Textual enrichment of three public multi-modal knowledge graph (MMKG)
benchmarks — MKG-W, MKG-Y, and DB15K — produced by the
Beyond Images pipeline
(ESWC 2026): entity images (both the originals shipped with each benchmark
and newly crawled Wikipedia images) are converted into textual descriptions
with BLIP-2, giving every entity a language view of its visual content.
Feeding these descriptions to standard MMKG completion models (MMRNS… See the full description on the dataset page:
https://huggingface.co/datasets/pengyu3/beyond-images-enriched.
