Research map / Graph learning / Knowledge and temporal graphs
Social Anchor-Unit Graph Regularized Tensor Completion for Large-Scale Image Retagging
Jinhui Tang, Xiangbo Shu, Zechao Li, Yu–Gang Jiang, Qi Tian
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use anchor-unit graph regularized tensor completion instead of sub-tensor completion (TTC2) | SUGAR-TC · NUS-WIDE-128 image retagging | average F-score: 0.458 -> 0.485 | Improved |
| add user information via tensor (tag-image-user) instead of image-tag matrix | MRTF/TTC/SUGAR-TC · NUS-WIDE-128 image retagging | average F-score: 0.417 -> 0.447 | Improved |
| solve the super-sparse tensor problem by sub-tensor division (TTC1 vs MRTF) | TTC1 · NUS-WIDE-128 image retagging | average F-score: 0.423 -> 0.447 | Improved |
| use anchor units instead of all data for tensor completion | SUGAR-TC · NUS-WIDE-128 image retagging computational cost | total computing time (h): 35.5 -> 4.2 | Improved |
| use anchor units instead of full tensor | SUGAR-TC · NUS-WIDE-128 computational cost | memory of tensor completion (GB): 6.6 -> 10.1 | Mixed |
| use co-clustering anchors rather than TTC tri-clustering | SUGAR-TC · NUS-WIDE-128 computational cost | clustering time (h): 4.3 -> 2.1 | Improved |
| vary number of image centers C_i in {10..500} (ablation) | SUGAR-TC · NUS-WIDE-128 image retagging | F-score: best at C_i=40 | Improved |
| vary number of user clusters C_u in {3..200} (ablation) | SUGAR-TC · NUS-WIDE-128 image retagging | F-score: best at C_u=12; stable in [6, 18] | No clear effect |
| vary tag-image vs tag-user weight gamma from 0 to 1 (ablation) | SUGAR-TC · NUS-WIDE-128 image retagging | F-score: best at gamma=0.8 | Improved |
| add graph regularization (lambda1=0.1, lambda2=0.05) vs 0 (ablation) | SUGAR-TC · NUS-WIDE-128 image retagging | F-score: best at lambda1=0.1, lambda2=0.05 | Improved |
| use SUGAR-TC tag ranking for tag-based retrieval | SUGAR-TC · NUS-WIDE-128 tag-based retrieval | MAP | Improved |
Closest papers by meaning
- Tri-Clustered Tensor Completion for Social-Aware Image Tag Refinement (TPAMI 2016 · similarity 0.93)
- Kill Two Birds With One Stone: Weakly-Supervised Neural Network for Image Annotation and Tag Refinement (AAAI 2018 · similarity 0.86)
- Weakly-Supervised Deep Nonnegative Low-Rank Model for Social Image Tag Refinement and Assignment (AAAI 2017 · similarity 0.85)
- Tensorized Label Learning on Anchor Graph (AAAI 2024 · similarity 0.85)
- Fast Incomplete Multi-view Clustering with Adaptive Similarity Completion and Reconstruction (AAAI 2025 · similarity 0.84)
- GNNs Getting ComFy: Community and Feature Similarity Guided Rewiring (ICLR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
