azpoliak / eco

Code and data related to "Efficient, Compositional, Order-Sensitive n-gram Embeddings" (EACL 2017)

Efficient, Compositional, Order-Sensitive n-gram Embeddings

A suite for creating & evaluating phrasal embeddings via the ECO model based on Efficient, Compositional, Order-Sensitive n-gram Embeddings (EACL 2017).

Data:

The Skip-Embeddings and English Wikipedia used to generate the skip-embeddings can be downloaded here.

Directories:

1. evaluations: data and scripts for different evaluation tasks to evaluate the embeddings.
2. skipEmbeds: the script used to generate the ECO Skip-Embeddings and vanilla word2vec embeddings. ⋅⋅1. We extended Debora Sujono's python version of word2vec. ⋅⋅2. We also have a local C version that is not tested. ⋅⋅3. The embeddings used in the paper and released were created using the python version.

Citation:

If you use our changes to the code or our skip-embeddings, please cite us:

@inproceedings{Poliak:2017EACL,
Title = {Efficient, Compositional, Order-sensitive n-gram Embeddings},
 Author = {Poliak, Adam and Rastogi, Pushpendre and Martin, M. Patrick and Van Durme, Benjamin},
 booktitle = {Proceedings of the 15th Conference of the European Chapter of the 
 Association for Computational Linguistics},
 Year = {2017},
 Publisher = {Association for Computational Linguistics},
 location = {Valencia, Spain}
}

Errata:

There is a typo in equations (6) and (7) in the EACL proceedings. The version found at https://www.cs.jhu.edu/~apoliak1/papers/ECO--EACL-2017.pdf has the correct equations.