@article{Pramanik24c,
TITLE = {{UNIQORN}: {U}nified Question Answering over {RDF} Knowledge Graphs and Natural Language Text},
AUTHOR = {Pramanik, Soumajit and Alabi, Jesujoba and Saha Roy, Rishiraj and Weikum, Gerhard},
LANGUAGE = {eng},
ISSN = {1873-7749},
DOI = {10.1016/j.websem.2024.100833},
PUBLISHER = {Elsevier},
ADDRESS = {Amsterdam},
YEAR = {2024},
MARGINALMARK = {$\bullet$},
DATE = {2024},
ABSTRACT = {Question answering over knowledge graphs and other RDF data has been greatly<br>advanced, with a number of good systems providing crisp answers for natural<br>language questions or telegraphic queries. Some of these systems incorporate<br>textual sources as additional evidence for the answering process, but cannot<br>compute answers that are present in text alone. Conversely, systems from the IR<br>and NLP communities have addressed QA over text, but barely utilize semantic<br>data and knowledge. This paper presents the first QA system that can seamlessly<br>operate over RDF datasets and text corpora, or both together, in a unified<br>framework. Our method, called UNIQORN, builds a context graph on the fly, by<br>retrieving question-relevant triples from the RDF data and/or the text corpus,<br>where the latter case is handled by automatic information extraction. The<br>resulting graph is typically rich but highly noisy. UNIQORN copes with this<br>input by advanced graph algorithms for Group Steiner Trees, that identify the<br>best answer candidates in the context graph. Experimental results on several<br>benchmarks of complex questions with multiple entities and relations, show that<br>UNIQORN, an unsupervised method with only five parameters, produces results<br>comparable to the state-of-the-art on KGs, text corpora, and heterogeneous<br>sources. The graph-based methodology provides user-interpretable evidence for<br>the complete answering process.<br>},
JOURNAL = {Journal of Web Semantics},
VOLUME = {83},
EID = {100833},
}