A Multi-Modal Fusion Method Based on Higher-Order Orthogonal Iteration Decomposition.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Fusion

Background:

• Multi-modal fusion leverages diverse data sources for improved predictive performance.
• Existing fusion techniques can suffer from redundant information and high dimensionality.
• Accurate data fusion is crucial for applications like sentiment analysis and trait recognition.

Purpose of the Study:

• To propose a novel multi-modal fusion method, Higher-order Orthogonal Iteration Decomposition and Projection (HOIDP).
• To enhance prediction accuracy by effectively removing inter-modal redundancy and minimizing information loss.
• To validate the proposed HOIDP method on diverse multi-modal datasets.

Main Methods:

• The Higher-order Orthogonal Iteration Decomposition algorithm is employed for data decomposition.
• Factor matrix projection is utilized to reduce dimensionality and remove redundant features.
• The HOIDP method is applied to multi-modal datasets for fusion and prediction.

Main Results:

• The HOIDP method demonstrated significant performance improvements across three multi-modal datasets.
• Sentiment analysis accuracy improved by 0.4% to 4%.
• Personality trait recognition accuracy improved by 0.3% to 8%.
• Emotion recognition accuracy improved by 0.2% to 25% compared to five other methods.

Conclusions:

• The proposed HOIDP method effectively reduces redundant information and parameters in multi-modal fusion.
• HOIDP achieves superior accuracy in sentiment analysis, personality trait recognition, and emotion recognition.
• This method offers a promising approach for enhancing multi-modal data fusion tasks.