NERSkill.Id: Annotated dataset of Indonesian's skill entity recognition

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Data Science

Background:

• Annotated datasets are crucial for training Natural Language Processing (NLP) models.
• Existing annotated corpora for the Indonesian language, particularly for skill entities, are scarce.
• Accurate skill identification is vital for talent acquisition, job matching, and educational program development.

Purpose of the Study:

• To introduce NERSkill.Id, a manually annotated Named Entity Recognition (NER) dataset for skill entities in Indonesian.
• To provide a valuable resource for advancing NLP research in Indonesian.
• To support stakeholders in improving talent acquisition and educational alignment with market needs.

Main Methods:

• Data collection from a job portal.
• Manual annotation of tokens following the BIO (Beginning, Inside, Outside) scheme.
• Processing using open-source libraries for dataset construction.

Main Results:

• The NERSkill.Id dataset contains 418,868 tokens.
• 15.51% of tokens are identified as named entities, categorized into hard skill, soft skill, and technology.
• The dataset is suitable for training and evaluating NER systems for skill extraction.

Conclusions:

• NERSkill.Id addresses the scarcity of annotated Indonesian corpora for skill entities.
• The dataset facilitates advancements in skill entity recognition for the Indonesian language.
• It offers significant benefits for NLP researchers, companies, recruiters, and educational institutions.