Sydney Deep Learning Meetup - CNN Text Classification
?
1 like?411 views
This talk features a simplified implementation of Kim Yoon's paper on convolutional neural networks for sentence classification. Keras models are configuration driven so that with simple changes of a json file, one can vary the dimensions of the convolutional layers between embedding and fully connected.
Sydney Deep Learning Meetup - CNN Text Classification
- 1. Text Classification Using Convolutional Network
- 2. https://au.linkedin.com/in/andyhuangyh - Head of Artificial Intelligence @ Servian - Promotes Open Source - Organise Apache Spark Meetups - Apache Spark Instructor - Not a pilot yet Who Am I
- 3. Since 2008 7 Locations 200+ People 50+ Clients Currently Hiring in Sydney, Melbourne, Canberra, Brisbane, Bangalore About Servian
- 4. - Text Classification - Why - Convolutional Neural Network (CNN) Overview - Using CNN with Text - Pros and Cons About This Talk
- 5. - Sentiment Analysis (Better understanding of market/social reaction, brand image etc) - Spam/Phishing Filters (You see this all the time) - Trading & Investment News Classification - Customer Service Augmentation (Enquiry and Complaints etc.) - Recruitment - Insurance Risk Rating And many more ¡... Text Classification Use Cases
- 6. - Well known for its computer vision capability (image & video) - LeNet-5 - ResNet, VGG, Inception ... - Audio processing - WaveNet - Natural Language Processing - Sentence classification and modelling - Search query retrieval & ranking optimisation - NLP from Scratch Convolutional Neural Network - Overview
- 7. All models are wrong - George Box
- 8. CNN Overview - Convolution Operation https://en.wikipedia.org/wiki/Convolution
- 9. CNN Overview - Filters http://deeplearning.stanford.edu/wiki/images/6/6c/Convolution_schematic.gif 3x3 Filter Stride = 1
- 10. CNN Overview - Pooling http://cs231n.github.io/assets/cnn/maxpool.jpeg
- 11. CNN Overview - Image Recognition Networks http://cs231n.github.io/assets/cnn/convnet.jpeg
- 12. CNN Overview - Text Classification Kim, Yoon. (2014). Convolutional Neural Networks for Sentence Classification
- 13. Demo Text CNN Input Embedding Convolution Pooling Dropout Convolution Pooling Dropout Concatenate Flatten Fully Connected # Layers Softmax Convolution Pooling Dropout Convolution Pooling Dropout Flatten Convolution Pooling Dropout Convolution Pooling Dropout Flatten ¡... ... ... ... Variable Depth Feature Extraction Layers VariableWidthFeatureExtractionLayers https://github.com/andyhyh/easytext
- 15. Demo