ESCO Entity Linking

This repository is for the paper

Entity Linking in the Job Market Domain

Mike Zhang, Rob van der Goot, and Barbara Plank. In EACL Findings 2024.

---

We make use of two separate models, BLINK and GENRE. Each have their separate folder. We suggest you to read each README of the respective repository to use the models and how to preprocess the data.

In each folder, we left an environment.yml to reproduce our experiments. We suggest to create a conda environment. Unfortunately, you have to use the environments separatly.

To install:

# BLINK
conda create env -f BLINK/environment.yml

# GENRE
conda create env -f GENRE/environment.yml

Once you followed the installation instructions in both BLINK and GENRE. You can train models.

Data

In this repository, you can find a .tar (data_entity_linking_esco.tar) file with the respective data for each model. We suggest putting the data in a /data/ folder per model directory.

Training

In both BLINK and GENRE, we left a folder ESCO_scripts for the ESCO-specific experiments.

To run these scripts. You can simply call them:

# BLINK
cd BLINK
bash ESCO_scripts/train_blink_biencoder.sh
bash ESCO_scripts/train_blink_biencoder_pretrained.sh

# GENRE
cd GENRE
bash ESCO_scripts/train_bart.sh
bash ESCO_scripts/train_genre_pretrained.sh

The *_pretrained.sh files are for further fine-tuning the models released by both BLINK and GENRE.

Evaluation

Also in the ESCO_scripts folders of both models, you can find the evaluation scripts BLINK/ESCO_scripts/eval_blink_biencoder.sh and GENRE/ESCO_scripts/evaluate_genre.sh.

Once you trained your models. These should be able to run. By default, it should also create the predictions.

Citation

If you have been using this work in your cool work, consider citing it:

@inproceedings{zhang-etal-2024-entity,
    title = "Entity Linking in the Job Market Domain",
    author = "Zhang, Mike  and
      Goot, Rob  and
      Plank, Barbara",
    editor = "Graham, Yvette  and
      Purver, Matthew",
    booktitle = "Findings of the Association for Computational Linguistics: EACL 2024",
    month = mar,
    year = "2024",
    address = "St. Julian{'}s, Malta",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.findings-eacl.28",
    pages = "410--419",
    abstract = "In Natural Language Processing, entity linking (EL) has centered around Wikipedia, but yet remains underexplored for the job market domain. Disambiguating skill mentions can help us get insight into the current labor market demands. In this work, we are the first to explore EL in this domain, specifically targeting the linkage of occupational skills to the ESCO taxonomy (le Vrang et al., 2014). Previous efforts linked coarse-grained (full) sentences to a corresponding ESCO skill. In this work, we link more fine-grained span-level mentions of skills. We tune two high-performing neural EL models, a bi-encoder (Wu et al., 2020) and an autoregressive model (Cao et al., 2021), on a synthetically generated mention{--}skill pair dataset and evaluate them on a human-annotated skill-linking benchmark. Our findings reveal that both models are capable of linking implicit mentions of skills to their correct taxonomy counterparts. Empirically, BLINK outperforms GENRE in strict evaluation, but GENRE performs better in loose evaluation (accuracy@k).",
}