No description or website provided.
Python Shell
Latest commit d7524f6 Dec 9, 2016 @rspeer rspeer committed on GitHub Fix the `uris_main` link. Fixes #40.


The main location of this repository is on GitHub, which may contain updates:

The code and paper for the previous version of this system, also referred to as the 'ConceptNet Vector Ensemble', are available in a branch of this repository:


This is the data from an ensemble that combines ConceptNet, word2vec, and GloVe using a variation on retrofitting. It was previously known as the "ConceptNet Vector Ensemble".

ConceptNet Numberbatch consists of high-quality term vectors (that is, word embeddings) that can be used directly as a representation of word meanings or as a starting point for further machine learning.

Here is a paper we wrote about the previous version in early 2016.

The code and paper were created as a research project of Luminoso Technologies, Inc., by Rob Speer and Joshua Chin.

License and attribution

These vectors are distributed under the CC-By-SA 4.0 license. In short, if you distribute a transformed or modified version of these vectors, you must release them under a compatible Share-Alike license and give due credit to Luminoso.

Some suggested text:

This data contains semantic vectors from Conceptnet Numberbatch, by
Luminoso Technologies, Inc. You may redistribute or modify the
data under the terms of the CC-By-SA 4.0 license.

If you build on this data, you should cite it. We recommend citing the arXiV preprint for now:

Robert Speer and Joshua Chin. "An Ensemble Method to Produce High-Quality Word Embeddings." arXiv preprint arXiv:1604.01692 (2016).

In BibTeX form, the citation is:

  title={An Ensemble Method to Produce High-Quality Word Embeddings},
  author={Speer, Robert and Chin, Joshua},
  journal={arXiv preprint arXiv:1604.01692},

This data is itself built on:

  • ConceptNet 5.5, which contains data from Wiktionary, WordNet, and many contributors to Open Mind Common Sense projects, edited by Rob Speer

  • GloVe, by Jeffrey Pennington, Richard Socher, and Christopher Manning

  • word2vec, by Tomas Mikolov and Google Research

Downloading the term vectors

The *.txt.gz files of term vectors are too large to include in the GitHub repository for this package. You should follow the links provided here to download them.

Format of the term vectors

The term vectors are in the text format used by GloVe, which is similar to the format used by word2vec but without a header line.

Each line contains a term label followed by 300 floating-point numbers, separated by spaces:

/c/en/absolute_value -0.1410 -0.0641 -0.0618 0.2627 -0.0177 -0.2862 -0...
/c/en/absolute_zero 0.0074 -0.0272 -0.1510 0.0817 0.0648 -0.2144 -0.03...
/c/en/absoluteless 0.3960 -0.0599 0.0875 0.1853 -0.1459 -0.0827 0.0142...
/c/en/absolutely -0.0455 -0.1817 0.1297 0.0954 0.0150 0.0417 0.0772 0....
/c/en/absolutely_convergent 0.4119 0.1030 -0.0579 0.2609 -0.0179 -0.29...

To make this into word2vec format, add a header line with two numbers representing the height and width of the matrix. These dimensions are listed below.

There are four files available for download, providing different representations of the data:

If you have the ConceptNet database, the extra data should be redundant, but it provides a convenient way to expand the vocabulary without looking terms up in ConceptNet.

We have included here the code necessary to convert text into ConceptNet URIs, in