4 ms·

Surprising there is no mention of any CNN based methods for NLP. A different approach that seems to provide results on par with the state of the art in some NLP

Surprising there is no mention of any CNN based methods for NLP. A different approach that seems to provide results on par with the state of the art in some NLP tasks [1]. [1] http://arxiv.org/abs/1408.5882 http://arxiv.org/abs/1408.5882

chejazi 11y ago
Since it's on GitHub, you could fork the repo, add any resources, and submit a merge request! Or create an issue that brings the omission to everyone's attention.
andrewtbham 11y ago
It is strange because CNNs are so popular in image processing. Are there anymore papers besides that one?
p1esk 11y ago
Here's one by Lecun: http://arxiv.org/abs/1502.01710 http://arxiv.org/abs/1502.01710
tianlins 11y ago
I think there are two major reasons why CNN might not be a good model. The first is that CNNs expect translation invariance, which is pretty common in images. But language sentences do not have this structure. Another reason is that in NLP outputs usually have varying length, this is why RNNs, LSTMs are so popular these days.
- juxtaposicion 11y ago
  I think you're right when thinking about CNNs on words. It's max-pool that's usually combined with CNNs that helps with translational invariance, less so the CNN filters themselves (which, if it were a full convolution, would show up in the complex phase). I think it makes more sense when CNNs are applied at the character level. The filter banks then activate for specific n-gram patterns of characters, like certain prefixes, suffixes, and root words. The higher level LSTMs are then relieved of having to understand that level of structure. Also, tokenization is hard, and might be especially wrong for media with grammatical abuse like Twitter, and this avoids that janky preprocessing. See: http://arxiv.org/abs/1508.06615 http://arxiv.org/abs/1508.06615
- chebyshev3 11y ago
  CNNs are used in NLP all the time for a range of problems. see Collobert and Weston's work, NLP (almost) from scratch. even in images, you're able to zero pad the input.

chejazi 11y ago

Since it's on GitHub, you could fork the repo, add any resources, and submit a merge request! Or create an issue that brings the omission to everyone's attention.

andrewtbham 11y ago

It is strange because CNNs are so popular in image processing. Are there anymore papers besides that one?

p1esk 11y ago

Here's one by Lecun: http://arxiv.org/abs/1502.01710 http://arxiv.org/abs/1502.01710

tianlins 11y ago

I think there are two major reasons why CNN might not be a good model. The first is that CNNs expect translation invariance, which is pretty common in images. But language sentences do not have this structure. Another reason is that in NLP outputs usually have varying length, this is why RNNs, LSTMs are so popular these days.

juxtaposicion 11y ago
I think you're right when thinking about CNNs on words. It's max-pool that's usually combined with CNNs that helps with translational invariance, less so the CNN filters themselves (which, if it were a full convolution, would show up in the complex phase). I think it makes more sense when CNNs are applied at the character level. The filter banks then activate for specific n-gram patterns of characters, like certain prefixes, suffixes, and root words. The higher level LSTMs are then relieved of having to understand that level of structure. Also, tokenization is hard, and might be especially wrong for media with grammatical abuse like Twitter, and this avoids that janky preprocessing. See: http://arxiv.org/abs/1508.06615 http://arxiv.org/abs/1508.06615
chebyshev3 11y ago
CNNs are used in NLP all the time for a range of problems. see Collobert and Weston's work, NLP (almost) from scratch. even in images, you're able to zero pad the input.