"Open-domain question answering using a knowledge graph and web corpus" by Uma Sawant, Soumen Chakrabarti and Ganesh Ramakrishnan with Martin Vesely as coordinator

Abstract

In Web search, entity-seeking queries often trigger a special Question Answering (QA) system. It uses syntactic patterns to extract structure from the query, identify a semantic interpretation and then return direct entity responses from a knowledge graph (KG). Such QA systems tend to be brittle. Minor query variations may fail to trigger the QA system. Moreover, KG coverage is patchy at best. Rather than fall off the "structure cliff" in such cases, we propose a more robust approach that degrades gracefully on a "structure ramp". Our system, called AQQUCN, accepts a broad spectrum of queries, between well-formed questions to short "telegraphic" keyword sequences. In the face of inherent query ambiguities, AQQUCN aggregates signals from KGs and large corpora to directly rank KG entities, rather than commit to one semantic interpretation of the query. AQQUCN models the ideal interpretation as an unobservable or latent variable. Pairs of interpretations and candidate entity responses are scored as pairs, by combining signals from multiple convolutional networks that operate collectively on the query, KG and corpus. On four public query workloads amounting to over 8,000 queries in different query formats, we see 16--18% absolute improvement in mean average precision (MAP), compared to recent systems. Our system is also competitive when compared to recent KBQA systems.