摘要
在基于知识图谱的智能问答系统中,属性映射模块结果的错误传播会导致最终无法得到正确答案,对此提出了一种基于多注意力多维文本的属性映射方法。首先通过拆分问题文本及结合属性信息得到多维文本表示;其次使用长短期记忆网络(long-short-term memory,LSTM)层生成各自的隐层表示;然后输入多注意力机制层后使问句和属性之间的关系及语义信息更加完善,利用属性之间的交互信息及多种角度来加强问句语义信息的理解;最后通过卷积神经网络(convolutional neural networks,CNN)提取局部特征并且采用softmax分类器实现属性映射。试验结果表明,在自然语言处理与中文计算会议(NLPCC 2018)中知识库问答(KBQA)任务所提供的开源数据集上,本方法相比主流属性映射模型其性能有显著提升,准确率最高提升6.62%。本模型可以补足单一文本表示与注意力机制的短板,有效解决属性映射模块中语义歧义的问题,这有助于后续提高智能问答系统的整体性能。
In the intelligent question answering system based on knowledge graph,the error propagation of the result of attribute mapping module can lead to the inability to get the correct answer ultimately,hence an attribute mapping method based on multi-attention and multi-dimensional text came to the rescue.Firstly,the multi-dimensional text representation was obtained by splitting the question text and combining the attribute information;secondly,the long-term and short-term memory(long-short-term memory,LSTM)networks were used to generate their respective hidden layer representations,and then the multi-attention mechanism layer was input to improve the relationship and semantic information between questions and attributes,and the interactive information between attributes and various angles were used to enhance the understanding of semantic information of questions;finally,local features were extracted by virtue of convolution neural network(convolutional neural networks,CNN)and attribute mapping was realized with the aid of softmax classifier.The experimental results show that on the open source data set provided by the knowledge base question and answer(KBQA)task in the natural language processing and Chinese computing conference(NLPCC 2018),the performance of this method is significantly improved compared with the mainstream attribute mapping model,with the accuracy up to 6.62%higher.This model can make up for the deficiency of single text representation and attention mechanism and effectively solve the problem of semantic ambiguity in attribute mapping module,which is conducive to improving the overall performance of intelligent question answering system.
作者
叶仕超
雷景生
杨胜英
YE Shichao;LEI Jingsheng;YANG Shengying(School of Information and Electronic Engineering,Zhejiang University of Science and Technology,Hangzhou 310023,Zhejiang,China)
出处
《浙江科技学院学报》
CAS
2022年第5期435-443,共9页
Journal of Zhejiang University of Science and Technology
基金
国家自然科学基金项目(61972357,61672337)
浙江省重点研发计划项目(2019C03135)。
关键词
智能问答
属性映射
多注意力
多维文本
知识图谱
intelligent question answering
attribute mapping
multi-attention
multi-dimensional text
knowledge graph
