popqa-sampling-condition-exp8r-6

IN premise — summaries/2026/08/24/mallen-2023-when-not-to-trust-sA-appendix.md

Created 2026-08-25T02:58:12+00:00

POPQA knowledge triples are included if f > exp(8R − 6) where R ~ U(0,1) and f is the exact-match term frequency of the subject entity's aliases in an 800 MB random sample of C4

Summary

This is a random filter that decides which POPQA facts to keep: a fact survives only if its subject entity's name shows up often enough in a large web-text sample, and the cutoff is randomized so that well-known entities almost always pass while obscure ones are dropped with high probability. In practice, it biases the retained knowledge toward commonly mentioned topics and away from rare or niche ones.