Multi-facet Universal Schema

Data preparation and training

cd multifacet_relation_extraction Follow the instructions in multifacet_relation_extraction/README.md

Evaluation of TAC

In order to follow the evaluation procedure in Verga et al., 2016, we need to perform following steps

Step1: cd to this repo. Set up the path using

export TAC_ROOT=`pwd`/tackbp2016-sf
export TH_RELEX_ROOT=`pwd`/torch-relation-extraction

Step2: Download and unzip the libraries needed for compiling the JAVA Code into tackbp2016-sf/components/pipeline/. To compile the JAVA code in ./tackbp2016-sf, assuming your jdk path is /usr/lib/jvm/java-1.8.0-openjdk-1.8.0.275.b01-0.el7_9.x86_64, you can run:

export JAVA_HOME="/usr/lib/jvm/java-1.8.0-openjdk-1.8.0.275.b01-0.el7_9.x86_64"
cd tackbp2016-sf
./components/pipeline/build.sh

Step3: Compile java codes in ./torch-relation-extraction

cd torch-relation-extraction
./setup-tac-eval.sh

Step4:

• Download this zip file and extract the data folder into ./torch-relation-extraction.
• In ./torch-relation-extraction, run:
  ./bin/tac-evaluation/test_all_NSD_formal_release.sh ../multifacet_relation_extraction/results/milestone_run_trans-b5-kb11_trans_results Then, the final scores will be stored in ../multifacet_relation_extraction/results/milestone_run_trans-b5-kb11_trans_results.

NOTE: We store the results of several different scoring functions in ../multifacet_relation_extraction/results/milestone_run_trans-b5-kb11 by default, which will make ./bin/tac-evaluation/test_all_NSD_formal_release.sh take a long time to finish. In order to make the code run faster, you can only keep the folder *_kmeans_avg in ../multifacet_relation_extration/results/milestone_run_trans-b5-kb11_trans_results, which stores the results we report in our paper.

F1 Score reported in our paper

To view the F1 score, run the jupyter notebook results/Results.ipynb.

Citation

If you use the codes in multifacet_relation_extraction for your paper, please cite Paul et al., 2021.

If you use the training data or codes in torch-relation-extraction, please cite Verga et al., 2016.

If you use the codes in tackbp2016-sf to perform slot filling, please cite Chang et al., 2016.

Rohan Paul*, Haw-Shiuan Chang*, and Andrew McCallum,
"Multi-facet Universal Schema."
Conference of the European Chapter of the Association for Computational Linguistics (EACL), 2021

Patrick Verga, David Belanger, Emma Strubell, Benjamin Roth, and Andrew McCallum,
"Multilingual Relation Extraction using Compositional Universal Schema."
Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (HLT/NAACL), 2016

Haw-Shiuan Chang, Abdurrahman Munir, Ao Liu, Johnny Tian-Zheng Wei, Aaron Traylor, Ajay Nagesh, Nicholas Monath, Patrick Verga, Emma Strubell, and Andrew McCallum,
"Extracting Multilingual Relations under Limited Resources: TAC 2016 Cold-Start KB construction and Slot-Filling using Compositional Universal Schema."
Text Analysis Conference, Knowledge Base Population (TAC/KBP), 2016