# Spherical Text Embedding

**Authors:** Yu Meng, Jiaxin Huang, Guangyuan Wang, Chao Zhang, Honglei Zhuang, Lance Kaplan, Jiawei Han  
**arXiv:** [1911.01196](https://arxiv.org/abs/1911.01196)  
**Date:** 2019-11-04  
**Summary page:** https://www.summarizepaper.com/en/paper/1911.01196v1/

## Key points

- The paper proposes a novel approach to unsupervised text embedding that addresses the gap between training and usage stages of text embedding.
- Traditional text embeddings are learned in Euclidean space, but directional similarity is often more effective for tasks such as word similarity and document clustering.
- The authors introduce a spherical generative model that jointly learns unsupervised word and paragraph embeddings.
- The proposed model optimizes text embeddings in the spherical space using an efficient optimization algorithm based on Riemannian optimization with convergence guarantee.
- The authors demonstrate that their model achieves state-of-the-art performance on various text embedding tasks including word similarity and document clustering.
- This approach has significant implications for natural language processing (NLP) tasks where accurate representation of textual data is essential for downstream applications such as sentiment analysis, machine translation, and information retrieval.
- The authors provide code for their implementation on GitHub which makes it accessible to researchers and practitioners alike.
- Overall, this paper presents a promising direction for future research in NLP by addressing a fundamental challenge in unsupervised text embedding.

## Summary

The paper "Spherical Text Embedding" proposes a novel approach to unsupervised text embedding that addresses the gap between the training and usage stages of text embedding. While traditional text embeddings are learned in Euclidean space, directional similarity is often more effective for tasks such as word similarity and document clustering. To overcome this limitation, the authors introduce a spherical generative model that jointly learns unsupervised word and paragraph embeddings. The proposed model optimizes text embeddings in the spherical space using an efficient optimization algorithm based on Riemannian optimization with convergence guarantee. The authors demonstrate that their model achieves state-of-the-art performance on various text embedding tasks including word similarity and document clustering. The proposed approach has significant implications for natural language processing (NLP) tasks, where accurate representation of textual data is essential for downstream applications such as sentiment analysis, machine translation, and information retrieval. Furthermore, the authors provide code for their implementation on GitHub which makes it accessible to researchers and practitioners alike. Overall, this paper presents a promising direction for future research in NLP by addressing a fundamental challenge in unsupervised text embedding. It offers an effective solution to improve accuracy of textual data representation which can be beneficial for various NLP applications.