Research map / Label-efficient and robust learning / Recognition and long-tail
StarSpace: Embed All The Things!
Ledell Wu, Adam Fisch, Sumit Chopra, Keith Adams, Antoine Bordes, Jason Weston
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 |
|---|---|---|---|
| Train directly on sentence pairs from the same document rather than at word level (ablation) | StarSpace · Wikipedia Sentence Matching (Task 2) | Hits@10: 16.41% -> 50.89% | Improved |
| Train on both word-level and sentence-pair objectives instead of sentence pairs alone (ablation) | StarSpace · Wikipedia Sentence Matching (Task 2) | Hits@10: 50.89% -> 45.21% | Hurt |
| Increase negative samples from k=50 to k=1000 under a fixed one-hour budget (ablation) | StarSpace, dim=50 · Freebase 15k | filtered Hits@10: 69.05% -> 54.14% | Hurt |
| Increase text-classification n-gram size from 2 to 5 (ablation) | StarSpace · Yelp15 | accuracy: 62.4 -> 65.3 | Mixed |
| Train sentence embeddings directly on sentence pairs instead of word-level context (ablation) | StarSpace · SentEval MR | accuracy: 73.8 -> 69.1 | Hurt |
| Train sentence embeddings directly on sentence pairs instead of word-level context (ablation) | StarSpace · SentEval STS14 | Pearson correlation: 0.65 -> 0.70 | Improved |
| Use word-plus-sentence multi-task training instead of sentence-pair training alone (ablation) | StarSpace · SentEval MR | accuracy: 69.1 -> 72.1 | No clear effect |
| Train the ranking model directly for the article-search task rather than use TFIDF | StarSpace · Wikipedia Article Search (Task 1) | Hits@10: 72.80% -> 78.14% | Improved |
| Train on the downstream sentence-matching task rather than use unsupervised fastText | StarSpace · Wikipedia Sentence Matching (Task 2) | Hits@10: 13.54% -> 50.89% | Improved |
| Use ngrams=5 rather than ngrams=2 (ablation) | StarSpace · Yelp15 | accuracy: 66.6 -> 65.3 | Hurt |
Closest papers by meaning
- Active Discriminative Text Representation Learning (AAAI 2017 · similarity 0.85)
- Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe (NeurIPS 2024 · similarity 0.85)
- Contextual Document Embeddings (ICLR 2025 · similarity 0.84)
- NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models (ICLR 2025 · similarity 0.83)
- From Topology to Retrieval: Decoding Embedding Spaces with Unified Signatures (NeurIPS 2026 · similarity 0.83)
- Some Like It Small: Czech Semantic Embedding Models for Industry Applications (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.
