面向篇章的框架关系预测的表示学习  

REPRESENTATION LEARNING OF TEXT-ORIENTED FRAME RELATIONSHIP PREDICTION

作　　者：王燕 吕国英[1] 李茹[1,2,3] 任国华 Wang Yan;LüGuoying;Li Ru;Ren Guohua(College of Computer and Information Technology,Shanxi University,Taiyuan 030006,Shanxi,China;Key Laboratory of Computer Intelligence and Chinese Information Processing of Ministry of Education,Shanxi University,Taiyuan 030006,Shanxi,China;Collaborative Innovation Center of Big Data Mining and Intelligent Technology in Shanxi,Taiyuan 030006,Shanxi,China)

机构地区：[1]山西大学计算机与信息技术学院,山西太原030006 [2]山西大学计算智能与中文信息处理教育部重点实验室,山西太原030006 [3]山西省大数据挖掘与智能技术协同创新中心,山西太原030006

出　　处：《计算机应用与软件》2023年第4期128-133,210,共7页Computer Applications and Software

基　　金：国家社会科学基金项目(18BYY009)。

摘　　要：框架关系可以描述框架与框架之间的语义关系,通过对篇章构建框架关系图发现图中存在孤立框架,一个篇章表达的是一个语义整体,框架间关系缺失阻碍了篇章句子之间建立联系。针对该问题,进行面向篇章的框架关系预测方法研究,分别使用WSABIE算法、Word2vec方法和TransE方法训练得到框架表示作为关系预测的输入;使用余弦相似度方法和以Hing-loss函数为优化目标的神经网络方法进行实验,并在神经网络中融入框架本身的属性信息——框架定义。实验结果验证了基于知识图谱表示学习方法(TransE方法)的优越性以及框架定义信息的有效性,提升了框架关系预测的性能。Frame relationship can describe the semantic relationship between frames.By constructing a frame relationship diagram for a text,it is found that there are isolated frames in the diagram,and a text expresses a semantic whole.The lack of the relationship between frames hinders the establishment of links between text sentences.To solve this problem,the text-oriented framework relationship prediction is studied.The WSABIE algorithm,Word2vec method and TransE method were used to train the framework representation as the input of relationship prediction.The cosine similarity method and neural network method with Hing-loss function as the optimization goal were used for experiments,and the attribute information of the framework itself-framework definition was integrated into the neural network.Experimental results show the superiority of knowledge map representation-based learning method(TransE method)and the effectiveness of frame definition information,which improves the performance of frame relationship prediction.

关 键 词：框架关系 框架表示 框架定义 框架关系预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]