Autoregressive Models for Knowledge Graph Generation
Thiviyan Thanapalasingam [email protected], Antonis Vozikis [email protected], Peter Bloem [email protected], Paul Groth [email protected] · 2026
Read Autoregressive Models for Knowledge Graph Generation by Thiviyan Thanapalasingam [email protected], Antonis Vozikis [email protected], Peter Bloem [email protected], Paul Groth [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-06707.txt.
Description
Thiviyan Thanapalasingam, Antonis Vozikis, Peter Bloem, and Paul Groth teach machines to generate whole knowledge graphs, not just fill in single missing links. A knowledge graph stores facts as triples of head, relation, and tail, and a good generator must keep the whole structure sensible, respecting types, timing, and relational patterns. Their models, called ARK, produce graphs triple by triple in sequence.
Because the models generate in order, they learn the rules of a domain straight from examples, with no hand-written supervision. On a public benchmark they produced graphs that scored between about 89 and 100 percent on semantic validity, while still containing graphs never seen in training. A variational version adds knobs for controlled generation and for completing a partial graph. The authors also found that model width matters more than depth for this task.
Catalog record
| CREATOR | Thiviyan Thanapalasingam [email protected], Antonis Vozikis [email protected], Peter Bloem [email protected], Paul Groth [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-06707