Improving Question Answering Quality Through Language Feature-based SPARQL Query Candidate Validation

Question Answering systems are on the rise and on their way to become one of the standard user interfaces. However, in conversational user interfaces, the information quantity needs to be kept low as users expect a limited number of precise answers (often it is 1) { similar to human-human communication. The acceptable number of answers in a result list is a key dfferentiator from search engines where showing more answers (10-100) to the user is widely accepted. Hence, the quality of Question Answering is crucial for the wide acceptance of such systems. The adaptation of natural-language user interfaces for satisfying the information needs of humans requires high-quality and not-redundant answers. However, providing compact and correct answers to the users' questions is a challenging task. In this paper, we consider a certain class of Question Answering systems that work over Knowledge Graphs. We developed a system-agnostic approach for optimizing the ranked lists of SPARQL query candidates produced by the Knowledge Graph Question Answering system that are used to retrieve an answer to a given question. We call this a SPARQL query validation process. For the evaluation of our approach, we used two well-known Knowledge Graph Question Answering benchmarks. Our results show a significant improvement in the Question Answering quality. As the approach is system-agnostic, it can be applied to any Knowledge Graph Question Answering system that produces query candidates.