Cross-Domain Sentiment Analysis Based on Feature Projection and Multi-Source Attention in IoT

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Machine Learning

Background:

• Social media serves as a valuable data source for smart systems, but requires sentiment analysis.
• Cross-domain sentiment analysis addresses data scarcity in target domains by leveraging labeled source domains.
• Existing methods often neglect the negative transfer problem from irrelevant source domains.

Purpose of the Study:

• To propose a novel cross-domain sentiment analysis method, Feature Projection and Multi-source Attention (FPMA).
• To alleviate negative transfer effects and enhance feature representation for improved sentiment classification.
• To enable Internet of Things (IoT) sensors to provide user-preferred data through effective sentiment analysis.

Main Methods:

• Utilized adversarial training with two feature extractors and a domain discriminator to capture shared and private features.
• Employed orthogonal projection to optimize extracted features for multi-source domain classification.
• Implemented a multi-source attention mechanism for weighted sentiment prediction in the target domain.

Main Results:

• The proposed FPMA method demonstrated superior performance compared to baseline models on common datasets.
• FPMA effectively mitigated negative transfer by incorporating a multi-source selection strategy.
• Improved feature representation led to enhanced sentiment classification accuracy.

Conclusions:

• FPMA offers a robust solution for cross-domain sentiment analysis, particularly in data-scarce scenarios.
• The method successfully addresses the challenge of negative transfer in domain adaptation.
• FPMA enhances the reliability of sentiment analysis for smart systems and IoT applications.