语义文本相似度计算方法研究综述  

A Review of Semantic Text Similarity Calculation Methods

作　　者：李莹 伍胜 徐聪 尹刚 张锦 LI Ying;WU Sheng;XU Cong;YIN Gang;ZHANG Jin(School of Computer and Communication Engineering,Changsha University of Science and Technology,Changsha 410076,China;EduCoder Teaching and Research Center,Changsha 410205,China)

机构地区：[1]长沙理工大学计算机与通信工程学院,湖南长沙410076 [2]头歌教学研究中心,湖南长沙410205

出　　处：《软件导刊》2024年第11期1-11,共11页Software Guide

基　　金：湖南省自然科学基金项目(2021JJ30456,2021JJ30734);工业控制技术国家重点实验室开放研究项目(ICT2022B60);国防科技重点实验室基金项目(2021-KJWPDL-17);国家自然科学基金项目(61972055)。

摘　　要：语义文本相似度计算是自然语言处理领域一个关键任务,旨在衡量两个文本之间的语义相似程度。对以往经典和当前主流的语义文本相似度计算方法进行归纳和总结,将这些方法划分为传统的方法和基于深度学习的方法两大类。传统的方法又划分为基于字面匹配、基于统计和基于规则的方法。基于深度学习的方法又划分为基于词嵌入、基于句向量和基于预训练模型的方法。在进一步细分每个类别的基础上,详细介绍了各子类的典型方法,并对各种方法的基本思想、优点和局限性进行了深入分析和总结。最后,对语义文本相似度计算方法可能的发展方向进行了展望。Semantic text similarity calculation is a key task in the field of natural language processing,which aims to measure the degree of se⁃mantic similarity between two texts.Based on the summary of the traditional and current mainstream semantic text similarity calculation meth⁃ods,these methods are divided into traditional methods and deep learning-based methods.The traditional methods are divided into literal matching,statistics and rule-based methods.The methods based on deep learning are further divided into the methods based on word embed⁃ding,sentence vector and pre-trained model.On the basis of further subdivision of each category,the typical methods of each subcategory are introduced in detail,and the basic ideas,advantages and limitations of each method are deeply analyzed and summarized.Finally,the possi⁃ble development direction of semantic text similarity calculation is prospected.

关 键 词：文本相似度 语义相似度 自然语言处理 深度学习 预训练模型 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]