基于增量聚类的电子政务短文本信息挖掘算法研究  

Research on mining short text information in E-government based on incremental clustering

作　　者：冷泳林[1] 郭颖 孙晓红[2] 曲珮漪 LENG Yonglin;GUO Ying;SUN Xiaohong;QU Peiyi(College of Information Science and Technology,Bohai University,Jinzhou 121013,China;School of Management,Bohai University,Jinzhou 121013,China)

机构地区：[1]渤海大学信息科学与技术学院,辽宁锦州121013 [2]渤海大学管理学院,辽宁锦州121013

出　　处：《渤海大学学报（自然科学版）》2023年第3期262-269,共8页Journal of Bohai University:Natural Science Edition

基　　金：辽宁省社会科学基金项目(No:L14AGL002,No:L13AGL002).

摘　　要：电子政务平台每天都会产生大量短文本数据,挖掘短文本数据对政府掌握民意有十分重要的作用.针对短文本信息量少,单一短文本向量表示模型产生的特征信息丢失问题,提出一种融合权重及主题特征的混合向量表示模型.该模型利用Word2vec和TF-IDF算法挖掘短文本的局部特征,利用BTM主题模型挖掘短文本全局特征,然后将两种特征向量进行连接构成短文本向量.针对短文本数据增量变化特征,通过增加限定阈值改进传统Single-Pass聚类算法,实现短文本的增量聚类.实验结果表明,该模型能够有效的提高短文本聚类效果.The E-government platforms generate a large amount of short texts every day.Clustering short texts play a very important role for the government's control of public opinion.This paper proposes a hybrid vector representation model that integrates weights and topic features to address the problem of feature information loss caused by a single short text vector representation model with limited information content.This model utilizes Word2vec and TF-IDF algorithms to mine local features of short texts,and utilizes BTM topic model to mine global features of short texts.Then,the two feature vectors are connected to form a short text vector.For the incremental changes of short texts,the Single-Pass clustering algorithm is improved by adding limited thresholds to achieve incremental clustering of short texts.The experimental results show that the hybrid vector representation model proposed in this paper can effectively improve clustering performance.

关 键 词：电子政务 短文本 向量表示模型 增量聚类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]