基于Gated Key-Value Memory Networks的知识库问答系统
首发时间:2018-12-20
摘要:当前,使用记忆网络来实现知识库问答(KBQA)系统成为受人关注的焦点,因此本文提出了以基于GatedKey-Value Memory Networks模型为核心的知识库问答系统以验证模型的有效性。该系统主要通过四个步骤来解决中文知识库问答问题:知识库预处理、问题主题词提取、候选答案集合选取和记忆网络答案预测。本文在NLPCC-ICCPOL 2017 KBQA任务组提供的问答数据集上进行了实验,实验结果表明使用GatedKey-Value Memory Networks模型解决KBQA问题是可行且有效的,同时加入门机制可以减少误差在多层模型中的累计,从而提高系统的准确率。
For information in English, please click here
A Knowledge Base Question Answering System based on Gated Key-Value Memory Networks
Abstract:At present, using memory networks to implement knowledge base question answering(KBQA) system has become the focus of attention, so this paper proposes a KBQA system based on Gated Key-Value Memory Networks to verify the validity of the model. The system solves large scale Chinese knowledge base question answering problem through four steps: knowledge base preprocessing, key word extraction of question, candidate answer set selection and answer prediction through memory networks. The experiment is carried out on the question and answer data set provided by NLPCC-ICCPOL 2017 KBQA Task Force. The results show that it is feasible and effective to solve KBQA problem by using Gated Key-Value Memory Networks. At the same time, adding gating mechanism can reduce the accumulation of errors in the multi-layer model, so as to improve the accuracy of the system.
Keywords: Knowledge Base Memory Networks KBQA Gating Mechanism
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于Gated Key-Value Memory Networks的知识库问答系统
评论
全部评论0/1000