EFO-1-QA Benchmark for First Order Query Estimation on Knowledge Graphs

This repository contains an entire pipeline for the EFO-1-QA benchmark. EFO-1 stands for the Existential First Order Queries with Single Free Varibale. The related paper has been submitted to the NeurIPS 2021 track on dataset and benchmark. OpenReview Link, and appeared on arXiv

If this work helps you, please cite

@article{EFO-1-QA,
  title={Benchmarking the Combinatorial Generalizability of Complex Query Answering on Knowledge Graphs},
  author={Wang, Zihao and Yin, Hang and Song, Yangqiu},
  journal={arXiv preprint arXiv:2109.08925},
  year={2021}
}

The pipeline overview.

Query type generation and normalization The query types are generated by the DFS iteration of the context free grammar with the bounded negation hypothesis. The generated types are also normalized to several normal forms
Query grounding and answer sampling The queries are grounded on specific knowledge graphs and the answers that are non-trivial are sampled.
Model training and estimation We train and evaluate the specific query structure

Query type generation and normalization

The OpsTree is represented in the nested objects of FirstOrderSetQuery class in fol/foq_v2.py. We first generate the specific OpsTree and then store then by the formula property of FirstOrderSetQuery.

The OpsTree is generated by binary_formula_iterator in fol/foq_v2.py. The overall process is managed in formula_generation.py.

To generate the formula, just run

python formula_generation.py

Then the file formula csv is generated in the outputs folder. In this paper, we use the file in outputs/test_generated_formula_anchor_node=3.csv

Query grounding and answer sampling

We first prepare the KG data and then run the sampling code

The KG data (FB15k, FB15k-237, NELL995) should be put into under 'data/' folder. We use the data provided in the KGReasoning.

The structure of the data folder should be at least

data
	|---FB15k-237-betae
	|---FB15k-betae
	|---NELL-betae

Then we can run the benchmark sampling code on specific knowledge graph by

python benchmark_sampling.py --knowledge_graph FB15k-237 
python benchmark_sampling.py --knowledge_graph FB15k
python benchmark_sampling.py --knowledge_graph NELL

Append new forms to existing data One can append new forms to the existing dataset by

python append_new_normal_form.py --knowledge_graph FB15k-237

Model training and estimation

Models

Examples

The detailed setting of hyper-parameters or the knowledge graph to choose are in config folder, you can modify those configurations to create your own, all the experiments are on FB15k-237 by default.

Besides, the generated benchmark, one can also use the BetaE dataset after converting to our format by running:

python transform_beta_data.py

Use one of the commands in the following, depending on the choice of models:

python main.py --config config/{data_type}_{model_name}.yaml

The data_type includes benchmark and beta
The model_name includes BetaE, LogicE, NewLook and Query2Box

If you need to evaluate on the EFO-1-QA benchmark, be sure to load from existing model checkpoint, you can train one on your own or download from here:

python main.py --config config/benchmark_beta.yaml --checkpoint_path ckpt/FB15k/Beta_full
python main.py --config config/benchmark_NewLook.yaml --checkpoint_path ckpt/FB15k/NLK_full --load_step 450000
python main.py --config config/benchmark_Logic.yaml --checkpoint_path ckpt/FB15k/Logic_full --load_step 450000

We note that the BetaE checkpoint above is trained from KGReasoning

Paper Checklist

For all authors..

(a) Do the main claims made in the abstract and introduction accurately reflect the paper's contributions and scope? Yes

(b) Have you read the ethics review guidelines and ensured that your paper conforms to them? Yes

(c) Did you discuss any potential negative societal impacts of your work? No

(d) Did you describe the limitations of your work? Yes
If you are including theoretical results...

(a) Did you state the full set of assumptions of all theoretical results? N/A

(b) Did you include complete proofs of all theoretical results? N/A
If you ran experiments...

(a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? Yes

(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? Yes

(c) Did you report error bars (e.g., with respect to the random seed after running experiments multiple times)? No

(d) Did you include the amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? No
If you are using existing assets (e.g., code, data, models) or curating/releasing new assets...

(a) If your work uses existing assets, did you cite the creators? Yes

(b) Did you mention the license of the assets? No

(c) Did you include any new assets either in the supplemental material or as a URL? Yes

(d) Did you discuss whether and how consent was obtained from people whose data you're using/curating? N/A

(e) Did you discuss whether the data you are using/curating contains personally identifiable information or offensive content? N/A
If you used crowdsourcing or conducted research with human subjects...

(a) Did you include the full text of instructions given to participants and screenshots, if applicable? N/A

(b) Did you describe any potential participant risks, with links to Institutional Review Board (IRB) approvals, if applicable? N/A

(c) Did you include the estimated hourly wage paid to participants and the total amount spent on participant compensation? N/A

CodingMice / EFO-1-QA-benchmark