Tagger

Implementation of Tagger, a schema extraction tool initially proposed in the article Extracting JSON Schemas with Tagged Unions by Stefan Klessinger, Meike Klettke, Uta Störl and Stefanie Scherzinger at DEco@VLDB '22.

Tagger utilizes conditional functional dependency discovery to extract tagged unions from JSON documents and translates them to if-then-else statements in JSON Schema. A number of heuristics are applied to prevent overfitting.

Tagger is designed for compatibility with third-party schema extraction approaches, allowing Tagger to focus exclusively on the detection of tagged unions while relying on other tools to extract a more general schema. As an example, a slightly adapted implementation (extract.py) of the approach Schema Extraction and Structural Outlier Detection for JSON-based NoSQL Data Stores by Meike Klettke, Uta Störl, and Stefanie Scherzinger (published at BTW 2015) is integrated here.

This repository is organized in two branches: The main development branch and the branch "deco", where a version of our code for the DEco Workshop at VLDB 2022 is maintned (see "Citation").

Citation

To refer to this work, please use these BibTeX entries.

@inproceedings{Klessinger:2022:JSONTaggedUnions,
  author    = {Stefan Klessinger and
               Meike Klettke and
               Uta St{\"{o}}rl and
               Stefanie Scherzinger},
  title     = {Extracting JSON Schemas with Tagged Unions},
  year      = {2022}
  booktitle = {Proc.\ DEco@VLDB 2022}
}

Setup

For simplifying the setup, a Pipfile containing all Python dependencies is provided. With pipenv installed, run pipenv install in the directory of the Pipfile.

Reproduction Package

A reproduction package, provided as a fully automated Docker container, including all input data used in our experiments, is available in a separate repository.