Pattern-revising Enhanced Simple Question Answering over Knowledge Bases

Proceedings of the 27th International Conference on Computational Linguistics, pages 3272–3282 Santa Fe, New Mexico, USA, August 20-26, 2018. 3272 Pattern-revising Enhanced Simple Question Answering over Knowledge Bases Yanchao Hao1,2, Hao Liu1, Shizhu He1, Kang Liu1,2, Jun Zhao1,2 1 National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China 2 University of Chinese Academy of Sciences, Beijing, 100049, China {yanchao.hao, hao.liu2017, shizhu.he, kliu, jzhao}@nlpr.ia.ac.cn Abstract Question Answering over Knowledge Bases (KB-QA), which automatically answer natural lan- guage questions based on the facts contained by a knowledge base, is one of the most important natural language processing (NLP) tasks. Simple questions constitute a large part of questions queried on the web, still being a challenge to QA systems. In this work, we propose to conduct pattern extraction and entity linking first, and put forward pattern revising procedure to mitigate the error propagation problem. In order to learn to rank candidate subject-predicate pairs to en- able the relevant facts retrieval given a question, we propose to do joint fact selection enhanced by relation detection. Multi-level encodings and multi-dimension information are leveraged to strengthen the whole procedure. The experimental results demonstrate that our approach sets a new record in this task, outperforming the current state-of-the-art by an absolute large margin. 1 Introduction As the amount of the knowledge bases (KBs) grows, such as DBpedia1, Freebase2, and WikiData3, people are paying more attention to seeking effective methods for accessing these precious intellectu- al resources. While knowledge bases are usually very large and not easily accessible for users. KB- QA (Unger et al., 2014), which takes natural language as query language, is a more user-friendly solu- tion, and has become a research focus in recent years. At the same time the growing amount of data has led to a heterogeneous data landscape where QA systems struggle to keep up with the volume, variety and veracity of the underlying knowledge. Simple question answering over knowledge bases, which answers a question using a single fact in knowledge bases, is not simple at all and far from being solved. This is the setup of the SimpleQuestions benchmark recently presented by Bordes et al. (2015). The task of KB-QA for simple questions could be put as follows. Let G = (si, pi, oi) be a background knowledge base represented as a set of triples, where si represents a subject entity, pi a predicate (also denoted as relation), and oi an object entity. Given a natural language question represented as an utterance q = w1, ..., wn, the task of simple question answering is to find a triple (ˆs, ˆp, ˆo) ∈ G such that ˆo is the intended answer for question q. This task can be formulated to finding the best matches of subject ˆs and predicate ˆp. ˆs, ˆp = arg max s,p∈G p(s, p|q) (1) The setup of using a single fact to answer a question seems simple, while there are many existing challenges faced by QA systems. One challenge is the …

Content truncated to 500 words