feat: add MELO and MELS datasets as ranking tasks (#37)

Introduce the MELO (Multilingual Entity Linking of Occupations) and MELS
(Multilingual Entity Linking of Skills) benchmarks as new ranking tasks.

MELO provides 42 evaluation datasets spanning 21 languages for job title
normalization into ESCO, built from crosswalks between national occupation
taxonomies and ESCO published by official EU labor organizations. MELS
follows the same methodology but targets skill normalization, covering
5 languages with 8 datasets.

- Add MELORanking task class with 42 datasets across 21 languages
- Add MELSRanking task class with 8 datasets across 5 languages
- Implement get_dataset_languages() for both tasks, supporting
  monolingual and cross-lingual dataset variants
- Add Austria and Belgium datasets to MELO (6 additional dataset IDs)
- Add unit tests for dataset ID filtering and language mapping
- Update README and example scripts with new tasks

Author: federetyk

README.md

@@ -207,12 +207,14 @@ lang_result_ci = summary["mean_per_language/en/f1_macro/ci_margin"]
 | Job to Skills WorkBench        | multi_label | 3039 queries x 13939 targets        | 28        |
 | Job Title Similarity           | multi_label | 105 queries x 2619 targets          | 11        |
 | Job Normalization              | single_label | 15463 queries x 2942 targets        | 28        |
+| Job Normalization MELO         | multi_label | 633 queries x 33813 targets         | 21        |
 | Skill to Job WorkBench         | multi_label | 13492 queries x 3039 targets        | 28        |
 | Skill Extraction House         | multi_label | 262 queries x 13891 targets         | 28        |
 | Skill Extraction Tech          | multi_label | 338 queries x 13891 targets         | 28        |
 | Skill Extraction SkillSkape    | multi_label | 1191 queries x 13891 targets        | 28        |
 | Skill Similarity SkillMatch-1K | single_label | 900 queries x 2648 targets          | 1         |
 | Skill Normalization ESCO       | multi_label | 72008 queries x 13939 targets       | 28        |
+| Skill Normalization MELS       | multi_label | 1722 queries x 19466 targets        | 5         |
 | Query-Candidate Matching       | multi_label | 200 queries x 4019 (x-lang) targets | 5         |
 | Project-Candidate Matching     | multi_label | 200 queries x 4019 (x-lang) targets | 5         |
 | **Classification**             

examples/run_benchmark_flat_average.py

@@ -59,7 +59,8 @@
         # Tasks with monolingual, cross-lingual, and multilingual datasets
         workrb.tasks.ProjectCandidateRanking(split=split, languages=langs),
         workrb.tasks.SearchQueryCandidateRanking(split=split, languages=langs),
-        # TODO: add MELO and MELS tasks when PR #37 is merged
+        workrb.tasks.MELORanking(split=split, languages=langs),
+        workrb.tasks.MELSRanking(split=split, languages=langs),
     ]
 
     # Evaluate

examples/run_benchmark_flat_average_all_langs.py

@@ -45,7 +45,8 @@
         # Tasks with monolingual, cross-lingual, and multilingual datasets
         workrb.tasks.ProjectCandidateRanking(split=split, languages=langs),
         workrb.tasks.SearchQueryCandidateRanking(split=split, languages=langs),
-        # TODO: add MELO and MELS tasks when PR #37 is merged
+        workrb.tasks.MELORanking(split=split, languages=langs),
+        workrb.tasks.MELSRanking(split=split, languages=langs),
     ]
 
     # Evaluate

examples/run_benchmark_language_weighted.py

@@ -56,7 +56,8 @@
         # Tasks with monolingual, cross-lingual, and multilingual datasets
         workrb.tasks.ProjectCandidateRanking(split=split, languages=langs),
         workrb.tasks.SearchQueryCandidateRanking(split=split, languages=langs),
-        # TODO: add MELO and MELS tasks when PR #37 is merged
+        workrb.tasks.MELORanking(split=split, languages=langs),
+        workrb.tasks.MELSRanking(split=split, languages=langs),
     ]
 
     # Evaluate

examples/run_benchmark_language_weighted_all_langs.py

@@ -42,7 +42,8 @@
         # Tasks with monolingual, cross-lingual, and multilingual datasets
         workrb.tasks.ProjectCandidateRanking(split=split, languages=langs),
         workrb.tasks.SearchQueryCandidateRanking(split=split, languages=langs),
-        # TODO: add MELO and MELS tasks when PR #37 is merged
+        workrb.tasks.MELORanking(split=split, languages=langs),
+        workrb.tasks.MELSRanking(split=split, languages=langs),
     ]
 
     # Evaluate

src/workrb/tasks/__init__.py

@@ -15,6 +15,8 @@
 from .ranking.job2skill import ESCOJob2SkillRanking
 from .ranking.job_similarity import JobTitleSimilarityRanking
 from .ranking.jobnorm import JobBERTJobNormRanking
+from .ranking.melo import MELORanking
+from .ranking.mels import MELSRanking
 from .ranking.skill2job import ESCOSkill2JobRanking
 from .ranking.skill_extraction import (
     HouseSkillExtractRanking,
@@ -39,6 +41,8 @@
     "ESCOSkillNormRanking",
     "JobBERTJobNormRanking",
     "JobTitleSimilarityRanking",
+    "MELORanking",
+    "MELSRanking",
     "HouseSkillExtractRanking",
     "TechSkillExtractRanking",
     "SkillSkapeExtractRanking",

src/workrb/tasks/abstract/ranking_base.py

@@ -38,6 +38,8 @@ def __init__(
         target_indices: list[list[int]],
         target_space: list[str],
         dataset_id: str,
+        allow_duplicate_queries: bool = True,
+        allow_duplicate_targets: bool = False,
     ):
         """Initialize ranking dataset with validation.
 
@@ -56,7 +58,7 @@ def __init__(
         self.target_indices = self._postprocess_indices(target_indices)
         self.target_space = self._postprocess_texts(target_space)
         self.dataset_id = dataset_id
-        self.validate_dataset()
+        self.validate_dataset(allow_duplicate_queries, allow_duplicate_targets)
 
     def validate_dataset(
         self,

src/workrb/tasks/ranking/__init__.py

@@ -14,6 +14,8 @@
 from workrb.tasks.ranking.job2skill import ESCOJob2SkillRanking
 from workrb.tasks.ranking.job_similarity import JobTitleSimilarityRanking
 from workrb.tasks.ranking.jobnorm import JobBERTJobNormRanking
+from workrb.tasks.ranking.melo import MELORanking
+from workrb.tasks.ranking.mels import MELSRanking
 from workrb.tasks.ranking.skill2job import ESCOSkill2JobRanking
 from workrb.tasks.ranking.skill_extraction import (
     HouseSkillExtractRanking,
@@ -30,6 +32,8 @@
     "HouseSkillExtractRanking",
     "JobBERTJobNormRanking",
     "JobTitleSimilarityRanking",
+    "MELORanking",
+    "MELSRanking",
     "ProjectCandidateRanking",
     "SearchQueryCandidateRanking",
     "SkillMatch1kSkillSimilarityRanking",