Research map / Label-efficient and robust learning / Recognition and long-tail
Spherical Text Embedding
Yu Meng, Jiaxin Huang, Guangyuan Wang, Chao Zhang, Honglei Zhuang, Lance Kaplan, Jiawei Han
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 embeddings directly in spherical space rather than Euclidean space | JoSE word embeddings · Word similarity: WordSim353, MEN, SimLex999 | Spearman rank correlation: WordSim353: 0.711 -> 0.739; MEN: 0.726 -> 0.748; SimLex999: 0.321 -> 0.339 | Improved |
| Use JoSE paragraph embeddings instead of the strongest listed baseline | Document clustering with K-Means · 20 Newsgroup | MI: 1.856 ± 0.020 -> 1.975 ± 0.026 | Improved |
| Use JoSE paragraph embeddings instead of the strongest listed baseline | Document clustering with SK-Means · 20 Newsgroup | NMI: 0.630 ± 0.007 -> 0.664 ± 0.010 | Improved |
| Use JoSE paragraph embeddings instead of the strongest listed baseline | Document clustering with SK-Means · 20 Newsgroup | ARI: 0.494 ± 0.012 -> 0.568 ± 0.020 | Improved |
| Use JoSE paragraph embeddings instead of the strongest listed baseline | Document clustering with SK-Means · 20 Newsgroup | Purity: 0.648 ± 0.017 -> 0.721 ± 0.029 | Improved |
| Use spherical K-Means instead of K-Means | JoSE paragraph embeddings · 20 Newsgroup | NMI: 0.663 ± 0.008 -> 0.664 ± 0.010 | No clear effect |
| Use JoSE paragraph embeddings instead of Avg. W2V | 20 Newsgroup k-NN classification · 20 Newsgroup | Macro-F1: 0.630 -> 0.703 | Improved |
| Use JoSE paragraph embeddings instead of Avg. W2V | 20 Newsgroup k-NN classification · 20 Newsgroup | Micro-F1: 0.631 -> 0.707 | Improved |
| Use JoSE paragraph embeddings instead of Avg. W2V | Movie Review k-NN classification · Movie Review | Macro-F1: 0.712 -> 0.764 | Improved |
| Use JoSE paragraph embeddings instead of Avg. W2V | Movie Review k-NN classification · Movie Review | Micro-F1: 0.713 -> 0.765 | Improved |
| Train with JoSE rather than Word2Vec | Word embedding training · Wikipedia dump | Training time per iteration: 0.81 hrs -> 0.73 hrs | Improved |
Closest papers by meaning
- From Topology to Retrieval: Decoding Embedding Spaces with Unified Signatures (NeurIPS 2026 · similarity 0.84)
- Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe (NeurIPS 2024 · similarity 0.84)
- Harnessing the Universal Geometry of Embeddings (NeurIPS 2025 · similarity 0.84)
- HyperSDFusion: Bridging Hierarchical Structures in Language and Geometry for Enhanced 3D Text2Shape Generation (CVPR 2024 · similarity 0.82)
- Text Is MASS: Modeling as Stochastic Embedding for Text-Video Retrieval (CVPR 2024 · similarity 0.82)
- SPHERE-JEPA: Spherical Prediction with Homogeneous Embeddings (NeurIPS 2026 · similarity 0.82)
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.
