Prerequisites

• Download Glove
  • Twitter 50 dim: wget https://nlp.stanford.edu/data/glove.twitter.27B.zip
• Tweebo parser
  • Download the pretrained models from here

Prepare the inputs

Tokenize:

bash …/TweeboParser/tokenize_and_tag_only.sh <root_dir> …/TweeboParser/ark-tweet-nlp-0.3.2 <working directory> …/pretrained_models/tagging_model …/TweeboParser/scripts input.csv

Extract the argument:

python discourseParsing/ArgumentExtractor.py input.csv_tagger.out n

Format into inputs:

python discourseParsing/utils/InputMaker.py input.csv_tagger.out_args.csv input.csv_tagger.out_args_train_meta.csv

Training:

python train.py --tr_file input.csv_tagger.out_args.dict --tr_meta_file input.csv_tagger.out_args_train_meta.odict --cuda --input_dim 200 --hidden_dim 200 --word_embedding_dict …/glove/glove_200.dict --attn_act ReLU

• Saves the model

Prediction:

python predict.py --test_file test_file.csv_tagger.out_args.dict --cuda --input_dim 200 --hidden_dim 200 --word_embedding_dict …/glove/glove_200.dict --attn_act ReLU --model model.ptstdict --split_one_arg

• Split one argument into two if that’s the only argument in the whole message, otherwise the message will be ignored if it’s too short to get any discourse relations

Pretrained models and dataset

Please send an email to vvaradarajan@cs.stonybrook.edu to get access the social media treebank and pretrained models.

Citation

If you are using our data or models, please cite our work:

@inproceedings{son-etal-2022-discourse,
    title = "Discourse Relation Embeddings: Representing the Relations between Discourse Segments in Social Media",
    author = "Son, Youngseo  and
      Varadarajan, Vasudha  and
      Schwartz, H. Andrew",
    booktitle = "Proceedings of the Workshop on Unimodal and Multimodal Induction of Linguistic Structures (UM-IoS)",
    month = dec,
    year = "2022",
    address = "Abu Dhabi, United Arab Emirates (Hybrid)",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.umios-1.5",
    doi = "10.18653/v1/2022.umios-1.5",
    pages = "45--55",
  }