One unified, integer-indexed interaction corpus glued from 15 recommendation sources for
foundation / sequential recommenders. Built in 3 additive versions that share one schema and
one global index space, so reading all of them together = the full corpus, collision-free.
Interactions only — no model, no embeddings. Item text is provided separately as title+description
metadata, joinable 1:1 by item_idx.
Full corpus: 882… See the full description on the dataset page:
https://huggingface.co/datasets/TOPAPEC/cross-domain-recsys-interactions.
