基于双向LSTM的人物关系抽取

Character Relationship Extraction Method Based on BiLSTM

为了解决传统人物关系抽取方法中特征选择困难及不全的问题,本文拟提出一种基于双向长短时记忆网络的人物关系抽取方法(Character Relationship Extraction Method based on BiLSTM,CRE-BiLSTM)。首先选取了基于CBOW的Word2vec模型训练词向量,并完成了对互动百科数据集词向量的转化工作。然后使用BiLSTM神经网络获取更为完整的上下文特征信息,从而提取出文本深度词向量特征。最后通过基于Softmax函数分类器对人物关系抽取进行分类。实验结果显示,本文提出的CRE-BiLSTM模型不需要人工定义复杂的特征就可以抽取人物关系且平均F值可达到84.5%。

In order to solve the difficulty and incompleteness of feature selection in traditional character relationship extraction methods,this paper proposes a character relationship extraction model based on BiLSTM(CREBiLSTM).Firstly,the CBOW-based Word2vec model training word vector is selected,and the interactive encyclopedia data set word vector transformation is completed.Then,the more complete text context information is obtained through the bidirectional LSTM neural network to extract the deep word vector features.Finally the character relationship is classified by softmax function classifier and realizes the extraction of the relationship between the five common characters.Experimental results show that CRE-BiLSTM model proposed in this paper does not need to manually define complex features to extract character relationships and the average value can reach 63.3%.