CNN for Text Classification

7
CNN for Text Classification Tim Lee and Will Kelly

Transcript of CNN for Text Classification

Page 1: CNN for Text Classification

CNN for Text ClassificationTim Lee and Will Kelly

Page 2: CNN for Text Classification

What are Convolutions?

A convolution is an integral that

expresses the amount of overlap of one function as it is shifted over another

function

Can be thought as "blending" functions

Pictures found on Christopher Olah's blog, originally from Wikipedia 2

Definition from Wolfram Alpha's page on convolution 1

Page 3: CNN for Text Classification

How does this work?

Computer Vision NLP

Word2VecGloVeOne-hot

Images via WildML 3

Page 4: CNN for Text Classification

Convolution Layers

Picture found on Christopher Olah's blog 4

Page 5: CNN for Text Classification

More complexity & Max-pooling

Picture found on Christopher Olah's blog 4

Image via WildML 3

Page 6: CNN for Text Classification

The Model

Adapted from Kim Yoon's 2014 paper, Convolutional Neural Networks for Sentence Classification 5

Page 7: CNN for Text Classification

References

1. http://mathworld.wolfram.com/Convolution.html 2.

https://en.wikipedia.org/wiki/Convolution#/media/File:Convolution_of_box_signal_with_itself2.gif

3. http://www.wildml.com/2015/11/understanding-convolutional-neural-networks-for-nlp/

4. http://colah.github.io/posts/2014-07-Conv-Nets-Modular/

5. http://www.wildml.com/2015/12/implementing-a-cnn-for-text-classification-in-tensorflow/