Timeline for How can I transform names in a confidential data set to make it anonymous, but preserve some of the characteristics of the names?
Current License: CC BY-SA 4.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| S Jun 17 at 3:47 | history | suggested | cottontail | CC BY-SA 4.0 |
minor touchup to improve grammar
|
| May 3 at 16:40 | review | Suggested edits | |||
| S Jun 17 at 3:47 | |||||
| Jun 17, 2014 at 20:27 | comment | added | neone4373 | Right, I would just filter the matrix to only include Levenshteins below a certain cutoff, so you are only populating where there is high likelihood of overlap. Additionally, when it comes to PII I am of the mindset that if you include enough information to determine a relationship among disparate entities in your datasets, its very unlikely you are preserving the customers anonymity. The point of anonymizing the data is to avoid potential PII related regulatory headaches down that line, (standards can always be tightened), so personally I wouldn't take the risk. | |
| Jun 17, 2014 at 19:48 | comment | added | Air | That's why I brought up the idea of a substitution cipher (ROT13) since it preserves the distance between strings; but it's not secure, and I suspect it may be impossible to securely encrypt the strings while preserving the edit distance. (Would love to be wrong!) | |
| Jun 17, 2014 at 19:44 | comment | added | Air | What interests me about the paper I linked is that it claims to show a method for performing this sort of computation without knowledge of both input strings. In the paper, each actor has knowledge of one string, which isn't useful for my purposes; I would need one actor to be able to perform the calculation without knowledge of either string. Calculating them beforehand is only feasible for very small datasets or very limited products; a full cross product of integer distances on my dataset would take ~10 PB of storage. | |
| Jun 17, 2014 at 18:42 | history | answered | neone4373 | CC BY-SA 3.0 |