Research map / Label-efficient and robust learning / Recognition and long-tail
LightXML: Transformer with Dynamic Negative Sampling for High-Performance Extreme Multi-label Text Classification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 dynamic rather than static negative-label sampling (ablation) | LightXML · Wiki-500K | P@1: 75.30 -> 76.19; 73.30 -> 76.19 | Helped |
| use dynamic rather than static negative-label sampling (ablation) | LightXML · Amazon-670K | P@1: 46.27 -> 47.14; 43.36 -> 47.14 | Helped |
| add additional text representation to static negative sampling (ablation) | LightXML static-sampling variant · Wiki-500K | P@1: 73.30 -> 75.30 | Helped |
| add additional text representation to static negative sampling (ablation) | LightXML static-sampling variant · Amazon-670K | P@1: 43.36 -> 46.27 | Helped |
| use the last five transformer layers instead of only the last layer (ablation) | LightXML · Wiki-500K and Amazon-670K | P@5: more than 1% | Helped |
| use the last five transformer layers instead of only the last layer (ablation) | LightXML · Wiki-500K and Amazon-670K | training convergence: half of total epochs | Helped |
| use LightXML-1 rather than AttentionXML-1 | single-model XMC · Wiki-500K | P@5: 44.41 -> 44.12 | Hurt |
| use LightXML-1 rather than AttentionXML-1 | single-model XMC · Amazon-670K | P@5: 36.94 -> 38.23 | Helped |
| use LightXML-1 rather than AttentionXML-1 | single-model XMC · Wiki-500K | model size: 3.11 -> 1.47 | Helped |
| use LightXML-1 rather than AttentionXML-1 | single-model XMC · Amazon-670K | model size: 5.52 -> 1.53 | Helped |
| use LightXML-1 rather than AttentionXML-1 | single-model XMC · Wiki-500K | average prediction time (ms/sample): 4.20 -> 3.75 | Helped |
| use LightXML-1 rather than AttentionXML-1 | single-model XMC · Amazon-670K | average prediction time (ms/sample): 8.59 -> 4.09 | Helped |
Closest papers by meaning
- ELMO : Efficiency via Low-precision and Peak Memory Optimization in Large Output Spaces (ICML 2025 · similarity 0.85)
- Harnessing the Power of Beta Scoring in Deep Active Learning for Multi-Label Text Classification (AAAI 2024 · similarity 0.85)
- T-MDML: Triplet-based Multiple Distance Metric Learning for Multi-Instance Multi-Label Classification with Label Correlation (AAAI 2025 · similarity 0.84)
- Addressing Multi-Label Learning with Partial Labels: From Sample Selection to Label Selection (AAAI 2025 · similarity 0.84)
- Taming the Sigmoid Bottleneck: Provably Argmaxable Sparse Multi-Label Classification (AAAI 2024 · similarity 0.83)
- Enhancing Multi-Label Classification via Dynamic Label-Order Learning (AAAI 2024 · 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.