基于权重计算的中文因果关系分析

Chinese causality analysis based on weight calculation

提出一种中文因果关系分析方法,以便更加细腻地表达因果关系.该方法由因果关系提取和权重计算组成.首先,构建了中文因果关系四元组数据集,将因果划分为核心名词和谓语状态,即原因中的核心名词、原因中的谓语或状态、结果中的核心名词、结果中的谓语和状态;然后,构建了中文因果关系抽取(CCE)模型,该模型由中文预训练的基于全词掩码训练的双向编码表示模型(BERT-wwm)和条件随机场(CRF)组成,在所构建的数据集上,四元组抽取F1分数为0.3;最后,提出基于因果强度的近似原因权重算法,用于计算同一结果不同原因的权重,减小对语料库数据量的依赖性,具有更好的鲁棒性和泛化性,能更加真实地反映不同原因对结果的重要程度.

A Chinese causality analysis method was proposed to express causality more delicately.The method consists of causality extraction and weight calculation.Firstly,the Chinese causal relationship quaternary data set were constructed and divide causality into core nouns and predicate states,that is,the core nouns in the cause and the result,the predicate or state in the cause and the result.Secondly,the Chinese causality extraction(CCE)model was constructed which was composed of a Chinese pretrained bidirectional encoder representations from transformers with whole word masking(BERT-wwm)and the conditional random field(CRF).The F1 values for the four-tuple is 0.3 in our dataset.Finally,an approximate cause weighting algorithm based on the strength of causality was proposed to calculate the weight of different causes for the same result,which reduces the dependence on the amount of corpus data,has better robustness and generalization,and can reflect the importance of different reasons to the result more truly.