Tri-Clustered Tensor Completion for Social-Aware Image Tag Refinement.

Area of Science:

• Computer Science
• Information Retrieval
• Machine Learning

Background:

• Social image tag refinement is crucial for effective social image search.
• Existing methods often overlook valuable user information, limiting tag accuracy.
• Improving tag quality involves completing missing tags and correcting noisy ones.

Purpose of the Study:

• To propose a novel tri-clustered tensor completion framework for social image tag refinement.
• To collaboratively explore visual, tag, and user information for enhanced tag accuracy.
• To overcome computational challenges in large-scale tensor factorization for social image data.

Main Methods:

• A tensor model represents inter-relations among users, images, and tags.
• Three regularizations capture intra-relations within users, images, and tags.
• A tri-clustering method divides the tensor into sub-tensors for efficient processing.
• Two strategies are employed for sub-tensor completion, considering dependencies.

Main Results:

• The proposed tri-clustered tensor completion framework significantly improves social image tag refinement.
• Experimental results demonstrate superior performance compared to state-of-the-art methods on a real-world dataset.
• The integration of user information proves vital for accurate tag completion and correction.

Conclusions:

• The novel framework effectively refines social image tags by leveraging user, image, and tag data.
• Tri-clustering tensor completion offers a scalable and efficient solution for large-scale social image analysis.
• This approach enhances the overall performance of social image search systems.