atoroghi/bikg

This repository contains the implementation of our AAAI-24 paper Bayesian Inference with Complex Knowledge Graph Evidence (BIKG).

In order to use the code, please follow these steps:

1- Install requirements

pip install -r requirements.txt

2- Download the Datasets

You can download the five datasets used in the paper from here and de-compress them.

3- Download Embedding Models or Train them

You can download the KGE models from here and de-compress them. Alternatively, you can train your own models using the following code with your desired arguments.

python -m kbc.learn data/Movielens_twohop --rank 100 --max_epochs 300 --batch_size 128 --model SimplE --valid 1 --model_save_schedule 10

4- Sequential Knowledge Graph-based Query Answering (KGQA)

In order to run the KGQA experiments, please run the following command with your desired arguments.

python -m kbc.cqd_beam data/NELL --model_path models/NELL-DistMult-model-rank-300-epoch-300-1690457575.pt --dataset NELL --candidates 1 --reasoning_mode bayesian1 --mode test --seq yes --chain_type 1_2_seq

5- Critiquing with Complex Evidence

In order to perform critiquing with complex evidence experiments, please run the following files with your desired arguments. Please note that this experiment requires hyperparameter tuning.

python -m kbc.cqd_beam_bpl data/Movielens_twohop --model_path models/Movielens_twohop-SimplE-model-rank-50-epoch-30-1687217986.pt --dataset Movielens_twohop --candidates 3 --quantifier marginal_ui --cov_anchor 1e-2 --cov_var 1e-2 --cov_target 1e-2

You can also do the hyperparameter tuning by running the following code that performs grid search:

python kbc.CC_simulator

Thank you for your attention!