Europeana Cultural Heritage Metadata Benchmark for LLM Querying

Europeana Cultural Heritage Metadata Benchmark for LLM Querying

This dataset is a benchmark package derived from Europeana metadata and designed for experiments on structured querying over large cultural heritage collections. It includes a cleaned tabular dataset (europeana.csv, in the zip file), a schema description (europeana.json), an ingestion script (ingest_europeana.sql), a SQL query set (queries_europeana.sql), and a natural-language query set (queries_europeana.txt). 

The benchmark is organized around a single table named `europeana` with the following fields:

1. title
2. creator
3. type
4. year
5. language
6. country
7. provider
8. dataProvider

The included query set contains 140 manually designed queries in both SQL and natural-language form. These queries cover a range of structured information needs, including:

• filtering by country, language, type, and year;
• temporal constraints over historical periods;
• top-k aggregations by country, provider, creator, language, and year;
• equality conditions between provider and dataProvider;
• retrieval of titles, creators, providers, and metadata attributes under controlled query patterns. 

This benchmark was created to support reproducible evaluation of querying strategies over cultural heritage data, with particular relevance to research on digital libraries, metadata access, and large language models used as structured knowledge access systems.