i69086 / twitter-sentiment-cnn

An implementation in TensorFlow of a convolutional neural network (CNN) to perform sentiment classification on tweets.

Twitter sentiment classification by Daniele Grattarola

This is an implementation in TensorFlow of a convolutional neural network (CNN) to perform sentiment classification on tweets. Rather than a pre-packaged tool to simply perform classification, this is a model that needs to be trained and fine tuned by hand and has more of an educational value. The dataset used in this example is taken from here (someone reported to me that the link to the dataset appears to be dead sometimes, so dataset_downloader.py might not work. I successfully ran the script on September 10, 2016, but feel free to contact me if you have any problems).

NOTE: this script is for Python 2.7 only

Setup

You'll obviously need TensorFlow >=1.1.0 and its dependecies installed in order for the script to work (see here). Download the source files and cd into the folder:

# Clone the repo
git clone https://gitlab.com/danielegrattarola/twitter-sentiment-cnn.git
cd twitter-sentiment-cnn

Before being able to use the script, some setup is needed; download the dataset from the link above by running:

# Create the twitter-sentiment-dataset folder and download the dataset
python dataset_downloader.py

Read the dataset from the CSV into two files (.pos and .neg) with:

# Create tw-data.neg and tw-data.pos inside the twitter-sentiment-dataset folder
python csv_parser.py

And generate a CSV with the vocabulary (and its inverse mapping) with:

# Create vocab.csv and vocab_inv.csv inside the twitter-sentiment-dataset folder
# This is needed to sucessfully restore saved sessions with different hyperparameters
python vocab_builder.py

The files will be generated into the twitter-sentiment-dataset folder. After this initial setup, create an output folder which will contain all the data associated to different sessions of the network:

mkdir output

Now everything is set up and you're ready to start using the script.

Usage

The simplest way to run the script is:

python twitter-sentiment-cnn.py

which will load the dataset in memory, create the computation graph and quit. Try to run the script like this to see if everything is set up correctly. To run a training session on the full dataset (and save the result so that we can reuse the network later, or perform more training) run:

python twitter-sentiment-cnn.py --train --save

When the training in finished, we can test the network as follows:

# replace 'runYYYYMMDD-HHmmss' with the actual name of the run folder (not the path, just the name of the folder)
python twitter-sentiment-cnn.py --load path/to/ckpt/folder/ --custom_input 'I love neural networks!'

which will eventually output:

...
Processing custom input: I love neural networks!
Custom input evaluation: POS
Actual output: [ 0.19249919  0.80750078]
...

We can also fine tune the parameters to have different performance.
By running:

python twitter-sentiment-cnn.py -h

the script will output a list of all customizable session parameters. The parameters are:

• train: whether the network should perform training (default: False)
• save: save session checkpoints (default: False)
• save_protobuf: save pre-trained model as binary protobuf (default: False)
• evaluate_batch: print the network output on a batch from the dataset (for debugging/educational purposes)
• load: restore the given session if it exists (Pass the name of the session folder: runYYYMMDD-hhmmss)
• custom_input: the program will print the network output for the given input string.
• filter_sizes: comma-separated filter sizes for the convolution layer (default: '3,4,5')
• reduced_dataset: use a reduced dataset to improve memory usage (default: 1; uses all dataset)
• embedding_size: size of character embedding (default: 128)
• num_filters: number of filters per filter size (default: 128)
• batch_size: batch Size (default: 100)
• epochs: number of training epochs (default: 3)
• valid_freq: set how many times per epoch to perform validation testing (default: 1)
• checkpoint_freq: set how many times per epoch to save the model (default: 1)
• test_data_ratio: percentual of the dataset to be used for validation (default: 10)

Model description

The network implemented in this script is a single layer CNN structured as follows:

• Embedding layer: takes as input the tweets (as strings) and maps each word to an n-dimensional space so that it is represented as a sparse vector (see word2vec).
• Convolution layers: performs a convolution using an arbitrary number of filters of arbitrary sizes. Each layer is associated to a filter size and outputs to a single pooling layer (by defalut: 128 filters per filter size, sizes 3,4,5; this means that the network has 3 parallel conv+pool sections). Note that a filter size is the number of words that the filter covers.
• Pooling layers: pools the output of a convolution layer using max-pooling.
• Concat layer: concatenates the output of the different pooling layers into a single tensor.
• Dropout layer: performs neuron dropout (some random neurons are not considered during training).
• Output layer: fully connected layer that uses a softmax activation function to perform classification.

To have a better visualization of the network, first run a training session, then:

tensorboard --logdir output/runYYYYMMDD-HHmmss/summaries/

and then navigate to localhost:6006 from your browser (you'll see the TF computation graph in the Graph section).