README.md

metadata

license: cc-by-nc-nd-4.0
task_categories:
  - text-retrieval
language:
  - en

Download

To download a dataset in the hahu test repository, create an access read token.

Run the following script to use the dataset as a pandas Dataframe:

from huggingface_hub import hf_hub_download
import pandas as pd
HF_TOKEN = "hf_..."
REPO_ID = "tabiya/hahu_test"
FILENAME = "..."

dataset = pd.read_csv(
    hf_hub_download(repo_id=REPO_ID, filename=FILENAME, repo_type="dataset", token=HF_TOKEN)
)

Description

The hahu_test.csv file is the original file provided by Hahu jobs with the following fields:

• title: The title of the job position, indicating the specific role and/or position within the organization.
• description: A detailed description of the job responsibilities, requirements, qualifications, and other relevant information pertaining to the job role. This might include sensitive information regarding the company which posted the job.
• general_classification: The broad classification or category of the job, providing insight into the sector or field to which the job belongs. It is one of 'Sales and Services', 'Wholesale and Retail Trade', 'Transport, Storage and ICT', 'Manufacturing, Construction and Engineering', 'Restaurant and Hospitality Services', 'Financial, Insurance and Business Services', 'Community, Social Services and Public Administration', 'Professional, Scientific and Technical', 'Education', 'Manufacturing and Construction', 'Health'.
• esco_code: The ESCO (European Skills, Competences, Qualifications, and Occupations) code associated with the job, facilitating standardized classification and comparison across different job listings.

This dataset offers 550 hand-labeled job offers available in Ethiopia, spanning diverse sectors and roles. The curation enhances its reliability for natural language processing tasks, such as entity linking to the ESCO database.

The hahu_test_with_id.csv and redacted_hahu_test_with_id.csv files are artifacts produced from the original file. The first one consists of the same fields with an additional ID field to allow the comparison with the other artifact. The ID are generated based on the original order of the rows.

The redacted_hahu_test_with_id.csv consists of 542 job offers, as 8 of them have removed because their code was deprecated in ESCO v1.1.1. It contains the following fields:

• ID: the ID of the row which allows to identify rows of the original dataset with rows of the redacted dataset.
• title: the title of the job position, indicating the specific role and/or position within the organization.
• redacted_description: a detailed description of the job responsibilities, requirements, qualifications, and other relevant information pertaining to the job role. The sensitive information in the original field have been redacted using Google DLP API.
• general_classification: the broad classification or category of the job, providing insight into the sector or field to which the job belongs. It is one of 'Sales and Services', 'Wholesale and Retail Trade', 'Transport, Storage and ICT', 'Manufacturing, Construction and Engineering', 'Restaurant and Hospitality Services', 'Financial, Insurance and Business Services', 'Community, Social Services and Public Administration', 'Professional, Scientific and Technical', 'Education', 'Manufacturing and Construction', 'Health'.
• esco_code: the ESCO (European Skills, Competences, Qualifications, and Occupations) code associated with the job, facilitating standardized classification and comparison across different job listings.
• synthetic_query: a query generated synthetically from the redacted description. This can be used for evaluation of linking user's requests for the Compass app to the ESCO Occupation Database.
• skills_essential: list of Tabiya UUIDs of essential skills linked to the occupation of interest. It can be used to evaluate how skills are linked to a job description or to the resulting synthetic query.
• skills_optional: list of Tabiya UUIDs of optional skills linked to the occupation of interest.

The synthetic_queries_translated.csv is a new version of redacted_hahu_test_with_id.csv with the following fields:

• ID: same as redacted_hahu_test_with_id.csv.
• esco_code: same as redacted_hahu_test_with_id.csv.
• synthetic_query: same as redacted_hahu_test_with_id.csv
• fr_synthetic_query: translation of synthetic_query to French with Gemini 1.0 Pro.
• fr_to_en_synthetic_query: back-translation of fr_synthetic_query from French to English with Gemini 1.0 Pro for sample evaluation.