Avature/MELO-Benchmark

MELO Benchmark

This dataset contains the Multilingual Entity Linking of Occupations (MELO) Benchmark for easy loading with the HuggingFace datasets library.

Dataset Description

MELO is a collection of 48 datasets for evaluating the linking of entity mentions in 21 languages to the ESCO Occupations multilingual taxonomy. It was built using high-quality, pre-existent human annotations.

Original paper abstract:

We present the Multilingual Entity Linking of Occupations (MELO) Benchmark, a new collection of 48 datasets for evaluating the linking of entity mentions in 21 languages to the ESCO Occupations multilingual taxonomy. MELO was built using high-quality, pre-existent human annotations. We conduct experiments with simple lexical models and general-purpose sentence encoders, evaluated as bi-encoders in a zero-shot setup, to establish baselines for future research.

Original repository: https://github.com/Avature/melo-benchmark

Dataset Structure

Each subset (configuration) contains two splits:

• queries: Query texts with their relevant corpus element indices
• corpus: Corpus element texts

Schema

queries split:

Column	Type	Description
text	string	The query surface form
labels	list[int]	Indices of relevant corpus elements

corpus split:

Column	Type	Description
text	string	The corpus element surface form

Usage

from datasets import load_dataset

# Load a specific subset
ds = load_dataset("federetyk/MELO-Benchmark", "ita_q_it_c_it")

# Access the data
query_surface_forms = ds["queries"]["text"]
corpus_surface_forms = ds["corpus"]["text"]
label_lists = ds["queries"]["labels"]

# Example: Get relevant corpus texts for the first query
query_idx = 0
print(f"Query: {query_surface_forms[query_idx]}")
print(f"Relevant corpus elements:")
for corpus_idx in label_lists[query_idx]:
    print(f"  - {corpus_surface_forms[corpus_idx]}")

