Research map / Learning theory and algorithms / Optimal transport
Hierarchical Optimal Transport for Document Representation
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 |
|---|---|---|---|
| Truncate LDA topic proportions before computing transport (ablation) | HOTT · k-NN classification across datasets | classification performance | No clear effect |
| Truncate topic proportions at varying levels (top 10, 20, 50, or 100 words) (ablation) | HOTT · Reuters | k-NN classification performance | No clear effect |
| Remove topic truncation (ablation) | HOTT · Reuters | k-NN classification performance | Mixed |
| Use trained word2vec embeddings instead of GloVe (ablation) | HOTT · Reuters | k-NN classification performance | No clear effect |
| Increase the number of LDA topics (ablation) | HOTT · Classic | k-NN classification performance | No clear effect |
| Truncate documents to their top 20 unique words (ablation) | WMD · k-NN classification across datasets | classification performance | Hurt |
| Use HOTT rather than RWMD for distance-based link prediction | Link Prediction TU · Φ0 through Φ4 | F1 Score: 71.60 -> 73.19; 74.90 -> 76.03; 75.20 -> 76.24; 77.16 -> 78.64; 82.92 -> 83.25 | Helped |
| Include the HOTT document distance rather than no distance feature | Link Prediction TU · Φ1 through Φ4 | F1 Score: 61.13 -> 76.03; 64.27 -> 76.24; 67.72 -> 78.64; 81.68 -> 83.25 | Helped |
| Use W2 instead of W1 (ablation) | HOTT · k-NN classification across datasets | classification performance | No clear effect |
| Apply stemming (ablation) | HOTT · k-NN classification across datasets | classification performance | No clear effect |
| Use W2 rather than W1 (ablation) | RWMD and WMD-T20 · k-NN classification across datasets | classification performance | Hurt |
Closest papers by meaning
- OTLDA: A Geometry-aware Optimal Transport Approach for Topic Modeling (NeurIPS 2020 · similarity 0.89)
- On the Affinity, Rationality, and Diversity of Hierarchical Topic Modeling (AAAI 2024 · similarity 0.83)
- Distribution-Driven Dense Retrieval: Modeling Many-to-One Query-Document Relationship (AAAI 2025 · similarity 0.83)
- Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe (NeurIPS 2024 · similarity 0.83)
- Multi-Level Optimal Transport for Universal Cross-Tokenizer Knowledge Distillation on Language Models (AAAI 2025 · similarity 0.83)
- Hierarchical Retrieval: The Geometry and a Pretrain-Finetune Recipe (NeurIPS 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.