Heuristic multi-scale feature fusion with attention-based CNN for sentiment analysis.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Machine Learning

Background:

• Sentiment analysis automates insights from user-generated content but faces challenges due to limited labeled data.
• Traditional sentiment analysis methods have evolved, with deep learning approaches showing promising performance.
• Existing deep learning models require further optimization for enhanced sentiment analysis accuracy.

Purpose of the Study:

• To propose an attention deep learning model for improved sentiment analysis.
• To address the scarcity of labeled data in Natural Language Processing (NLP) for sentiment analysis.
• To enhance the accuracy and efficiency of sentiment analysis using a novel heuristic approach.

Main Methods:

• Data was collected from public resources and pre-processed to remove irrelevant information.
• A Multiscale Feature Fusion-based Adaptive and Attention-based Convolution Neural Network (MFF-AACNet) was developed.
• Features were extracted using Bidirectional Encoder Representations from Transformers (BERT), Transformers, and word2vector, then fused and processed by MFF-AACNet.
• Parameter tuning was performed using an improved Fitness Opposition of Rat Swarm Optimizer (FORSO).

Main Results:

• The proposed MFF-AACNet model demonstrated superior performance compared to traditional sentiment analysis methods.
• Feature extraction from BERT, Transformers, and word2vector, combined with multiscale fusion, improved analysis.
• The FORSO optimizer effectively tuned model parameters, enhancing overall accuracy.

Conclusions:

• The developed attention deep learning framework offers a significant advancement in sentiment analysis.
• The proposed model effectively mitigates the challenge of insufficient labeled data in NLP.
• This approach provides a more accurate and robust solution for analyzing user-generated content sentiments.