项目名称: 基于关系Markov网的多关系数据聚类分析方法研究
项目编号: No.61202308
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 计算机科学学科
项目作者: 高滢
作者单位: 吉林大学
项目金额: 24万元
中文摘要: 多关系数据的聚类分析作为多关系数据挖掘的重要任务之一,对于分析多关系数据的拓扑结构、理解其功能、发现其隐含模式和预测其行为等具有重要意义,在生物信息学、WWW、社会网、地理信息系统等领域具有广泛应用,但现有方法对大规模数据、复杂关系及不确定环境的处理能力有限。关系Markov网将Markov网与数据库的关系模式相结合,具有出色的多关系表示及不确定性处理能力,成功解决了链接预测、协作分类等任务,但其判别学习的特点妨碍了其在聚类分析中的应用,且其学习效率有待进一步提高。本项目以多关系数据的聚类分析为目的,在发挥关系Markov网优势基础上,对其进行扩展与改进,增强其表达能力,并将其改造成产生式模型;结合智能优化技术研究能提高其学习效率的方法;并进一步从横向、纵向两个角度,提出基于关系Markov网混合模型和基于关系团模板的聚类分析方法。该研究对促进多关系数据挖掘研究具有重要理论意义和应用价值。
中文关键词: 多关系数据挖掘;聚类分析;统计关系模型;参数优化;
英文摘要: Clustering analysis for multi-type relational data is one of the improtant tasks of multi-relational data mining, which is very important for analyzing the topological structures, understanding the functions, recognizing the hidden patterns, and predicting the behaviors of multi-type relational data. It can be widely used in bioinformatics, World Wide Webs, social networks, geographic information system and so on. The current methods of clustering analysis for multi-type relational data have limitions on large scale data, complex and uncertain environment. Recently, with rising of statistical relational learning, a lot of work use statistical relational models to deal with multi-relational learning problem in the complex and uncertain environment, and have achieved good results. Relational markov network is one of representative statistical relational models, which combines markov network and relational schema, and has excellent capabilities of uncertainty processing and expressing complex relationship among data. Relational markov network has been used successfully in link prediction, collaborative classification, information extraction and so on. However, it can not be used for clustering analysis because of its discriminative learning characteristics, and its learning efficiency should be improved further.Thi
英文关键词: Multi-Relational Data Mining;Clustering Analysis;Statistical Relational Models;Parameter Optimization;