Movie Dialog classification

Cornell Movie Dialog classification with convolution layers.

It is slightly simplified implementation of Kim's Convolutional Neural Networks for Sentence Classification paper in Tensorflow.

Usage

• To (Download and) preprocess data ./data.py
• To create model graph, and test data flow ./model.py
• To train (existing or new) model ./train.py
• To get actual options use -h or --help flag after ./script.py
• Dependency: (((data) model) train)

Requirements

• Python 3
• Tensorflow > 0.12
• Numpy
• Possibly [update][update] required for Google Protobuf library if TFRecord is slow [update]: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/g3doc/get_started/os_setup.md#protobuf-library-related-issues

TensorBoard snapshots

Loss and accuracy

Graph overall view

Classifier

Convolution

Embedding

Reader

References

• Yoon Kim's github repo
• Denny's blog post on text classification
• Suriya Deepan blog post on preprocessing Cornell Data
• A Sensitivity Analysis of (and Practitioners' Guide to) Convolutional Neural Networks for Sentence Classification
• Indico blog post on Protobufs and Queues