions:

• The novel hybrid model effectively leverages the strengths of each component (doc2vec, CNN, BiLSTM, attention) to improve sentiment analysis.
• The proposed model shows superior performance and effectiveness compared to existing methods, offering a significant advancement in text sentiment analysis.