APRILS - Prophesying Axiom Relevance
by Latent Semantic Analysis
Idea
- Used for analysing relationship between documents
- Analogy - Formula = Document; Symbol = Word
Computing LSA Relevance
- Compute initial contextual symbol relationship strengths
- Iterate to take transitivity into account
- Compute formula symbol strength vectors
- LSA relevance between formulae is dot product of their vectors
Evaluation vs. Prophet
- 1337 problems with proofs by EP
- Compare highest ranks of axioms in proof
- APRILS wins 49%, Prophet wins 28%
- APRILS better on larger problems