- Keywords: knowledge extraction, entity-oriented querying, hybrid reasoning
- Abstract: The use of large Knowledge Bases (KBs) in modern semantic intensive applications is gradually becoming the norm. Entities from a KB can be used to annotate and semantically enrich data (e.g., holiday destinations or patient profiles) as well as build intelligent services (e.g., holiday booking or symptom checking engines). To access such information or services users usually form entity-centric queries expressed using noun phrases like ``Small Mediterranean Islands'', ``Cheap Pizza Restaurants'', or ``I have a pain in my left leg''. Answering such queries or matching them to the annotated entities is a challenging task since it requires a deeper understanding of their semantics as well as taking into account the axioms defined in the KB (both hierarchy and relations). In the current paper we propose a framework for KB-based understanding and reasoning with such queries that is heavily based on formal Semantic Web technologies. First, we develop a method which given such short phrases it extracts a Description Logic concept in an attempt to formally capture their semantics. Second, we develop a novel reasoning algorithm which given two concepts constructed using entities from a KB compares them by possibly also exploiting information that is encoded in their text labels. The latter is necessary due to possible ``underspecification'' of KBs which hinders recall of formal reasoning methods. We demonstrate the feasibility and practical relevance of our approach by testing it in a real-world industrial strength application in Anonymous which develops an AI-based symptom checking dialogue system. Our approach coupled with sentence embeddings as a fall-back is used to activate Anonymous' symptom checking engine and provides the best combination of precision and recall compared to any other method used alone.
- Archival status: Archival
- Subject areas: Natural Language Processing, Question Answering, Reasoning, Knowledge Representation, Semantic Web