Question Answering on DBpedia - Answer Type Dataset

This dataset contains 37,279 questions and their respective answer category and type (list). The questions are extracted from the QALD 1-9, LC-QuAD 1.0 and SimpleDBpediaQA train and test datasets - based on DBpedia. The answer types are adopted from the SMART-task challenge of ISWC 2020 - more details at https://smart-task.github.io/.

The dataset is structured as JSON array and each question record contains the following fields:

id: The question ID of the original KGQA dataset

source: The name of the original KGQA dataset

question: The actual natural language question.

category: The answer category - either literal, resource, or boolean.

type: The type list.

For the type list, the following cases are applicable: In case of "resource" as answer category, the type list contains the ontology class(es) assigned to the correct answer resource. Sometimes, this list is empty as the SPARQL query of the original dataset did not require a type setting of the answer. In case, the answer category is literal, the type list contains only one string: either "string", "date", or "number". In case the answer category is "boolean", the type list contains the type "boolean".