Research map / Learning theory and algorithms / Optimal transport
Re-evaluating Word Mover’s Distance
Ryoma Sato, Makoto Yamada, Hisashi Kashima
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 |
|---|---|---|---|
| L1-normalize BOW vectors before kNN classification | BOW · All eight datasets | relative kNN classification error: 3.931 -> 1.000 | Improved |
| Use optimal transport with word-embedding geometry after L1 normalization | WMD versus BOW (L1/L1) · All eight datasets | relative kNN classification error: 1.000 -> 0.917 | Improved |
| Use WMD instead of the strongest L1/L1 classical baseline | WMD · bbcsport, twitter, recipe, ohsumed, classic, reuters, amazon, and 20news | kNN classification error | Mixed |
| Use TF-IDF weighting for WMD marginals | WMD · Eight document-classification datasets | kNN classification error: approximately a six percent improvement from the normalized TF-IDF | Improved |
| Use TF-IDF weighting for WMD marginals | WMD · Clean datasets | kNN classification error: two percent improvement | Improved |
| Remove duplicate documents from the evaluation data | BOW, TF-IDF, and WMD · Eight document-classification datasets | kNN classification error | No clear effect |
| Include words outside the word2vec vocabulary in BOW and TF-IDF | BOW and TF-IDF · All-word datasets | kNN classification error | No clear effect |
| Use 5-dimensional instead of 300-dimensional embeddings (ablation) | WMD · Eight document-classification datasets | kNN classification error | Hurt |
| Vary embedding normalization and distance metric in WMD's cost function (ablation) | WMD · Eight document-classification datasets | kNN classification error | No clear effect |
| Use exponential distance weighting in weighted kNN instead of standard kNN voting | BOW, TF-IDF, WMD, and WMD-TF-IDF · Eight document-classification datasets | classification error | Improved |
Closest papers by meaning
- Supervised Word Mover's Distance (NeurIPS 2016 · similarity 0.91)
- Style-transfer and Paraphrase: Looking for a Sensible Semantic Similarity Metric (AAAI 2021 · similarity 0.83)
- DOCS: Quantifying Weight Similarity for Deeper Insights into Large Language Models (ICLR 2025 · similarity 0.81)
- UltraTWD: Optimizing Ultrametric Trees for Tree-Wasserstein Distance (ICML 2025 · similarity 0.80)
- Bayesian Test-Time Inference of Task-Aligned Similarity from Weak Interactive Feedback (NeurIPS 2026 · similarity 0.79)
- Exploiting Vocabulary Frequency Imbalance in Language Model Pre-training (NeurIPS 2025 · similarity 0.79)
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.
