Code for Mitigating Data Sparsity for Short Text Topic Modeling by Topic-Semantic Contrastive Learning (EMNLP 2022)

Check our latest topic modeling toolkit TopMost !

EMNLP 2022

Usage

1. Prepare Environment

python==3.7
pytorch==1.7.1
scipy==1.7.3
scikit-learn==0.23.2
pyyaml==6.0

2. Training

Training without data augmentation:

python run.py --data_dir data/${dataset} --model TSCTM --num_topic {number of topic}

Training with data augmentation:

python run.py --data_dir data/${dataset} --model TSCTM_aug --num_topic {number of topic}

After training, outputs are in ./output/{dataset}/, including topic words, topic distributions of short texts, and topic-word distribution matrix.

3. Evaluation

topic coherence: coherence score

topic diversity:

python utils/TU.py --data_path {path of topic word file}

Citation

If you want to use our code, please cite as

@inproceedings{wu-etal-2022-mitigating,
    title = "Mitigating Data Sparsity for Short Text Topic Modeling by Topic-Semantic Contrastive Learning",
    author = "Wu, Xiaobao  and
      Luu, Anh Tuan  and
      Dong, Xinshuai",
    booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing",
    month = dec,
    year = "2022",
    address = "Abu Dhabi, United Arab Emirates",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.emnlp-main.176",
    pages = "2748--2760",
}