Question Answering over Knowledge Base with Neural Attention Combining Global Knowledge Information

• DOI: 10.48550/arxiv.1606.00979
• Publication Date: 2016-01-01
• Authors: Yuanzhe Zhang; Kang Liu; Shizhu He; Guoliang Ji; Zhanyi Liu; Hua Wu; Jun Zhao
• Keywords: Leverage (statistics); Commonsense knowledge; Representation

With the rapid growth of knowledge bases (KBs) on web, how to take full advantage them becomes increasingly important. Knowledge base-based question answering (KB-QA) is one most promising approaches access substantial knowledge. Meantime, as neural network-based (NN-based) methods develop, NN-based KB-QA has already achieved impressive results. However, previous work did not put emphasis representation, and converted into a fixed vector regardless its candidate answers. This simple representation strategy unable express proper information question. Hence, we present attention-based model represent questions dynamically according different focuses various answer aspects. In addition, leverage global inside underlying KB, aiming at integrating rich KB And it also alleviates out vocabulary (OOV) problem, which helps attention more precisely. The experimental results WEBQUESTIONS demonstrate effectiveness proposed approach.