documentations

• the paper
• the interim report
• the pitch
• experiment log
• post-related results

quick start

dependencies

• python3
• tensorflow
• sentencepiece
• nltk
• scikit-learn
• pandas
• matplotlib
• tqdm

datasets

mkdir data

download and unzip the following datasets into data

• reason identification and classification for evaluation
• claim sentences search for sentence modeling (use dirname ibm_claim)
• internet argument corpus v1.1 for post modeling (use dirname iac_v1.1)

to reproduce our results

mkdir trial trial/data trial/ckpt
cd src

prepare the data

run ./data_ibm.py or ./data_iac.py

train a new model

./train.py

check ./train.py --help for arguments and see config.json for more configurations.

evaluation and exploration

this part is messy due to the experimental nature of this project, but basically:

• run eval_prep_data_*.py to prepare the evaluation data
• run eval_embed*.py to embed the prepared data using the trained model
• run eval_classification*.py for classification and eval_clustering*.py for clustering
• there are also a few scripts named explore*.py for explorations

default paths, git ignored

• data for storing raw data
• trial/data for storing processed data
• trial/ckpt for model checkpoints
• trial/log for tensorboard summaries