Ubaid Azam, Imran Razzak, Steven Schockaert, Shoaib Jameel · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3839959
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
While deep reasoning models (e.g., NBFNet, ULTRA) have advanced knowledge graph completion, they often sacrifice the efficiency and interpretability of geometric embeddings for increased expressivity. We present PROCS, a framework that combines the best of both worlds through region-based representations inspired by Gärdenfors' Conceptual Spaces, realised as a scalable probabilistic embedding layer. Unlike deterministic point embeddings, PROCS models entities as Gaussian distributions where variance explicitly encodes vagueness and conflict. By introducing a scalable geometric inference mechanism that avoids intractable sampling, PROCS achieves 0.417 MRR on FB15k-237, matching the performance of heavy reasoning models while retaining the O(1) inference of embedding methods. Its uncertainty estimates enable risk-aware retrieval, validated by a Defence-domain case study. Our code and data are available at https://github.com/ubaidazam/PROCS.
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
No comments yet — start the discussion below.