Research map / Learning theory and algorithms / Prediction and decision losses
On the Dimensionality of Word Embedding
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 |
|---|---|---|---|
| Choose dimensionality by minimizing estimated PIP loss | PPMI LSA on Text8 · WordSim353 and MTurk771 | Agreement with empirical optimal dimensionality: 214; WS353 opt. 127; MT771 opt. 116 | Helped |
| Choose dimensionality by minimizing estimated PIP loss | Skip-gram Word2Vec on Text8 · WordSim353, MT771, Google analogy | Agreement with empirical optimal dimensionality: 129; 56, 102, 220 | Helped |
| Choose dimensionality by minimizing estimated PIP loss | GloVe on Text8 · WordSim353, MT771, Google analogy | Agreement with empirical optimal dimensionality: 719; 220, 860, 560 | Helped |
| Increase α in the factorization embedding | PPMI LSA · WordSim353 and MTurk771 | Robustness to over-parametrization | Helped |
| Use α=0.25 instead of α=0 (ablation) | PPMI LSA · Text8 word relatedness dimensionality selection | PIP loss-minimizing dimensionality: 214 -> 138 | Mixed |
| Use α=0.5 instead of α=0 (ablation) | PPMI LSA · Text8 word relatedness dimensionality selection | PIP loss-minimizing dimensionality: 214 -> 108 | Mixed |
| Use α=0.75 instead of α=0 (ablation) | PPMI LSA · Text8 word relatedness dimensionality selection | PIP loss-minimizing dimensionality: 214 -> 90 | Mixed |
| Use α=1 instead of α=0 (ablation) | PPMI LSA · Text8 word relatedness dimensionality selection | PIP loss-minimizing dimensionality: 214 -> 82 | Mixed |
| Increase embedding dimensionality substantially | Skip-gram Word2Vec and GloVe · WordSim353, MTurk771, and analogy tests | Functionality-test performance under over-parametrization: 80% to 90%; k=10000 | No clear effect |
Closest papers by meaning
- Invariance and identifiability issues for word embeddings (NeurIPS 2019 · similarity 0.89)
- Fast PMI-Based Word Embedding with Efficient Use of Unobserved Patterns (AAAI 2019 · similarity 0.88)
- Imposing Category Trees Onto Word-Embeddings Using A Geometric Construction (ICLR 2019 · similarity 0.88)
- By Tying Embeddings You Are Assuming the Distributional Hypothesis (ICML 2024 · similarity 0.87)
- On the Emergence of Linear Analogies in Word Embeddings (NeurIPS 2025 · similarity 0.86)
- Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe (NeurIPS 2024 · similarity 0.86)
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.