JPIS: A Joint Model for Profile-Based Intent Detection and Slot Filling with Slot-to-Intent Attention

We propose a joint model, namely JPIS, designed to enhance profile-based intent detection and slot filling. JPIS incorporates the supporting profile information into its encoder and introduces a slot-to-intent attention mechanism to transfer slot information representations to intent detection. Experimental results show that our JPIS substantially outperforms previous profile-based models, establishing a new state-of-the-art performance in overall accuracy on the Chinese benchmark dataset ProSLU.

Please CITE our paper whenever our JPIS implementation is used to help produce published results or incorporated into other software.

@inproceedings{JPIS,
title     = {{JPIS: A Joint Model for Profile-based Intent Detection and Slot Filling with Slot-to-Intent Attention}},
author    = {Thinh Pham and Dat Quoc Nguyen},
booktitle = {Proceedings of the 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
year      = {2024}
}

Model installation, training and evaluation

Installation

Python version >= 3.8
PyTorch version >= 1.7.1

    git clone https://github.com/VinAIResearch/JPIS.git
    cd JPIS/
    pip3 install -r requirements.txt

Training and evaluation

To train the model, you can run the experiments by the following command:

python train.py \
        --gpu \
        --early_stop \
        --save_dir model_dir \
        --use_crf \
        --num_epoch 50 \
        --s2i \
        --i2s \
        --up \
        --ca \
        --use_pretrained \
        --model_type RoBERTa

If you have any questions, please issue the project or email me (v.thinhphp1@vinai.io or thinhphp.nlp@gmail.com) and we will reply soon.

Acknowledgement

Our code is based on the implementation of the ProSLU paper from https://github.com/LooperXX/ProSLU