基于上下文嵌入和叠加注意力的机器阅读理解OA
Machine Reading Comprehension Based on Context Embedding and Superimposed Attention
自然语言处理研究方向之一机器阅读理解,其目的是提升计算机对文本内容的阅读并理解的能力.由于之前经典模型并没有考虑到长期语境依赖与一词多义现象,单一的注意力机制也不能充分表达文本的意思.根据上述问题,论文提出了一种算法模型,该算法在经典模型基础上,在嵌入层通过理解上下文提高了词嵌入的准确性,对一词多义问题的准确理解有一定的提升,通过叠加计算增强了问题到文本和文本到问题之间的相关性,完善了注意力表达的文本含义.在SQuAD数据集上,实验结果表明,该模型对比基线模型性能上有显著提高.
Machine reading comprehension,one of the research directions of natural language processing,aims to improve the ability of computers to read and understand text content.Because the previous classical models do not consider long-term con-text dependence and polysemy,a single attention mechanism cannot fully express the meaning of the text.According to the above problems,this paper proposes an algorithm model,which improves the accuracy of word embedding by understanding the context in the embedding layer on top of the previous classical model.The accurate understanding of polysemy has a certain improvement,and the correlation between the question to the text and the text to the question is enhanced by superposition calculation,and the text meaning of attention expression is improved.On SQuAD dataset,the experimental results show that the performance of the model is significantly improved compared with the baseline model.
何青山;尹祎
武汉科技大学计算机科学与技术学院 武汉 430065武汉科技大学智能信息处理与实时工业系统湖北省重点实验室 武汉 430065
数理科学
机器阅读理解ELMO注意力机制叠加注意力
machine reading comprehensionELMOattention mechanismsuperimposed attention mechanism
《计算机与数字工程》 2026 (1)
23-27,5
湖北省教育厅科学研究计划指导性项目(编号:B2022002)资助.
评论