Text classification with

What is this?

It is text classification model, a Convolutional Neural Network has been trained on 1.4M Amazon reviews, belonging to 7 categories, to predict what the category of a product is based solely on its reviews. You can try it live above, type your own review for an hypothetical product and check the results, or pick a random review.

How does it work?

• The characters of each review are one-hot encoded, and the review is represented as a matrix. This review is passed through 1D (or temporal) convolutions and pooling layers all the way to a classification step where the review is classified as one of the N categories

Source: Character-level Convolutional Networks for Text Classification

• Check this video series that explains what convolutions are and how to apply them to Natural Language Processing:

Resources

You can find the code to train the same model to perform text classification with convolutional neural networks in this interactive notebook

Apache MXNet official website

Apache MXNet tutorials

Learning Deep Learning with Apache MXNet/Gluon through interactive notebooks/textbook: MXNet, The straight dope

Serve deep learning models through an HTTP API using MXNet Model Server and AWS Elastic Container Service

Earlier implementation of crepe using the Apache MXNet/Module API (github, demo, blog post)

References

• Paper: Xiang Zhang, Junbo Zhao, Yann LeCun. Character-level Convolutional Networks for Text Classification . Advances in Neural Information Processing Systems 28 (NIPS 2015)
• Icons credits:
  • Icons made by Vectors Market from www.flaticon.com is licensed by CC 3.0 BY
  • Icons made by Freepik from www.flaticon.com is licensed by CC 3.0 BY
• Amazon Review Dataset:
  • Image-based recommendations on styles and substitutes. J. McAuley, C. Targett, J. Shi, A. van den Hengel. SIGIR, 2015
  • Inferring networks of substitutable and complementary products. J. McAuley, R. Pandey, J. Leskovec. Knowledge Discovery and Data Mining, 2015