# Efficient Estimation of Word Representations in Vector Space

**Authors:** Tomas Mikolov, Kai Chen, Greg Corrado, Jeffrey Dean  
**arXiv:** [1301.3781](https://arxiv.org/abs/1301.3781)  
**Date:** 2013-01-16  
**Summary page:** https://www.summarizepaper.com/en/paper/1301.3781v1/

## Key points

- Paper introduces two novel model architectures for generating continuous vector representations of words
- Representations evaluated in a word similarity task and compared to existing techniques
- Significant improvements in accuracy while reducing computational costs
- High-quality 300-dimensional vectors produced for a vocabulary of one million words from a dataset containing 1.6 billion words in less than a day using just one CPU
- Generated vectors outperform other methods on the authors' test set for measuring different types of word similarities
- Authors plan to make the test set available for use by the wider research community
- Research submitted to ICLR 2013 highlights potential of efficient and effective word representation learning methods for natural language processing tasks

## Summary

The paper "Efficient Estimation of Word Representations in Vector Space" by authors Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean introduces two novel model architectures for generating continuous vector representations of words from large data sets. These representations are evaluated in a word similarity task and compared to existing techniques based on various neural networks. The results show significant improvements in accuracy while reducing computational costs. The authors were able to produce high-quality 300-dimensional vectors for a vocabulary of one million words from a dataset containing 1.6 billion words in less than a day using just one CPU. Furthermore, the generated vectors outperform other methods on the authors' test set for measuring different types of word similarities. They also plan to make this test set available for use by the wider research community. This research was submitted to ICLR 2013 and highlights the potential of efficient and effective word representation learning methods that can benefit natural language processing tasks.