Recent trends suggest that neuralnetwork-inspired word embedding models outperform traditional count-based distributional models on word similarity and analogy detection tasks. We reveal that much of the performance gains of word embeddings are due to certain system design choices and hyperparameter optimizations, rather than the embedding algorithms themselves. Furthermore, we show that these modifications can be transferred to traditional distributional models, yielding similar gains. In contrast to prior reports, we observe mostly local or insignificant performance differences between the methods, with no global advantage to any single approach over the others.
Transactions of the Association for Computational Linguistics
pages
211--225
volume
3
language
en
file
Levy et al - Improving Distributional Similarity with Lessons Learned from Word Embeddings.pdf:C\:\\Users\\Admin\\Documents\\Research\\_Paperbase\\Word Embeddings\Łevy et al - Improving Distributional Similarity with Lessons Learned from Word Embeddings.pdf:application/pdf
%0 Journal Article
%1 levy_improving_2015
%A Levy, Omer
%A Goldberg, Yoav
%A Dagan, Ido
%D 2015
%J Transactions of the Association for Computational Linguistics
%K imported
%P 211--225
%R 10.1162/tacl_a_00134
%T Improving Distributional Similarity with Lessons Learned from Word Embeddings
%U https://www.mitpressjournals.org/doi/abs/10.1162/tacl_a_00134
%V 3
%X Recent trends suggest that neuralnetwork-inspired word embedding models outperform traditional count-based distributional models on word similarity and analogy detection tasks. We reveal that much of the performance gains of word embeddings are due to certain system design choices and hyperparameter optimizations, rather than the embedding algorithms themselves. Furthermore, we show that these modifications can be transferred to traditional distributional models, yielding similar gains. In contrast to prior reports, we observe mostly local or insignificant performance differences between the methods, with no global advantage to any single approach over the others.
@article{levy_improving_2015,
abstract = {Recent trends suggest that neuralnetwork-inspired word embedding models outperform traditional count-based distributional models on word similarity and analogy detection tasks. We reveal that much of the performance gains of word embeddings are due to certain system design choices and hyperparameter optimizations, rather than the embedding algorithms themselves. Furthermore, we show that these modifications can be transferred to traditional distributional models, yielding similar gains. In contrast to prior reports, we observe mostly local or insignificant performance differences between the methods, with no global advantage to any single approach over the others.},
added-at = {2020-02-21T16:09:44.000+0100},
author = {Levy, Omer and Goldberg, Yoav and Dagan, Ido},
biburl = {https://www.bibsonomy.org/bibtex/227f41754adc4077be035bc2381f8bdf6/tschumacher},
doi = {10.1162/tacl_a_00134},
file = {Levy et al - Improving Distributional Similarity with Lessons Learned from Word Embeddings.pdf:C\:\\Users\\Admin\\Documents\\Research\\_Paperbase\\Word Embeddings\\Levy et al - Improving Distributional Similarity with Lessons Learned from Word Embeddings.pdf:application/pdf},
interhash = {a0fc3aa54b75452d40774fc4835f8be5},
intrahash = {27f41754adc4077be035bc2381f8bdf6},
issn = {2307-387X},
journal = {Transactions of the Association for Computational Linguistics},
keywords = {imported},
language = {en},
month = dec,
pages = {211--225},
timestamp = {2020-02-21T16:09:44.000+0100},
title = {Improving {Distributional} {Similarity} with {Lessons} {Learned} from {Word} {Embeddings}},
url = {https://www.mitpressjournals.org/doi/abs/10.1162/tacl_a_00134},
urldate = {2019-12-11},
volume = 3,
year = 2015
}