Label Prediction on Dynamic Social Networks with Concept Drifting

• Main Authors: Yu, Hsiang-Min, 游詳閔
• Other Authors: Shan, Man-Kwan
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/sp455r

碩士 === 國立政治大學 === 資訊科學系 === 099 === Label prediction is one of the central questions of social network research. The core of label prediction is the use of labeled nodes to predict labels of un-labeled nodes in a social network. The definition of a labeled social network is a social network of partial or complete labeled nodes. The nodes in the same social network have a mutual impact on each other’s labels. Previous research on label prediction have been focused on static social networks. However, social networks are more dynamic in reality. In a dynamic social network, the links of nodes, even the labels of nodes, can be changed with time. The mutual influence of nodes can also be changed. The changing is called “Concept Drift.” This thesis predicts the labels on a dynamic labeled social work. We address the problems of classification for a dynamic social network. The technique of label prediction on static social networks and algorithms used to tackle concept drift are combined to solve the label prediction problem on dynamic social networks. Experiments were performed on a labeled social network constructed from the Internet Movie Database. The results show that we can use the evolution of dynamic social networks to generate a more precise prediction of labels.