KERL

Knowledge Graphs and Pre-trained Language Models enhanced Representation Learning for Conversational Recommender Systems

KERL leverages the power of knowledge graphs and pre-trained language models to generate semantically rich entity representations, enabling more accurate recommendations and engaging user interactions.

Installation Instructions

To run the provided code, follow these steps:

1. Download and install Miniconda:

   wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
   bash Miniconda3-latest-Linux-x86_64.sh

2. Create a new conda environment with provided environment.yml file:

   conda env create -f environment.yml
   conda activate kerl

ℹ️ Note:

1. Code tested only on WSL2 and Linux-based systems.
2. The provided Miniconda installation commands are for Linux. For other systems, download the appropriate installer from the official website.

Quick-Start

To train the KERL model, use the following commands. Select the appropriate configuration file based on the dataset:

# For training on the Inspired dataset
python main.py -c config/inspired_kerl.yaml

# For training on the ReDial dataset
python main.py -c config/redial_kerl.yaml

ℹ️ Note: If you wish to log the training metrics, you will need a wandb account. Alternatively, you can use a CSV logger. For more details on setting up a wandb account, please refer to the wandb quickstart guide.

Saved Models

You can download the saved models for two datasets from the following links:

1. Inspired Model: Download
2. Redial Model: Download

Place the downloaded saved folder in the root directory. In the configuration file for the respective dataset, set <phase>_reload to True, and <phase>_model_path with the path of the downloaded model.

Citation

@article{10530439,
	title        = {Knowledge Graphs and Pretrained Language Models Enhanced Representation Learning for Conversational Recommender Systems},
	author       = {Qiu, Zhangchi and Tao, Ye and Pan, Shirui and Liew, Alan Wee-Chung},
	year         = 2024,
	journal      = {IEEE Transactions on Neural Networks and Learning Systems},
	volume       = {},
	number       = {},
	pages        = {1--15},
	doi          = {10.1109/TNNLS.2024.3395334}
}

Acknowledgements

Parts of our implementation were inspired by the excellent CRSLab toolkit. We are grateful to the CRSLab developers for open sourcing their valuable work.