Dialogue Act Classification

Implementation and comparison of several solutions for Dialogue Act Classification.

Dataset

swda GitHub repo is used to obtain the dataset.

Model	Accuracy, %
Tf-Idf + LightGBM without context	63.89
Fasttext + LightGBM without context	66.57
Pretrained Bert + LightGBM without context	66.61
Fasttext + Hierarchical RNN	76.63
Pretrained Bert + RNN	76.56
Fine-tuned Bert + RNN	78.05

Clone the repo: git clone --recurse-submodules https://github.com/JandJane/DialogueActClassification.git
Unzip data: unzip DialogueActClassification/swda/swda.zip -d DialogueActClassification/swda/swda
Install requirements: pip install -r DialogueActClassification/requirements.txt
Run notebooks 01-07

Name		Name	Last commit message	Last commit date
Latest commit History 14 Commits
dialogue_classification		dialogue_classification
swda @ 3baa1fe		swda @ 3baa1fe
.gitmodules		.gitmodules
01_EDA.ipynb		01_EDA.ipynb
02_TF-IDF.ipynb		02_TF-IDF.ipynb
03_Fasttext.ipynb		03_Fasttext.ipynb
04_Pretrained_Bert.ipynb		04_Pretrained_Bert.ipynb
05_Fasttext_Hierarchical_RNN.ipynb		05_Fasttext_Hierarchical_RNN.ipynb
06_Bert_RNN_Frozen.ipynb		06_Bert_RNN_Frozen.ipynb
07_Bert_RNN_Finetuning.ipynb		07_Bert_RNN_Finetuning.ipynb
README.md		README.md
requirements.txt		requirements.txt