Available Subsets

The subset names follow the pattern: {country}_q_{query_lang}_c_{corpus_lang(s)}

Subset ID	Task Name	Source Taxonomy	Query Lang	# Queries	Target Taxonomy	Corpus Lang	# Corpus
usa_q_en_c_en	USA-en-en	O*NET	en	633	ESCO v1.1.0	en	33,813
usa_q_en_c_de_en_es_fr_it_nl_pl_pt	USA-en-xx †	O*NET	en	633	ESCO v1.1.0	xx	150,140
aut_q_de_c_de	AUT-de-de	Austria	de	1,120	ESCO v1.1.0	de	19,782
aut_q_de_c_en	AUT-de-en	Austria	de	1,120	ESCO v1.1.0	en	33,813
bel_q_fr_c_fr	BEL-fr-fr	Belgium	fr	328	ESCO v1.0.3	fr	15,227
bel_q_fr_c_en	BEL-fr-en	Belgium	fr	328	ESCO v1.0.3	en	33,609
bel_q_nl_c_nl	BEL-nl-nl	Belgium	nl	328	ESCO v1.0.3	nl	24,070
bel_q_nl_c_en	BEL-nl-en	Belgium	nl	328	ESCO v1.0.3	en	33,609
bgr_q_bg_c_bg	BGR-bg-bg	Bulgaria	bg	4,438	ESCO v1.0.3	bg	21,082
bgr_q_bg_c_en	BGR-bg-en	Bulgaria	bg	4,438	ESCO v1.0.3	en	33,609
cze_q_cs_c_cs	CZE-cs-cs	Czechia	cs	988	ESCO v1.0.9	cs	13,333
cze_q_cs_c_en	CZE-cs-en	Czechia	cs	988	ESCO v1.0.9	en	33,583
deu_q_de_c_de	DEU-de-de	Germany	de	1,779	ESCO v1.0.3	de	19,135
deu_q_de_c_en	DEU-de-en	Germany	de	1,779	ESCO v1.0.3	en	33,609
dnk_q_da_c_da	DNK-da-da	Denmark	da	734	ESCO v1.0.8	da	10,410
dnk_q_da_c_en	DNK-da-en	Denmark	da	734	ESCO v1.0.8	en	33,583
esp_q_es_c_es	ESP-es-es	Spain	es	1,580	ESCO v1.0.8	es	16,502
esp_q_es_c_en	ESP-es-en	Spain	es	1,580	ESCO v1.0.8	en	33,583
est_q_et_c_et	EST-et-et	Estonia	et	1,068	ESCO v1.0.8	et	4,956
est_q_et_c_en	EST-et-en	Estonia	et	1,068	ESCO v1.0.8	en	33,583
fra_q_fr_c_fr	FRA-fr-fr	France	fr	1,435	ESCO v1.0.9	fr	15,217
fra_q_fr_c_en	FRA-fr-en	France	fr	1,435	ESCO v1.0.9	en	33,583
hrv_q_hr_c_hr	HRV-hr-hr	Croatia	hr	2,347	ESCO v1.0.3	hr	17,390
hrv_q_hr_c_en	HRV-hr-en	Croatia	hr	2,347	ESCO v1.0.3	en	33,609
hun_q_hu_c_hu	HUN-hu-hu	Hungary	hu	362	ESCO v1.0.8	hu	16,923
hun_q_hu_c_en	HUN-hu-en	Hungary	hu	362	ESCO v1.0.8	en	33,583
ita_q_it_c_it	ITA-it-it	Italy	it	362	ESCO v1.0.8	it	16,199
ita_q_it_c_en	ITA-it-en	Italy	it	362	ESCO v1.0.8	en	33,583
ltu_q_lt_c_lt	LTU-lt-lt	Lithuania	lt	3,849	ESCO v1.0.8	lt	17,824
ltu_q_lt_c_en	LTU-lt-en	Lithuania	lt	3,849	ESCO v1.0.8	en	33,583
lva_q_lv_c_lv	LVA-lv-lv	Latvia	lv	3,251	ESCO v1.0.8	lv	9,733
lva_q_lv_c_en	LVA-lv-en	Latvia	lv	3,251	ESCO v1.0.8	en	33,583
nld_q_nl_c_nl	NLD-nl-nl	Netherlands	nl	2,605	ESCO v1.0.3	nl	24,070
nld_q_nl_c_en	NLD-nl-en	Netherlands	nl	2,605	ESCO v1.0.3	en	33,609
nor_q_no_c_no	NOR-no-no	Norway	no	96	ESCO v1.0.8	no	7,821
nor_q_no_c_en	NOR-no-en	Norway	no	96	ESCO v1.0.8	en	33,583
pol_q_pl_c_pl	POL-pl-pl	Poland	pl	1,937	ESCO v1.0.3	pl	8,879
pol_q_pl_c_en	POL-pl-en	Poland	pl	1,937	ESCO v1.0.3	en	33,609
prt_q_pt_c_pt	PRT-pt-pt	Portugal	pt	379	ESCO v1.0.3	pt	11,671
prt_q_pt_c_en	PRT-pt-en	Portugal	pt	379	ESCO v1.0.3	en	33,609
rou_q_ro_c_ro	ROU-ro-ro	Romania	ro	3,273	ESCO v1.0.8	ro	14,833
rou_q_ro_c_en	ROU-ro-en	Romania	ro	3,273	ESCO v1.0.8	en	33,583
svk_q_sk_c_sk	SVK-sk-sk	Slovakia	sk	2,040	ESCO v1.0.8	sk	12,899
svk_q_sk_c_en	SVK-sk-en	Slovakia	sk	2,040	ESCO v1.0.8	en	33,583
svn_q_sl_c_sl	SVN-sl-sl	Slovenia	sl	3,222	ESCO v1.0.8	sl	15,487
svn_q_sl_c_en	SVN-sl-en	Slovenia	sl	3,222	ESCO v1.0.8	en	33,583
swe_q_sv_c_sv	SWE-sv-sv	Sweden	sv	2,883	ESCO v1.1.1	sv	7,506
swe_q_sv_c_en	SWE-sv-en	Sweden	sv	2,883	ESCO v1.1.1	en	33,802

† The usa_q_en_c_de_en_es_fr_it_nl_pl_pt subset contains a multilingual corpus with texts in German, English, Spanish, French, Italian, Dutch, Polish, and Portuguese.

Subset Naming Convention

• {country}: ISO 3166-1 alpha-3 country code (e.g., ita for Italy, usa for USA)
• q_{lang}: Query language (ISO 639-1 code)
• c_{lang(s)}: Corpus language(s) (one or more ISO 639-1 codes)

Examples:

• ita_q_it_c_it: Italian queries, Italian corpus (Italy)
• ita_q_it_c_en: Italian queries, English corpus (Italy)
• usa_q_en_c_de_en_es_fr_it_nl_pl_pt: English queries, multilingual corpus (USA)

Citation

If you use this dataset, please cite the original MELO Benchmark paper:

@inproceedings{retyk2024melo,
  title        = {{MELO: An Evaluation Benchmark for Multilingual Entity Linking of Occupations}},
  author       = {Federico Retyk and Luis Gasco and Casimiro Pio Carrino and Daniel Deniz and Rabih Zbib},
  booktitle    = {Proceedings of the 4th Workshop on Recommender Systems for Human Resources
                  (RecSys in {HR} 2024), in conjunction with the 18th {ACM} Conference on
                  Recommender Systems},
  year         = {2024},
  url          = {https://recsyshr.aau.dk/wp-content/uploads/2024/10/RecSysHR2024-paper_2.pdf},
}

License

This dataset is licensed under the MIT License. See the LICENSE file for more information.