摘要
为了学习非结构化文本与对应的结构化语义知识之间的嵌入语义对应关系,本文提出了一种用于自然语言理解(Natural Language Understanding,NLU)的分布式语义向量学习框架.该语义框架使用长短期记忆对输入序列进行编码以生成文本向量,然后将意图标签、时隙标记和时隙值向量合并生成分布式语义向量,通过最小化文本输出向量与语义框架向量的距离,将语义等价向量放置在向量空间中,最后采用意图重构和时隙标签生成损失作为目标得分以学习鲁棒的语义向量.实验结果表明,所学习的语义向量包含语义信息,该语义框架在NLU结果重新排列方面均优于测试的NLU系统.
In order to learn the embedded semantic correspondence between unstructured text and its corresponding structured semantic knowledge,a distributed semantic vector learning framework has been proposed in this paper for natural language understanding(NLU).The semantic framework,with long-term memory,aims at encoding the input sequence to generate the text vector,and then at combining the intention tag,timeslot tag and timeslot value vector to generate the distributed semantic vector.By minimizing the distance between the text output vector and the semantic framework vector,the semantic equivalency vector is placed in the vector space,and finally uses the intention reconstruction and timeslot tag generation loss as the goal score is to learn the robust semantic vector.Experimental results show that the learned semantic vector contains semantic information,and the proposed semantic framework is better than the NLU system in terms of NLU results rearrangement.
作者
李潇雯
朱齐亮
LI Xiao-wen;ZHU Qi-liang(School of Computer and Information Engineering, Shanxi Technology and Business College, Taiyuan 030006, China;School of Information Engineering, North China University of Water Resources and Electric Power, Zhengzhou 450045, China)
出处
《西南师范大学学报(自然科学版)》
CAS
2021年第1期19-24,共6页
Journal of Southwest China Normal University(Natural Science Edition)
基金
教育部产学合作协同育人项目(201901195002)
山西省教育科学“十三五”规划2020年度“互联网+教育”专项课题(HLW-20141)
山西省1331工程立德树人好老师课程建设计划支持人选项目(201832)
山西工商学院2019年“助推计划”专项课题(201953).
关键词
自然语言理解
分布式表示
语义向量学习
语义框架重构
natural language understanding
distributed representation
semantic vector learning
semantic framework reconstruction