基于混合CNN的自然语言因果关系抽取方法

NATURAL LANGUAGE CAUSALITY EXTRACTION METHOD BASED ON MIXED CNN

针对现有自然语言因果关系抽取模型的低准确性和自由参数过多等问题,提出高效的面向知识通道和面向数据通道相结合的混合卷积神经网络(Mixed Convolution Neural Network,MCNN)方法。面向知识通道结合词汇知识库中因果关系的语言知识,自动生成的卷积过滤器以捕获因果关系的重要语言线索。实验表明与SingleCNN和MultiCNN相比,该方法在因果关系提取方面效果更优。

To solve the problem of low accuracy and excessive free parameters of existing natural language causality extraction models,an efficient mixed convolution neural network(MCNN)is proposed,which combines knowledge-oriented channel with data-oriented channel.Knowledge-oriented channel was combined with the language knowledge of causality in the lexical knowledge base,and the convolution filter was automatically generated to capture the important language clues of causality.The experimental results show that compared with SingleCNN and MultiCNN,this method has better effect than other two traditional CNN models on causal relationship extraction.