klaijan / 11711_f22_anlp_hw3_seq2emo

Seq2Emo: A Sequence to Multi-Label Emotion Classification

This repo presents the script to replicate the results of Seq2Emo, CC, BR, BR-att as described in the paper.

Requirements

The dependent packages are listed in the requirements.txt. Note that a cuda device is required.

We have included the data and preprocessing script under data folder.

We use GloVe 840B 300d pretrain embedding, you can extract and put the txt file under data folder. or set the python argument --glove_path to point at it.
You also need to use the argument --download_elmo to download the ELMo embedding for first time of running the code.

Training/evaluation

For Seq2Emo, you can get the classification result of SemEval'18 dataset by the following script.

python trainer_lstm_seq2emo.py --dataset sem18 --batch_size 16 --glove_path data/glove.840B.300d.txt --download_emo --seed 0 

Note that the results reported in the paper are based 5 runs with --seed set to 1, 2, 3, 4, 5 respectively.

To change the dataset to GoEmotions, you can specify the --dataset option as goemotions. In the paper, we set the batch size to 32 for GoEmotions dataset to accelerate training.

Similarly, you can generate the results for CC by the following:

python trainer_lstm_cc.py --dataset sem18 --batch_size 16 

For BR, there are two variants: with self-attention and w/o self-attention. You can specify the model by setting the --attention option to self or None.

python trainer_lstm_binary.py --dataset sem18 --batch_size 16 --attention [self|None]