Schema Extraction and Structural Outlier Detection for JSON-based NoSQL Data Stores

Although most NoSQL Data Stores are schema-less, information on the structural properties of the persisted data is nevertheless essential during application development. Otherwise, accessing the data becomes simply impractical. In this paper, we introduce an algorithm for schema extraction that is operating outside of the NoSQL data store. Our method is specifically targeted at semi-structured data persisted in NoSQL stores, e.g., in JSON format. Rather than designing the schema up front, extracting a schema in hindsight can be seen as a reverse-engineering step. Based on the extracted schema information, we propose set of similarity measures that capture the degree of heterogeneity of JSON data and which reveal structural outliers in the data. We evaluate our implementation on two real-life datasets: a database from the Wendelstein 7-X project and Web Performance Data.