dataset

changemyview:https://chenhaot.com/data/cmv/cmv.tar.bz2 ADU recognition, dependency structure parsing: please contact Katsuhide Fujita, [email protected]

BIO parsing

https://github.com/seasa2016/arg_parsing_bert

preprocess (stage 1)

python preprocess_parsing_full.py train_period_data.jsonlist ./preprocess_data/train/ python preprocess_parsing_full.py heldout_period_data.jsonlist ./preprocess_data/heldout/

do training for op pos neg

python train.py --model_name_or_path bert-base-uncased --data_dir {PATH_TO_DATA} --do_train --do_eval --learning_rate 5e-5 --num_train_epochs 3.0 --output_dir {MODEL_FOLDER}

do testing for op pos neg

python train.py --bert_model {PATH_TO_CHECKPOINT_FOLDER} --data {PATH_TO_DATA} --output_dir {MODEL_FOLDER} --pred_name {PREDICT_PATH} --do_test

dependency structure parsing

https://github.com/seasa2016/span_pytorch_para

preprocess

python preprocess_tree.py ./mapping/train/mapping ./mapping/train/mapping ./preprocess_data/train/ python preprocess_tree.py "predict file" "mapping file" ./preprocess_data/heldout/

Trainging

python train.py --use-elmo 1 --data-path ${PATH_TO_DATA} --elmo-path ${PATH_TO_ELMO_EMBEDDING} --optimizer Adam --lr 0.003 --ac-type-alpha 0.25 --link-type-alpha 0.25 --batchsize 16 --epoch 64 --dropout 0.5 --dropout-lstm 0.1 --lstm-ac --lstm-shell --lstm-ac-shell --lstm-type --elmo-layers avg --train --dev --test --save-dir ./saved_model/

Testing

python test.py --test-data ${PATH_TO_DATA} --test-elmo ${PATH_TO_ELMO_EMBEDDING} --result-path ${OUTPUT_PATH} --save-model ${CHECK_POINT} --lstm-ac --lstm-shell --lstm-ac-shell --lstm-type --use-elmo 1 --elmo-layers avg

persuasiveness prediction

preprocess

allennlp elmo cmv_elmo_pos.txt cmv_elmo_pos.hdf5 --all --cuda-device 0 allennlp elmo cmv_elmo_neg.txt cmv_elmo_neg.hdf5 --all --cuda-device 0

python preprocess_persuasive.py ./preprocess_data/train/ ./../../span_pytorch_para/train 2 python preprocess_persuasive.py ./preprocess_data/heldout/ ./../../span_pytorch_para/heldout 1

run

python train.py --data-path {PATH_TO_DATA} --pair-path {PATH_TO_PAIR_INFO} --use-elmo 1 --elmo-layers avg --elmo-path {PATH_TO_ELMO_EMBEDDING} --optimizer Adam --lr 0.0003 --batchsize 2 --accumulate 8 --epoch 12 --dropout 0.3 --nhid 512 --dropout-lstm 0.1 --dropout-word 0.7 --extractor pool --criterion bce --nheads 4 --graph_layers 2 --ngraph 2 --graph GAT --rnn LSTM --lstm-ac-shell --para_decoder --grad_clip 10 --lr_gamma 0.8 --direct left --top 3 --final all --weight_decay 1e-5 --save-path ./qq34_rgat/