Text classification of 38,416 bore-log intervals, 2,811 of which describe nothing recovered
A research poster from the School of Underlying Conditions trains a linear text classifier on 38,416 free-text interval descriptions from 1,204 legacy water-bore logs and compares three treatments of the 2,811 entries that record no recovery: forcing a rock class, adding a tenth class, and rejecting below a confidence threshold. A probe trained on the loss wording alone names the rock bracketing a lost interval with accuracy 0.58 against a majority baseline of 0.29, and the authors propose reporting that suggestion apart from the rock classes.
| Authors | Ragnhild Uzoma , Joost Nwosu , Edda Nwafor |
|---|---|
| School | School of Underlying Conditions |
| Output type | Research poster |
| Published | |
| DOI | 10.5555/slop.vkgxes |
| Pages | 1 |
| Cited by | 1 output |
| Version | 1.0 |
| Licence | CC BY 4.0 |
Cites
Cited by
Cite as
@misc{slop_vkgxes,
author = {Ragnhild Uzoma and Joost Nwosu and Edda Nwafor},
title = {What the Driller Wrote Instead: Text classification of 38,416 bore-log intervals, 2,811 of which describe nothing recovered},
year = {2026},
publisher = {Slop University},
doi = {10.5555/slop.vkgxes},
url = {https://slop.university/outputs/slop-poster-sorting-drillers-free-text-vkgxes/},
version = {1.0},
note = {Research poster},
}