whybe-choi/kovidore-v2-hr-beir

KoViDoRe v2 : HR

This dataset, HR, is a corpus of reports on workforce outlook and employment policy in korea, intended for complex-document understanding tasks. It is one of the 4 corpora comprising the KoViDoRe v2 Benchmark.

Links

• Github: https://github.com/whybe-choi/kovidore-benchmark
• Collection: https://huggingface.co/collections/whybe-choi/kovidore-benchmark-beir-v2
• Data Generation Pipeline: https://github.com/whybe-choi/kovidore-data-generator

Dataset Summary

For this specific dataset (HR):

• Description: Consists of 9 reports on workforce outlook and employment policy in korea

• Language: ko

• Domain: HR

• Document Types: Reports

Dataset Statistics

• Total Documents : 9
• Total Pages : 2109
• Total Queries : 221
• Average number of pages per query : 3.3

Languages

The documents in this dataset are in korean.

Queries per Document

Doc ID	Context	Summary	Count
2018_blind_recruitment_guidebook	2	2	4
2023_advanced_chemical_materials_industrial_technical_manpower_outlook_report	8	15	23
2023_high_tech_fiber_materials_industrial_technical_manpower_outlook_report	6	5	11
2023_korea_occupational_outlook	45	8	53
2023_new_metal_materials_industrial_technical_manpower_outlook_report	6	11	17
2023_next_generation_ceramic_materials_industrial_technical_manpower_outlook_report	12	20	32
2023_secondary_battery_industrial_technical_manpower_outlook_report	10	13	23
2023_survey_on_the_status_of_work_life_balance	26	8	34
2024_labor_conditions_by_employment_type_report	15	9	24
Total	130	91	221

Queries type

Query Type	Count
Compare-Contrast	152
Numerical	76
Open-Ended	144
Enumerative	26
Boolean	5
Extractive	23
Multi-Hop	104

Queries format

Query Format	Count
Question	99
Instruction	96
Keyword	26
Total	221

Dataset Structure

1. Corpus

Contains the full collection of documents to be searched.

Data instance of a single item from the corpus subset:

{
  "corpus_id": <int>,
  "image": <PIL.Image>,
  "doc_id": <str>,
  "markdown": <str>,
  "elements": <str>,
  "page_number_in_doc": <int>
}

• corpus_id : A unique numerical identifier for the corresponding corpus document.
• image <PIL.Image> : The page
• doc_id : name of the document from where the image was extracted
• markdown : Extracted text from the Image using an OCR pipeline
• elements : JSON-serialized list of extracted layout elements with bounding boxes and text from the Image using an OCR pipeline
• page_number_in_doc : Original page number inside the document

2. Queries

Contains set of questions or search queries. Data Instance of a single item from the queries subset:

{
  "query_id": <int>,
  "query": <str>,
  "language": <str>,
  "query_types": <List[str]>,
  "query_format": <str>,
  "source_type": <str>,
  "query_type_for_generation": <str>,
  "answer": <str>
}

• query_id : A unique numerical identifier for the query.
• query : The actual text of the search question or statement used for retrieval.
• language : The language of the query text.
• query_types <List[str]> : A list of categories or labels describing the query's intent.
• query_format : The syntactic format of the query ("intruction", "keyword" or "question").
• source_type : "summary" or "context", metadata about the type of information used by the annotation pipeline to create the query
• query_type_for_generation : The specific type requested when the query was generated
• answer : The answer extracted from the source documents, merged from human annotations using an LLM.

3. Qrels

Maps queries to their corresponding relevant documents. Data Instance of a single item for the qrels subset:

{
  "query_id": <int>,
  "corpus_id": <int>,
  "score": <int>,
}

• query_id : A unique numerical identifier for the query.
• corpus_id : A unique numerical identifier for the corresponding corpus document.
• score : Relevance score for the pair <query, corpus>. Can be either 1 (Critically Relevant) or 2 (Fully Relevant):
  • Fully Relevant (2) - The page contains the complete answer.
  • Critically Relevant (1) - The page contains facts or information that are required to answer the query, though additional information is required.

4. Original PDFs

All the original pdfs used to build the corpus are distributed in the "pdfs" folder of this directory.

License information

All annotations, query-document relevance judgments (qrels), and related metadata generated for this corpus are distributed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

The licensing status of the original source documents (the corpus) and any parsed text (markdown column in the corpus) are inherited from their respective publishers. The specific license governing each original document is provided in the documents_metadata["license"] field of that document's entry.

For documents subject to the Korea Open Government License (KOGL) Type 1 and Type 2, the sources are attributed as follows:

Title	Doc ID	Type	Attribution Text
2024년 고용형태별근로실태조사	2024_labor_conditions_by_employment_type_report	Type 1	본 저작물은 고용노동부에서 2025년 작성하여 공공누리 제 1유형으로 개방한 '2024년 고용형태별근로실태조사 보고서'를 이용하였으며, 해당 저작물은 고용노동부 고용노동통계에서 무료로 다운받으실 수 있습니다.
2023 한국직업전망	2023_korea_occupational_outlook	Type 2	본 저작물은 한국고용정보원에서 2023년 작성하여 공공누리 제 2유형으로 개방한 '2023 한국직업전망(최영순 외 5명)'을 이용하였으며, 해당 저작물은 한국고용정보원 발간자료에서 무료로 다운받으실 수 있습니다.

Data Privacy and Removal Requests

While this dataset is released under open licenses, we respect the privacy of individuals and the ownership of source content. If you are a data subject, author, or publisher and are uncomfortable with the inclusion of your data or documents in this release, please contact us at whybe.choi@gmail.com. We will promptly review your request.

Acknowledgements

This dataset is inspired by the ViDoRe V3, and we thank the ViDoRe team for their foundational work. We also thank the Upstage x AWS AI Initiative for granting us free access to their API services. Finally, We acknowledge the datasets provided by the Public Data Portal(공공데이터포털), which were utilized to construct the tasks in KoViDoRe v2.