this projects attempts to combine:

• graph embedding
• ConvNet

for the purpose of multi-label text classification.

I compared three methods on stackexchange datasets, where the goal is to predict the tags of posts.

If you wan to know more, here are some slides

• scripts/preprocessing_pipeline.sh: all the preprocessing, data splitting, feature extractio, etc
• sample_random_walks.py: sample random walks on a graph
• extract_embedding_labels.py: extract labels for embedding visualization

• fastxml_experiment.py: experiment for fastxml
• kim_cnn_experiment.py: experiment for cnn
• combined_model_experiment.py: experiment for cnn + deepwalk