Research map / Label-efficient and robust learning / Recognition and long-tail
Poincare Glove: Hyperbolic Word Embeddings
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 |
|---|---|---|---|
| Replace Euclidean GloVe with Poincaré GloVe using cosh² distance and the initialization trick | GloVe word embeddings · Word similarity benchmarks | Spearman correlation: 0.3798 -> 0.4187 (RareWord); 0.5901 -> 0.6209 (WordSim); 0.2963 -> 0.3208 (SimLex); 0.1675 -> 0.1915 (SimVerb); 0.6524 - | Helped |
| Use Poincaré GloVe with cosh² distance and the initialization trick | Word analogy embeddings · Google and MSR analogy benchmarks | accuracy: 0.6167 -> 0.6339 (Google); 0.4868 -> 0.4971 (MSR) | Helped |
| Use unsupervised Poincaré GloVe with x² distance | Hypernymy detection · WBLESS | accuracy: 0.575 -> 0.652 | Helped |
| Use unsupervised Poincaré GloVe with x² distance | Hypernymy detection · HyperLex | Spearman correlation: 0.284 -> 0.341 (x² model); 0.284 (cosh² model); baseline 0.345 | Mixed |
| Add 400 generic and 400 specific WordNet words to identify the Gaussian-variance direction (ablation) | 50x2D Poincaré GloVe, x² · HyperLex | Spearman correlation: 0.341 -> 0.421 | Helped |
| Add 400 generic and 400 specific WordNet words to identify the Gaussian-variance direction (ablation) | 50x2D Poincaré GloVe, x² · WBLESS | accuracy: 0.652 -> 0.790 | Helped |
| Replace x² distance with cosh² distance (ablation) | 50x2D Poincaré GloVe with unsupervised hypernymy scoring · WBLESS | accuracy: 0.652 -> 0.575 | Hurt |
| Initialize from a model trained on the restricted 50k vocabulary (ablation) | Poincaré GloVe · Word similarity benchmarks | Spearman correlation: RareWord 0.3996 -> 0.4187; WordSim 0.6486 -> 0.6209 | Mixed |
| Use Riemannian Adagrad instead of Riemannian SGD (ablation) | Poincaré GloVe optimization · Word embedding experiments | performance | Helped |
| Average target and context word vectors instead of using only the target vector (ablation) | Word embedding evaluation · Analogy benchmarks | accuracy: For 100D Poincaré GloVe, cosh², init: SemGoogle 0.6641 -> 0.6776; Google 0.6339 -> 0.6210 | Mixed |
Closest papers by meaning
- Representation Tradeoffs for Hyperbolic Embeddings (ICML 2018 · similarity 0.88)
- Learning Mixed-Curvature Representations in Product Spaces (ICLR 2019 · similarity 0.88)
- Numerically Accurate Hyperbolic Embeddings Using Tiling-Based Models (NeurIPS 2019 · similarity 0.88)
- Learning Continuous Hierarchies in the Lorentz Model of Hyperbolic Geometry (ICML 2018 · similarity 0.87)
- Poincaré Embeddings for Learning Hierarchical Representations (NeurIPS 2017 · similarity 0.87)
- Hyperbolic Embeddings of Supervised Models (NeurIPS 2024 · similarity 0.87)
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.