OneNet

Repository for OneNet: A Fine-Tuning Free Framework for Few-Shot Entity Linking via Large Language Model Prompting. @ EMNLP 2024, main. [arxiv]

Set Up

Any environment with a reasonable Huggingface Transformers installation should be fine. You can also refer to the environments we provided in requirements.txt.

Datasets

For Wiki-based data, please visit NER4EL

For ZeShEL, please visit ZeShEL

Since OneNet is a multi-step pipeline, some steps may be time-consuming. Therefore, we also provide some intermediate results (e.g. summaries, filtered entities), please visit this link.

Usage

To start, you can just run the run.sh

bash run.sh

run.sh is the script that calls the LLM inference in new_code/prompt, you just need to change func_name, you can run the inference of different steps in OneNet, the func_name description is as follows

func_name	file format	Description
summary	summary.json	Get summary of enitities
point_wise	pointwise_raw.json	Filter the irrelevant entities
category	listwise_filter.json	get the category of mention
context	listwise_filter.json	entity linking based on context
prior	listwise_filter.json	entity linking based on prior
merge	listwise_merge.json	merge EL result from context and prior

We have organized most of the core code into new_code\. Specifically, prompt.py contains the prompts for all modules, while data_process.py contains most of the data processing code. Other details can also be found in other folders

Evaluation

Please run eval.sh.

bash eval.sh

We also provide the result file in the datasets, please visit link.

Citation

If you find our work interesting/helpful, please consider citing OneNet