Constraint-Based Question Answering with Knowledge Graph

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2503–2514, Osaka, Japan, December 11-17 2016. Constraint-Based Question Answering with Knowledge Graph Junwei Bao†∗, Nan Duan‡ , Zhao Yan♯ , Ming Zhou‡ , Tiejun Zhao† †Harbin Institute of Technology, Harbin, China ‡Microsoft Research, Beijing, China ♯Beihang University, Beijing, China †baojunwei001@gmail.com ‡{nanduan, mingzhou}@microsoft.com ♯yanzhao@buaa.edu.cn †tjzhao@hit.edu.cn Abstract WebQuestions and SimpleQuestions are two benchmark data-sets commonly used in recent knowledge-based question answering (KBQA) work. Most questions in them are ‘simple’ ques- tions which can be answered based on a single relation in the knowledge base. Such data-sets lack the capability of evaluating KBQA systems on complicated questions. Motivated by this issue, we release a new data-set, namely ComplexQuestions1, aiming to measure the quality of KBQA systems on ‘multi-constraint’ questions which require multiple knowledge base relations to get the answer. Beside, we propose a novel systematic KBQA approach to solve multi-constraint questions. Compared to state-of-the-art methods, our approach not only obtains comparable re- sults on the two existing benchmark data-sets, but also achieves significant improvements on the ComplexQuestions. 1 Introduction Knowledge-based question answering is a task that aims to answer natural language questions based on existing knowledge bases (KB). In the last decades, large scale knowledge bases, such as Free- base (Bollacker et al., 2008), have been constructed. Based on Freebase, two benchmark data-sets, WebQuestions (Berant et al., 2013) and SimpleQuestions (Bordes et al., 2015) are constructed and used in most of KBQA work (Berant and Liang, 2014; Bordes et al., 2014a; Fader et al., 2014; Yang et al., 2014; Bao et al., 2014; Reddy et al., 2014; Dong et al., 2015; Yih et al., 2015). Forest Whitaker Mark Rydell Even Money acted_films director 𝑪𝒏���𝒏𝒓���𝒓���𝒓𝒂𝒊𝒏𝒓��� Forest Whitaker acted_films Even Money Southpaw …… Which films star by Forest Whitaker Which films star by Forest Whitaker and are directed by Mark Rydell Figure 1: Simple and multi-constraint questions. However, about 85% of questions (Yao, 2015) of WebQuestions and all questions in SimpleQuestions are ‘simple’ questions, where a ‘simple’ question denotes that it can be answered based on a single KB relation. For example in Figure 1, “Which films star by Forest Whitaker” is a simple question that can be answered by the KB triples like ⟨Forest Whitaker,acted films,?⟩ with a single KB relation acted films. This leads to the fact that such data-sets cannot measure the capability of KBQA systems on ‘multi- constraint’ questions, where ‘multi-constraint’ means a question containing multiple semantic constraints expressed with different expressions to restrict the answer set. To answer a multi-constraint question, we have to base on multiple KB re- lations. For example in Figure 1, “Which films star by Forest Whitaker and are directed by Mark Ry- dell” is a multi-constraint question with a constraint “directed by Mark Rydell”, which requires multiple ∗This work was finished while the author was visiting Microsoft Research Asia. 1https://github.com/JunweiBao/MulCQA/tree/ComplexQuestions This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/ 2503 Constraint Category Example Percentage …

Content truncated to 500 words