一种改进的针对合著关系网络的链接预测方法_郭景峰.pdf

2008V ol1 35 l 1 12 * ) ( 燕山大学信息科学与工程学院 秦皇岛066004) 主要针对那些实体类标号属性未知的社会网络进行链接预测由于实体的类标号属性与具体的社会网络有 , 因此具体解决对作者之间合著 系网络图的链接预测问题首先, 给出了合著 系图的结构表示, 然后把一个作 者是否是多产的定义为合著 系图中作者实体的类标号属性另外, 还提出了一种改进的利用有指导学习进行链接 预测的方法在改进的链接预测方法中为每对作者新引入了一个特征属性) ) ) 是否至少有 一个是多产的当所要预 测的合著 系图中作者实体的类标号属性不完全已知时, 用改进后的I CCL P 算法对合著 系进行预测, 以提高链接 预测的性能改进后的ICCLP 算法中采用上面提到的改进后的链接预测方法 链接预测, 类标号属性, ICCL P , 合著, 多产的 Improved Link Prediction Method for Co- authorship Network GU O Jing- eng W AN G Chun- y an Z OU Xiao- hong Z HA O Peng- ei ZH A N G J ian ( College o In orm at ion Science and Engineerin g, Y ansh an U ni versit y , Q inh uan gdao 066004, C hina) Abstract T his paper specially predicted links in the social net w o rk w here the class labels o the objects are unknow n. Since that the in or mation o the o bj ects. class labels is r elated to concrete social netw or k , it specially so lves the pro b- lem o link pr ediction or co- autho rship netw or k . A t irst , the structure o the co- autho rship netw o rk w as or mulated. T hen the class labels o the author objects in this netw or k w ere de ined as either proli ic o r not . And o ne impr oved su- perv ised learning method o r predict co- author ship links w as proposed . In our superv ised learning method , o ne eature ( either o r both pro li ic) that r e er s to the class labels o the author objects w as increased. When the class labels o the autho r objects ar e unkno w n , the impr ov ed ICCL P w hich u ses the impr oved supervised learning o r link prediction is used . T he ex per iment resul