feat: add Job Title Similarity ranking task

[federetyk]

closes #24

Description

Added the Job Title Similarity dataset (Avature/Job-Title-Similarity) as a new ranking task in WorkRB. This task evaluates a model's ability to rank job titles by semantic similarity to a query job title. The dataset includes 11 languages (en, de, es, fr, it, ja, ko, nl, pl, pt, zh) with ~105 queries and ~2,500 corpus job titles per language.

Changes:

• Added new task group RankingTaskGroup.JOBSIM in src/workrb/types.py
• Added support for Chinese (zh), Japanese (ja), and Korean (ko) languages in the Language enum in src/workrb/tasks/abstract/ranking_base.py
• Created JobTitleSimilarityRanking task class in src/workrb/tasks/ranking/job_similarity.py
• Exported the new task in src/workrb/tasks/__init__.py and src/workrb/tasks/ranking/__init__.py
• Added tests for the new task in tests/test_task_loading.py

Task characteristics:

• Task type: Ranking
• Label type: Multi-label (each query has multiple relevant corpus documents)
• Query/Target input type: Job titles
• Default evaluation metrics: MAP, MRR, RP@5, RP@10 (as used by Zbib et al., 2022)

References:

• Zbib et al. (2022): "Learning Job Titles Similarity from Noisy Skill Labels" (link)
• Deniz et al. (2024): "Combined Unsupervised and Contrastive Learning for Multilingual Job Recommendations" (link)

Checklist

• Added new tests for new functionality
• Tested locally with example tasks
• Code follows project style guidelines
• Documentation updated
• No new warnings introduced

[Mattdl]

Thanks @federetyk, super valuable contribution!

Code looks good to me. Would just add some clarifications in the JobSImilarity task as it is the first task of its kind (see review).

[Mattdl]

Would be great to give a bit more context here on the JobTitleSimilarityTask.

1. Provide a link to the huggingface repo, and shortly how it is used (corpus and query sets) and languages it covers.
2. How this task differs from JobNormalization (for each job title you have multiple similar job titles, and others are deemed non-similar), whereas jobNormalization maps to a single best-matching canonical job title.
3. Give an example of a query and labels.

[Mattdl]

Additionally, to create visibilitiy, would add your dataset entry in the README.md table of datasets (along with nb targets x nb queries).

[federetyk]

Code looks good to me. Would just add some clarifications in the JobSImilarity task as it is the first task of its kind (see review).

@Mattdl Thanks for the review! I have updated the docstring with the requested context and added the entry to the README table as suggested.

[Mattdl]

Looks great, ready to merge.