This is the code and data for ACL 2019 long paper "Learning Representation Mapping for Relation Detection in Knowledge Base Question Answering".
As we discuss, for SimpleQuestion, 99% of the relations in the test set also exist in the training data. In order to evaluate unseen relation detection and seen relation detection fairly, we re-organize SimpleQuestion to SimpleQuestion_Balance(SQB), the dataset is released at Data/SQB and the script for re-organize this dataset is mix_dataset.py.
The main code for this paper is qa+adapter.
Our relation embeddings are trained by JointNRE between FB2M and wikipedia, please see this link for detail.
cd qa+adapter
bash script/run_baseline.sh
cd qa+adapter
bash script/run_baseline-star.sh
cd qa+adapter
bash script/run-other.sh $card
We use FocusPrune to annotated the entity, please refer to https://github.com/wudapeng268/KBQA-Baseline for detail.
cd qa+adapter
bash script/test-all-kbqa.sh $card_num
Our data for this experiment at Data/Number_relation_in_training created by this script.
You can use following script to reproduce this result:
cd qa+adapter
bash script/run_tl.sh
If you use our code or data, please kindly cite the paper about it!
@inproceedings{peng19acl,
title = {Learning Representation Mapping for Relation Detection in Knowledge Base Question Answering},
author = {Peng Wu, Shujian Huang, Rongxiang Weng, Zaixiang Zheng, Jianbing Zhang, Xiaohui Yan and Jiajun Chen},
booktitle = {The 57th Annual Meeting of the Association for Computational Linguistics (ACL)},
address = {Florence, Italy},
month = {July},
year = {2019}
}