Talk by Happy Buzaaba. (University of Tsukuba)

Speaker: Happy Buzaaba (University of Tsukuba)

Title: A Modular and Efficient Approach for Question Answering Over Knowledge Bases

Abstract:
Question answering over knowledge base (KBQA) is an important and challenging task which aims at correctly answering natural language questions posed by humans. It has a wide range of application in natural language processing (NLP) and information retrieval (IR) such as search, dialogue systems, information extraction to mention but a few. Studies on KBQA have been conducted using large scale knowledge bases (KB), such as Freebase [1], and DBpedia [2] as knowledge sources. Such large scale knowledge bases, consist of real world entities as nodes and the relations between them as edges. Each directed edge along with its head and tail entity, constitute a triple that is to say (head entity, relation, tail entity).

Given a natural language question such as where was Barack Obama born?, the KBQA system is tasked with identifying a triple (Barack Obama, people/person/place of birth, Honololu) from the the knowledge base such that the tail entity (Honololu) is the answer to the question. Several existing KBQA systems [3, 4, 5] exploit complex end-to-end neural network architectures to solve the KBQA task. While these methods achieve good performance, it comes at a high memory and computational cost and also the end-to-end approaches, make it even more difficult to conduct a detailed performance analysis.

Our current research focuses on how to efficiently perform question answering over knowledge bases. In this study, the question answering task is decomposed into three different components of entity detection, entity linking, and relation prediction, and we solve each of the components separate to come up with the correct answer candidate to the question [6]. This approach performs reasonably well compared to complex neural network methods and it provides us with the opportunity to understand the problem structure. However, our current study assumes a static knowledge base yet knowledge changes over time in the real world. It becomes difficult for a typical KBQA system to answer questions related to the new knowledge when the KB evolves since it lacks the ability to detect knowledge unseen in the training. Going forward, we believe that casting the KBQA task as a life-long learning problem would overcome the above challenge.

References:
[1] Kurt Bollacker, Colin Evans, Praveen Paritosh, Tim Sturge, and Jamie Taylor. Freebase: a collabo-ratively created graph database for structuring human knowledge. In Proceedings of the 2008 ACM SIGMOD international conference on Management of data, pages 1247–1250, 2008.
[2] Jens Lehmann, Robert Isele, Max Jakob, Anja Jentzsch, Dimitris Kontokostas, Pablo N Mendes, Sebastian Hellmann, Mohamed Morsey, Patrick Van Kleef, S¨oren Auer, et al. Dbpedia–a large-scale, multilingual knowledge base extracted from wikipedia. Semantic web, 6(2):167–195, 2015.
[3] Antoine Bordes, Sumit Chopra, and Jason Weston. Question answering with subgraph embeddings. arXiv preprint arXiv:1406.3676, 2014.
[4] David Golub and Xiaodong He. Character-level question answering with attention. arXiv preprint arXiv:1604.00727, 2016.
[5] Wenpeng Yin, Mo Yu, Bing Xiang, Bowen Zhou, and Hinrich Sch¨utze. Simple question answering by attentive convolutional neural network. In Proceedings of COLING 2016, the 26th International Con-ference on Computational Linguistics: Technical Papers, pages 1746–1756, Osaka, Japan, December 2016. The COLING 2016 Organizing Committee.
[6] Happy Buzaaba and Toshiyuki Amagasa. Question answering over knowledge base: A scheme for integrating subject and the identified relation to answer simple questions. SN Comput. Sci., 2(1):25, 2021.