摘要
针对已有知识树知识热点不突出、知识分类不准确以及结构不断演化等问题,本文面向维基百科的中文数据库"服务计算"领域密集型数据,提出了扩展的中文分词算法,抽取、分类出多种主题知识及其结构化信息,结合服务计算领域文档提出基于LDA改进的DKHM(文档-主题-热点)模型,使用Gibbs抽样算法对数据集采样,并消除原词条歧义分类,以建立演化知识树.实验结果表明:基于DKHM的聚类准确度高于一般的贝叶斯聚类,通过聚类发现的热点与真实热点的匹配度达60%以上,从而验证了演化知识树比维基百科原有知识树结构更合理,热点趋势效果更明显.
Because knowledge tree hotspots are not prominent,knowledge classification is not accurate,and the structure keeps evolving,an extensional algorithm of Chinese segmentation based on Chinese Wikipedia database"service computing"was proposed to extract a variety of themes knowledge and structural information.The evolution knowledge tree was constructed by improved DKHM(Document-Themes-Hotspot Model)based on the research in service computing domain,and using Gibbs sampling,the ambiguity of entry was eliminated.The experiments results showed that the accuracy of DKHM clustering is higher than that of the Bayes algorithm,and the matching rate reach60% by clustering to find hotspot.Thus,the evolution knowledge tree is more reasonable than the original Wikipedia knowledge tree structure and the hotspot trend is more obvious.
出处
《武汉大学学报(理学版)》
CAS
CSCD
北大核心
2015年第4期331-338,共8页
Journal of Wuhan University:Natural Science Edition
基金
国家重点基础研究发展计划(973)(2014CB340404)资助项目