摘要
提出基于聚类及链接分析的挖掘模型LinkNetClus，该模型将对象类型分为目标类型及属性类型，并假设目标对象属于每个簇的概率依赖于与之相关的其他对象，在目标对象上进行迭代的聚类操作，最终得到具有多样性的聚类结果。该模型充分利用了异质信息网络中的关联关系，得到多维的挖掘结果来解决数据冗余的问题，结果的可解释性也优于排序序列。通过实验证明了使用LinkNetClus得到的聚类结果比已有的基准方法提高大概30％～50％左右。
In this paper, the author proposes a mining model named LinkNetClus based on clustering and link analysis.Firstly, a star network is constructed, which has attribute objects and target objects. Assume that the generation probability of a target object is based on these associated attribute objects. Then these target objects will be clustered and get diversity results.The LinkNetClus model also uses the homogeneous and heterogeneous relationship between objects, and gets multidimensional ranking result to eliminate redundant information.Compared with the two -phrase ranking result, the result made by LinkNetClus is more understandable.Compared with the existing base method, experiments show that the mining results based on LinkNetClus model is enhanced as many as 30%-50%.
出处
《工业仪表与自动化装置》
2014年第6期11-14,46,共5页
Industrial Instrumentation & Automation
关键词
异质信息网络
聚类算法
链接分析
排序模型
多样性挖掘技术
heterogeneous information network
clustering algorithm
link analysis
ranking model
diversity data mining
