Text Classifying With A Convolutional Neural Network

python tensorflow juptyer

Easy to follow text classifying implementation using a Conv. Neural Network (Tensorflow)

What You Need

1. Python 3.4 >=
2. Tensorflow >= 0.8
3. Jupyter Notebook or IPython Notebook [Deprecated]

Training The Network

Training the network with Tensorflow CPU-only (2013 i7 3770K) took about 70 mins. Note: If you try train in Jupyter/Ipython Notebook remove all print statements in the training loop. These print statement will lock up your browser and slow things down a lot.

To train the network on the Movie Review dataset just run:

./train.py

Training the network on the Customer Product Review data is

./train.py --dataset_option="products"

Evaluating The Network

You first have to train the network so see above.

To evaluate the network we need to pass in some arguments to the evaluate.py. Run this command, replacing checkpoint_dir with the last savepoint of the trained model.

python evaluate.py --checkpoint_dir="runs/*some numbers*/checkpoints/"

or for product review dataset_option

python evaluate.py --checkpoint_dir="runs_product/*some numbers*/checkpoints/"

Playing With The Network

You first have to train the network so see above.

To test out individual sentences on the network, open experiment.ipynb and follow instruction within the notebook.

python experiment.py --checkpoint_dir="runs/*some numbers*/checkpoints/" --sent="This is the sentence you want to test."

Credit To

Yoon Kim, Hu and Liu - KDD-2004, Denny Britz

Citations

Minqing Hu and Bing Liu. "Mining and Summarizing Customer Reviews." Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2004), Aug 22-25, 2004, Seattle, Washington, USA,