Answer Category and Type Prediction with BERT Classifier Using Wikidata

Answer type prediction aims to identify the types of answer results to filter out irrelevant results which particularly improves QA systems. Answers categorized into boolean (true/false), numeric and string data types.

Two stage workflow

1. Multi-class classification that predicts the answer category of the question - (boolean, number, string, date, and resource) by fine-tuning BERT classifier.

2. BERT Classifier applied to predict the general types for each question in resource category.

More can be found in google slides: