---
language:
  - "en"
pretty_name: "MultiSource-ESCO-Skills: A Unified Dataset for Skill Extraction"
tags:
  - nlp
  - esco
  - skills
  - text-extraction
  - dataset
license: "cc0-1.0"
task_categories:
  - text-classification
  - sentence-similarity
---

# MultiSource-ESCO-Skills: A Unified Dataset for Skill Extraction

This dataset aggregates data from multiple sources—course descriptions, CV content, and job descriptions—all linked to ESCO skills. It is designed to help researchers and practitioners develop and fine-tune NLP models (e.g., BERT or SentenceTransformer-based models) for automated skill extraction.

## Dataset Overview

- **Name:** MultiSource-ESCO-Skills
- **Sources:**
  - **Course Content:** Educational course materials
  - **CV Content:** Curriculum vitae and resumes
  - **Job Descriptions:** Listings and descriptions of job roles
- **Data Format:** CSV file
- **Structure:** Each row in the CSV represents a single sentence extracted from the original JSON files. The key fields include:
  - **escoid:** The unique identifier linking to the ESCO skill (URI).
  - **preferredLabel:** The standardized label or name of the ESCO skill.
  - **description:** Detailed information about the ESCO skill.
  - **sentence:** An individual sentence extracted from the source text.
  - **sentence_type:** Indicates whether the sentence is from the "explicit" or "implicit" category.
  - **extract:** A categorical label indicating the source of the data (`course`, `cv`, or `job`).

## Data Creation Process

The dataset is generated by merging three JSON files (one for courses, one for CVs, and one for jobs). Each JSON file contains an array of objects that include:
- **ESCO Skill Metadata:** `escoid`, `preferredLabel`, and `description`
- **Sentence Examples:** Two lists of sentences, one under `"explicit"` and one under `"implicit"`

For each object, every sentence from both the explicit and implicit lists is extracted into individual rows, while retaining the associated ESCO metadata and adding an `extract` field to indicate the original data source.

## Intended Use Cases

- **Skill Extraction:** Fine-tune models to extract and map relevant skills from various textual inputs.
- **Semantic Similarity:** Develop embedding-based models to compare free-form text with standardized ESCO skill descriptions.
- **NLP Research:** Serve as a resource for studying language understanding and information extraction in vocational domains.