As of May 15, 2023, the Auto-redaction accuracy for Private Identifying Information type, phone number, reached nearly 100%. Using the Insurrection files (2020-11, 2020-12, and 2021-01) as a sample the auto-redaction script achieved improved accuracy overall.
Auto-redaction Accuracy by PII Type
Author Phone Number auto-Redacted 97.21%
Author Email Address auto-Redacted 90.91%
Author Physical Address auto-Redacted 23.03%
Overall, accuracy is close to 70%.
Although PII Type Physical Address has the lowest accuracy rate, it occurs much less frequently than the other two types. However, it continues to necessitate time-sensitive manual review.
We continue to refine processing as we observe and learn new format varieties increasing accuracy with each batch. Our goal is fully automate the redaction phase.