SemEval 2019 Task 3 - Emotion Detection in Textual Conversations

Dependencies

• Python 3.5+
• Numpy 1.15+
• PyTorch 1.0+
• torchtext 0.3.1+
• Spacy 2.0+

Model

• Recurrent Convolutional Neural Network (RCNN) (Lai 2015, AAAI)

External Word Representations

• Word embedding from datastories: https://github.com/cbaziotis/datastories-semeval2017-task4
• Word emmedding from trained sentiment classifier using CNN (Kim 2014, EMNLP), follow https://github.com/cbaziotis/datastories-semeval2017-task4
• ELMo: https://github.com/allenai/allennlp/blob/master/tutorials/how_to/elmo.md
• BERT: https://github.com/huggingface/pytorch-pretrained-BERT/blob/master/examples/extract_features.py

External Sentence Representations:

• DeepMoji: https://github.com/huggingface/torchMoji/blob/master/examples/encode_texts.py
• InferSent: https://github.com/facebookresearch/InferSent/blob/master/encoder/demo.ipynb

Run:

• Set your hyper-parameters in train.json, each key can have multiple values for grid search
• All word and sentence embeddings are in numpy array of shape (num_examples, embedding_size)
• Run python train.py --config train.json

Credit:

If you find the code or paper useful, please cite:

@article{zhong2019ntuer, title={ntuer at SemEval-2019 Task 3: Emotion Classification with Word and Sentence Representations in RCNN}, author={Zhong, Peixiang and Miao, Chunyan}, journal={arXiv preprint arXiv:1902.07867}, year={2019} }