# Bikidata

Queries over Wikidata dumps. 

For more information, also see: https://github.com/ISE-FIZKarlsruhe/bikidata 

Wikidata is an invaluable resource in our research activities. But in many cases we need to do complex queries, or extract large swathes of data. This is an example on how we can convert [offline dumps of Wikidata](https://www.mediawiki.org/wiki/Wikibase/Indexing/RDF_Dump_Format) into a format that we can query in a structured manner, or run extractions on.

## Summary

We download a ["truthy" dump in NT format](https://dumps.wikimedia.org/wikidatawiki/entities/latest-truthy.nt.bz2). Split the uncompressed file into smaller chunks. Index the RDF structure using a [fast hash-algorithm](https://xxhash.com/) into [parquet files](https://parquet.apache.org/) of 3 unsigned 64-bit columns (subject-predicate-object). Index the IRIs and Literals into another parquet file of two columns, (hash, literal/IRI) and query the resulting files using [DuckDB](https://duckdb.org/).