网络嵌入在计算社会科学中的发展与应用

Development and application of network embedding in computational social science

网络嵌入利于计算存储,可将数据对象表示成稠密、实值、低维向量,被广泛运用在社会计算中对象间的语义计算等问题上,以支持个性推荐、知识计算、社会网络分析等应用.基于网络嵌入的分布式表示学习模型可以习得词汇间的隐含关系,发现词汇语义层级,建立跨语言词汇表示,从而检测词汇语义变迁.本文在针对网络嵌入技术发展介绍的基础上,对基于随机游走、结合外部信息、基于深度学习、以及基于异构信息网络的四大类典型算法进行对比与分析,并讨论了网络嵌入在计算社会科学的不同应用场景中的实现过程与意义.

Data object which is conducive to computational storage can be represented as dense,real value,low-dimensional vectors by network embedding. It has been widely used in semantic computation among objects in social computing to support the personalized recommendation,knowledge computing,social network analysis,and so on. The distributed representation learning model based on network embedding can acquire the implicit relationships among words,discover the semantic hierarchy of words,establish the cross-language vocabulary representation,and detect the semantic changes of words. Four kinds of typical algorithms based on random walk,combining external information,deep learning,and heterogeneous information networks are compared and analyzed by this paper based on the introduction of the development of network embedding technology. The procedure of implementation and significance of network embedding in different application scenarios of computational social science are also be discussed.