首页|自注意力机制下复杂文本实体关系抽取方法

自注意力机制下复杂文本实体关系抽取方法

扫码查看
为了能够明确语义关系,快速识别所需信息,提出一种自注意力机制下复杂文本实体关系抽取方法.通过复杂文本向量模型把全部词映射为低度实数向量,使文本转变成矢量模式,按照单词的外在情境学习嵌入,可将句子内的所有字转化为嵌入矩阵.利用LSTM网络建立文本向量,存取以往和将来的上下文,融合前后两种输出矢量.运用激活函数压缩单词维度,计算句子上下文本语义贡献权重,在双向LSTM层和输出层间加入自注意力机制,从多角度获取矩阵层次的句子语义,计算组合句子特征矢量在关系上的分数,根据给定概率随机抽样权值参变量,完成复杂文本实体关系抽取.通过实验证明所提方法对复杂文本实体关系抽取效果较好,具有极高的精准度.
A method for Extracting Entity Relations from Complex Texts Based on Self Attention Mechanism
In order to clarify the semantic relationship and quickly identify information,a method of extracting en-tity relation from complex texts under self-attention mechanism was put forward.Firstly,by using a complex text vec-tor model to map all words into low degree real vectors,the text was transformed into a vector pattern.Then,by learn-ing embeddings based on the external context of words,all words within a sentence were transformed into an embed-ding matrix.Moreover,LSTM network was used to create text vectors,for accessing the previous and future contexts,thus fusing the two output vectors.The activation function was used to compress the word dimension and calculate the semantic contribution weight of the upper and lower text.After that,the self-attention mechanism was added between two-way LSTM layer and output layer.Furthermore,the sentence semantics of the matrix was drawn from multiple an-gles.And the score of feature vector of the combined sentence on the relationship was calculated.Finally,the complex text entity relation was extracted by the given parameter variable of probability random sampling weight.The experi-mental results prove that the proposed method has a good effect on the extraction of complex text entity relations as well as high accuracy.

Self-attention mechanismRelation extractionText vector modelComplex text entity

针钰、马晓宁

展开 >

中国民航大学计算机科学与技术学院,天津 300300

自注意力机制 关系抽取 文本向量模型 复杂文本实体

中央高校基本科研业务费专项中国民航大学专项天津市教委科研项目

3122014C0182019KJ127

2024

计算机仿真
中国航天科工集团公司第十七研究所

计算机仿真

CSTPCD
影响因子:0.518
ISSN:1006-9348
年,卷(期):2024.41(4)
  • 15