In_Context_EL

This is our coding implementation for the paper:

@inproceedings{ding2024chatel,
 title={ChatEL: ChatEL: Entity Linking with Chatbots},
 author={Yifan Ding, Qingkai Zeng, Tim Weninger},
 booktitle={COLING-LREC},
 year={2024}
}

@article{ding2024entgpt,
  title={EntGPT: Linking Generative Large Language Models with Knowledge Bases},
  author={Yifan Ding, Amrit Poudel, Qingkai Zeng, Tim Weninger, Balaji Veeramani, Sanmitra Bhattacharya},
  journal={arXiv preprint arXiv:2402.06738},
  year={2024}
}

Author: Yifan Ding ([email protected])

installation

1. this repo is built on top of REL and BLINK. Install REL and BLINK from the corresponding websites.
2. install In_context_EL as a package

git clone https://github.com/yifding/In_Context_EL.git
cd In_Context_EL
pip install -e ./

3. prepare opeai API key, create a file named openai_key.py with openai API under the directory of In_Context_EL/in_context_el

OPENAI_API_KEY = "sk-*****"

4. download dataset files from google drive, unzip the file under the directory of In_Context_EL/data

run the code of ChatEL & EntGPT-P (GPT3.5 results)

0. donwload the running script with running outputs from google drive
1. obtain REL entity candidates

sh In_Context_EL/RUN_FILES/public/blink_candidates/KORE50.sh

2. obtain BLINK and REL entity candidates

sh In_Context_EL/RUN_FILES/public/rel_blink_candidates/KORE50.sh

3. obtain summary and context augmentation prompt response

sh In_Context_EL/RUN_FILES/public/mention_prompt/KORE50.sh

4. obtain multi-choice selection prompt response

sh In_Context_EL/RUN_FILES/public/entity_candidate_prompt/KORE50.sh

5. evaluation

sh In_Context_EL/RUN_FILES/public/evaluation/KORE50.sh

Reproduced Performances (GPT-3.5, zero-shot)

Dataset	Precision	Recall	F1
AIDA-YAGO	0.830	0.812	0.821
msnbc	0.900	0.835	0.866
aquaint	0.830	0.757	0.791
ACE 2004	0.913	0.856	0.884
clueweb - WNED-CWEB (CWEB)	0.767	0.662	0.711
wikipedia - WNED-WIKI (WIKI)	0.800	0.745	0.771
KORE50	0.814	0.639	0.716
OKE15	0.847	0.701	0.767
OKE16	0.855	0.701	0.770
Reuters-128	0.840	0.737	0.785
RSS-500	0.850	0.769	0.808