Multimodal Feature Fusion Based Hypergraph Learning Model.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Hypergraph learning models rely on hypergraph structure quality and incidence matrices.
• Existing models use single feature extraction methods, limiting their abstract expression ability.
• A need exists for improved feature fusion in hypergraph learning.

Purpose of the Study:

• To propose a multimodal feature fusion method for hypergraph learning.
• To enhance the feature extraction and abstract expression capabilities of hypergraph models.
• To reduce the computational cost associated with multimodal hypergraph learning.

Main Methods:

• Built single modal hypergraph structures using different feature extraction methods.
• Extended hypergraph incidence and weight matrices for multimodal feature fusion.
• Proposed a Laplacian matrix fusion method for computational efficiency and performance enhancement.

Main Results:

• The multimodal feature fusion model demonstrated superior classification performance compared to single modal models.
• Laplacian matrix fusion reduced average computation time by approximately 40% compared to extended incidence matrices.
• The F1 index improved by 8.4% after Laplacian matrix fusion, indicating enhanced classification accuracy.

Conclusions:

• Multimodal feature fusion significantly improves hypergraph learning model performance.
• Laplacian matrix fusion offers an efficient and effective approach to integrating multimodal information.
• The proposed method enhances both the feature expression ability and classification accuracy of hypergraph learning.