Towards Protecting Sensitive Text with Differential Privacy
Bibliographic record
Abstract
Natural language processing can often require handling privacy-sensitive text. To avoid revealing confidential information, data owners and practitioners can use differential privacy, which provides a mathematically guaranteeable definition of privacy preservation. In this work, we explore the possibility of applying differential privacy to feature hashing. Feature hashing is a common technique for handling out-of-dictionary vocabulary, and for creating a lookup table to find feature weights in constant time. Traditionally, differential privacy involves adding noise to hide the true value of data points. We show that due to the finite nature of the output space when using feature hashing, a noiseless approach is also theoretically sound. This approach opens up the possibility of applying strong differential privacy protections to NLP models trained with feature hashing. Preliminary experiments show that even common words can be protected with (0.04,10−5)-differential privacy, with only a minor reduction in model utility.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.014 | 0.064 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.002 | 0.003 |
| Science and technology studies | 0.002 | 0.007 |
| Scholarly communication | 0.006 | 0.021 |
| Open science | 0.004 | 0.012 |
| Research integrity | 0.003 | 0.007 |
| Insufficient payload (model declined to judge) | 0.004 | 0.003 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".