value-anonymizer

Janus/value-anonymizer is part of a peer-reviewed publication. Please see the paper here or a preprint on arXiv.

@inproceedings{pallas2022configurable,
  title={Configurable Per-Query Data Minimization for Privacy-Compliant Web APIs},
  author={Frank Pallas and David Hartmann and Paul Heinrich and Josefine Kipke and Elias Grünewald},
  booktitle={Web Engineering: 22nd International Conference, ICWE 2022, Bari, Italy, July 5--8, 2022},
  volume={13362},
  pages={325},
  year={2022},
  organization={Springer Nature}
}

Usage

Installation

npm install value-anonymizer

Overview

The anonymizer package consists of three anonymizing methods, that are working for the folloing data types:

noise: Number, Date
generalize: Number, String, Date
hash: Number, String, Date

Each function needs some anonymization parameters. In general, the call of the function is

functionName(value, parameters)

Which type of parameters object is required will be explained in the following.

Noise

In general, when using noise, one has to define a certain distribution that is used to sample the noise value that is added to the original value. Therefore, every distribution of the probability-distributions package (https://www.npmjs.com/package/probability-distributions) is available. To use them, pass parameter object as follows:

{   
    typeOfDistribution:"normal", 
    distributionParameters: {
        mean: 100,
        standardDeviation: 1
    }, 
    valueParameters: {
        isInt: true
    }
}

There are the following distributions available:

"binomial", parameters: ["number", "probability"]
"beta", parameters: ["alpha", "beta"]
"cauchy", parameters: ["scale", "location"]
"chiSquared", parameters: parameters: ["degrees"]
"exponential", parameters: ["rate"]
"fdistribution", parameters: ["degree1", "degree2"]
"gamma", parameters: ["shape", "rate"]
"laplace", parameters: ["mean", "standardDeviation"]
"logNormal", parameters: ["logMean", "logStandardDeviation"]
"negativeBinomial", parameters: ["failures", "probability"]
"normal", parameters: ["mean", "standardDeviation"]
"poisson", parameters: ["mean"]
"uniform", parameters: ["min", "max"]
"uniformInt", parameters: ["min", "max"]

For the valueParameters, on can specify for numbers wheter the output should be and Integer by isInt and for dates which unit the noise has by addNoiseToUnit.

Generalization

Strings

Pass a parameter object like

{
    generalizationParameters: {
        hideCharactersFromPosition: 4,
        numberOfHideCharacters: 3
    }
}

Returns "This***" for "ThisIsText" as input.

Numbers

Pass a parameter object like

{
    generalizationParameters: {
        stepSize: 10
    }
}

Values are always rounded down to the next smaller step size boundary. By way of example: If the stepsize is 10, there would be limits of 0 (representing 0-9), 10 (for 10-19), 20 (for 20-29), and so on. This ensures that exactly one number (0,10,20,...) represents a range. Thus, returns 20 for 29 as input.

Dates

{
    generalizationParameters: {
        dateUnit: "day"
    }
}

The resulting date is truncated to this Unit. Returns Date(2021,10,0) for Date(2021,10,23,2) as input.

Hashing

Uses the SHA3 hash function of crypto-js.

{
    hashingParameters: {
        outputLength: 256
    }
}

PrivacyEngineering / janus-value-anonymizer