schependom/synthology
Ontology-Based Synthetic Data Generation for Neuro-Symbolic Knowledge Graph Reasoning.
What's novel
Ontology-Based Synthetic Data Generation for Neuro-Symbolic Knowledge Graph Reasoning.
Code Analysis
19 files read · 6 roundsGenerates synthetic knowledge graph training data for neurosymbolic link prediction by constructing backward-chaining proof trees from OWL 2 RL ontologies, guaranteeing multi-hop derivations and producing hard negatives by corrupting base facts within those proof trees.
Strengths
The backward-chaining proof construction engine is a genuinely novel and deeply implemented approach to synthetic data generation — rule indexing, recursive cycle detection via DFS, individual pooling, constraint-aware generation, and proof-based negative sampling are all non-trivial algorithms implemented from scratch. The project is well-organized as a UV workspace with clear separation between the core data structures, ontology parser, chainer, negative sampler, validator, and the RRN evaluat
Weaknesses
Test coverage is sparse relative to the project's complexity — the core chainer (60KB), negative sampler (45KB), and parser (34KB) have minimal dedicated test files, and the ont_generator tests directory appears to contain only a few files. Some code paths in the chainer and negative sampler lack defensive error handling for edge cases like empty rule sets or degenerate ontologies.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
107
Contributors
1
Files
105
Active weeks
6
Repository
Language
Python
Stars
2
Forks
0
License
MIT