Generalizability of Cross-Category Retrieval Benefits Across Technology Domains

Determine whether the hyperbolic representation's cross-category retrieval benefit is general across technology domains or specific to AI and software patents.

Background

The primary study evaluates the hyperbolic and Euclidean variational autoencoders primarily on AI and software patents. Because performance may depend on the structure and diversity of a technology domain, the authors identify domain generalizability as unresolved: the available single-domain evidence cannot establish whether the observed retrieval advantage extends broadly or is specific to the original patent categories. Study 2 addresses this limitation empirically by examining ten domains, but the quoted passage presents the unresolved question motivating that extension.

References

Study 1's principal limitation is that it examines a single technology domain, so it cannot say whether Section~\ref{sec:cross-category}'s retrieval benefit is general or specific to AI and software patents.

— Benchmarking Embedding Geometries and Clustering Algorithms for Unsupervised Patent Claim Classification  (2610.12997 - Subramanian et al., 9 Oct 2026) in Section 4.2, “Multi-Domain Extension (Study 2): Where Does This Approach Work Best”