基于混合机器学习模型的短文本语义相似性度量算法  

Short Text Semantic Similarity Measurement Algorithm Based on Hybrid Machine Learning Model

作　　者：韩开旭 袁淑芳 HAN Kaixu;YUAN Shufang(College of Electronics and Information Engineering,Beibu Gulf University,Qinzhou 535011,Guangxi Zhuang Autonomous Region,China;College of Sciences,Beibu Gulf University,Qinzhou 535011,Guangxi Zhuang Autonomous Region,China)

机构地区：[1]北部湾大学电子与信息工程学院,广西钦州535011 [2]北部湾大学理学院,广西钦州535011

出　　处：《吉林大学学报（理学版）》2023年第4期909-914,共6页Journal of Jilin University:Science Edition

基　　金：国家自然科学基金面上项目(批准号:61374127);广西高校中青年教师科研基础能力提升项目(批准号:2021KY0434;2020KY10019).

摘　　要：为提高短文本语义相似性度量准确性,设计一种基于混合机器学习模型的短文本语义相似性度量算法.先对短文本实施预处理,基于混合机器学习模型构建短文本的字词向量模型,对短文本进行特征扩展;然后组合短文本的多样度量特征,对多样度量特征进行维度规约;最后通过构建一个集成学习模型,计算语义相似性结果,实现语义相似性的度量.使用“Quora Question Pairs”比赛数据集测试该方法的性能,测试结果表明,该方法的准确性较高,对数损失和度量均方差均较低,说明该方法的相似性度量准确性较高.In order to improve the accuracy of short text semantic similarity measurement,we designed a short text semantic similarity measurement algorithm bas ed on a hybrid machine learning model.Firstly,we preprocessed the short text,constructed a word vector model of the short text based on the hybrid machine learni ng model,and extended the features of the short text.Secondly,we combined the various metric features of the short text,implemented dimensional reduction on the various met ric features.Finally,we constructed an ensemble learning model to calculate the semantic similarity results and achieve the semantic similarity measurement.We tested the performance of the method by using the“Quora Question Pairs”competition dataset,the test results show that the accuracy of the meth od is high,the logarithmic loss,and the measurement mean square error are both low,indicating that the similarity measurement accuracy of the method is high.

关 键 词：混合机器学习模型 短文本 文本分词 语义相似性 卡方检验 相似性度量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]