MMFS directories for LargeRDFBench datasets

The file mmfs.tar.zst contains all cleaned (RDF 1.1) datasets from the LargeRDFBench, encoded with the Mapped-Memory Friendly Store  (MMFS) format, which is used in experiments which code, scripts and results are also available as a dataset in Zenodo: 10.5281/zenodo.13960678.

Uncompressed, the 13 directories will cost 36.317 GiB in disk space. The list of directories inside mmfs.tar.zst is:

Affymetrix

ChEBI

DBPedia-Subset

DrugBank

GeoNames

Jamendo

KEGG

LinkedTCGA-A

LinkedTCGA-E

LinkedTCGA-M

LMDB

NYT

SWDFood

Each MMFS directory contains 4 files:

shared: suffixes and prefixes that are shared between multiple RDF terms;

strings: The dictionary that maps the N-Triples representation of each RDF term in the graph to a unique 64-bit sequential identifier. This dictionary also decomposes the N-Triples strings into a local part, stored at the end of this file, and a shared prefix/suffix, stored at the shared file;

lexical: This is a small list L of bitsets, where L[hash(t) % L.length] is the set of literal suffixes for which there MAY be a term u such that str(u) = str(t).

spo: This is a triples index which stores a sorted (by ID) list of predicate and object pairs for each subject;

pso: Similar to spo, but stores subject-object pairs for a given predicate;

ops: Similar to spo, but stores predicate-subject pairs given an object.