基于自注意力卷积神经网络的实体关系抽取  

Entity relation extraction based on self-attention convolutional neural network

作　　者：张婷婷 李卫疆[1] 李涛[1] ZHANG Ting-ting;LI Wei-jiang;LI Tao(School of Information Engineering and Automation,Kunming University of Technology,Kunming 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650500

出　　处：《信息技术》2022年第1期11-15,共5页Information Technology

基　　金：国家自然科学基金资助项目(62066022)。

摘　　要：在自然语言处理解领域中,实体关系抽取作为信息抽取中的一个重要分支,旨在从自然文本中提取出两个实体之间的语义关系。大多数研究工作都是基于NLP系统的特征,特征提取工程和预处理过程十分冗杂,并且由LTP工具提取出来的特征会在模型中迭代而产生错误传播。为了避免对NLP系统的滥用,提出一种基于端到端的自注意力卷积神经网络模型来提取实体对之间的语义关系。实验结果表明,该方法在SemEval-2010 Task 8数据集上的F_(1)值提高了约1.3%。In the field of natural language processing, entity relation extraction has been seen as an important branch of information extraction which aims at extracting semantic relation between two entities from natural texts. In the past, most research works were based on the features of NLP systems. However, the feature extraction engineering and pre-processing processes are very tedious and the features extracted by LTP tools can be iterated in the model and produce error propagation. To avoid the abuse of NLP systems, we propose an end-to-end self-attentive convolutional neural network-based model to extract semantic relationships between entity pairs. Experiment results show that the method improves the F_(1)value by about 1.3% on the SemEval-2010 Task 8 data set.

关 键 词：信息抽取 关系抽取 自注意力 卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]