一种基于知识库和SimBert在短文本场景下的实体链接方法  

An entity linking method based on knowledge base and SimBert in short text scenarios

作　　者：吴青华 缪云海 高恩伟 张坤元 WU Qing-hua;MIAO Yun-hai;GAO En-wei;ZHANG Kun-yuan(China Mobile(Hangzhou)Information Technology Co.,Ltd.,Hangzhou 310023,China)

机构地区：[1]中移(杭州)信息技术有限公司,杭州310023

出　　处：《电信工程技术与标准化》2024年第2期59-63,共5页Telecom Engineering Technics and Standardization

摘　　要：随着互联网和人工智能技术的发展,用户输入自由文本越来越不规范,在提升用户搜索结果的准确性和用户体验感方面,实体链接技术在其中发挥了很大的作用。传统的实体链接技术大多只适用于长文本场景,对于短文本场景下的实体链接效果泛化能力差。为了解决短文本场景下的实体链接问题,本文基于知识库构建前缀树进行实体匹配,同时利用SimBert模型进行实体词向量的定制化调整,提升短文本实体链接的准确度,然后分为前缀树构建模块、实体编码模块、实体链接模块3个模块设计实现了一种基于知识库和SimBert的中文短文本实体链接方法。该方法可提升在短文本场景下实体链接的准确性,避免出现实体链接错误,同时也减少了人工参与。With the development of the Internet and artificial intelligence technology,and the increasingly non-standard input of free text by users,entity linking technology has played a significant role in improving the accuracy of user search results and user experience.Most traditional entity linking techniques are only suitable for long text scenarios,and their generalization ability for entity linking in short text scenarios is poor.This article aims to solve the problem of entity linking in short text scenarios.By constructing a prefix tree based on a knowledge base for entity matching,and using the SimBert model to customize and adjust entity word vectors,the accuracy of short text entity linking is improved.Then,a Chinese short text entity linking method based on knowledge base and SimBert is designed and implemented in three modules:prefix tree construction module,entity encoding module,and entity linking module.This method can improve the accuracy of entity linking in short text scenarios,avoid entity linking errors,and also reduce human involvement.

关 键 词：实体 短文本 链接模型 

分 类 号：TN915[电子电信—通信与信息系统]