Due to the popularity of the JavaScript Object Notation (JSON), a need has arisen for the creation of schema documents for the purpose of validating the content of other JSON documents. Existing automatic schema generation tools, however, have not adequately considered the scenario of an array of JSON objects with different types of structures. These tools work off the assumption that all objects have the same structure, and thus, only generate a single schema combining them together. To address this problem, this thesis looks to improve upon schema generation for heterogeneous JSON arrays. We develop an algorithm to determine a set of keys that identifies what type of structure each element has. These keys are then used as the basis for a schema decision tree. The objective of this tree is to help in the validation process by allowing each element to be compared against a single, more tailored, schema.
