基于Bert模型的框架类型检测方法  

Bert-Based Frame Identification Method

作　　者：高李政 周刚 罗军勇 黄永忠 GAO Lizheng;ZHOU Gang;LUO Junyong;HUANG Yongzhong(State Key Laboratory of Mathematical Engineering and Advanced Computing,Zhengzhou 450001,China;School of Computer Science and Information Security,Guilin University of Electronic Technology,Guilin 541000,China)

机构地区：[1]数学工程与先进计算国家重点实验室,河南郑州450001 [2]桂林电子科技大学计算机与信息安全学院,广西桂林541000

出　　处：《信息工程大学学报》2020年第2期214-220,共7页Journal of Information Engineering University

基　　金：国家自然科学基金资助项目(61602508,61866008)。

摘　　要：FrameNet是一种著名的语义资源,不仅定义完备的框架体系,且提供丰富的标注语料,其数据经常被用于语义角色标注、事件检测、情感分析等NLP任务。受词义消歧(Word Sense Disambiguation,WSD)原理启发,提出一种基于Bert模型的框架类型检测方法。与传统方法只考虑词元上下文信息不同,该方法同时考虑词元的解释性信息。方法利用序列标注模型提取词元的上下文信息;利用语句关系检测模型判断语句与词元定义之间的相关性。在FrameNet框架检测实验中分别与传统的序列标注模型和语句关系检测模型进行了对比。实验结果表明,文章模型性能优于传统模型,从而证明方法的有效性。FrameNet is a famous semantic resource which not only defines a complete system of semantic frames but also provides a rich annotated corpus.Its annotated data are often used in some NLP tasks,such as semantic role labeling,event identification,sentiment analysis,etc.Inspired by the task of word sense disambiguation,we present a novel method of frame identification based on the famous pretrained model named Bert.Different from the studies which only consider the context information,our method also takes the definitions of lexical units into account.We construct a sequence labeling model to extract the context information and a sentence pair classification model to learn the relations between target sentences and lexical units definitions.We compare our models with traditional models on the frame identification task.The experimental results show that the Bert-based models significantly outperform the traditional models,which proves the effectiveness of our method.

关 键 词：FRAMENET 框架检测 预训练模型 语句关系检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]