[CaCL] glove paper

[Marten van Schijndel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

That's a nice bit of post-hoc work. Lucky he already had the pipeline
lying around.

Marty

On 11/12/2014 12:13 AM, Michael White wrote:
> It turns out that Yoav Goldberg has a blog post and writeup saying
> that glove's performance is quite similar to word2vec in an
> apples-to-apples comparison:
> 
> https://plus.google.com/114479713299850783539/posts/BYvhAbgG8T2
> 
> On Wed, Oct 15, 2014 at 10:52 AM, Marten van Schijndel < 
> van-schijndel.1 at buckeyemail.osu.edu> wrote:
> 
> GloVe http://www-nlp.stanford.edu/projects/glove/glove.pdf
> 
> Abstract: Recent methods for learning vector space representations
> of words have succeeded in capturing fine-grained semantic and
> syntactic regularities using vector arithmetic, but the origin of
> these regularities has remained opaque. We analyze and make
> explicit the model properties needed for such regularities to
> emerge in word vectors. The result is a new global log-bilinear
> regression model that combines the advantages of the two major
> model families in the literature: global matrix factorization and
> local context window methods. Our model efficiently leverages
> statistical information by training only on the nonzero elements in
> a word-word co-occurrence matrix, rather than on the entire sparse
> matrix or on individual context windows in a large corpus. The
> model produces a vector space with meaningful substructure, as
> evidenced by its performance of 75% on a recent word analogy task.
> It also outperforms related models on similarity tasks and named
> entity recognition.
>> 
>> _______________________________________________ CaCL mailing
>> list CaCL at mail.ling.ohio-state.edu 
>> http://mail.ling.ohio-state.edu/cgi-bin/mailman/listinfo/cacl
>> 
> 
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
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=dwBz
-----END PGP SIGNATURE-----