Social network analysis is the study of the relationships among social entities, and on the patterns, structures, and implications of these relationships. We are building a Web-based software system for social network analysis. The motivation is to provide a web site where users can analyze their collections of empirical data. Our analyses are currently based on the decomposition of a social network into strongly connected components as the constituting groups, and on the computation and visualization of network characteristics. We use Bayesian belief networks to model the composition of individual attributes (e.g., gender and other socio-metrics) in groups, as well as peer influence (as in a group) upon those attributes.
Much of this work is still in progress. For a preliminary summary, however, please take a look at "The Two Webs: Towards a Web-based System for Social Network Analysis", a paper presented at 3rd Workshop on Information Technology and Social Transformation (Academia Sinica, Taipei, Taiwan, December 20-21, 1999). Currently, the Web-based system is used to analyze adolescent friendship networks from a dataset collected by Chyi-in Wu at the Institute of Sociology, Academia Sinica. The dataset is taken from a multi-year survey on 1,434 junior high students (7th - 9th grade). The students are from 44 classes in 33 Taipei city junior high schools. Students in the 44 classes was surveyed in 1996, 1997, and 1998. One of question in the survey asked each of the student to name three of their best friends. We used the students' answers to build social networks: 3 networks for each class (one for each of 1996, 1997, and 1998), and for all 44 classes. This results in 3 X 44 = 132 adolescent friendship networks.
The entire collection of the 132 networks, as well as a Bayesian model of group influence upon individual attributes, can be accessed by the following Java Applet.
The social networks are represented with carefully chosen visual elements so one can easily grasp the structure of the networks, as well as explore its details. For example, the global view of a network clearly shows the relative sizes of the major groups in a class, and their differences in link density. The detailed view of the network makes clear the existence of social cliques within groups. We use oval-shaped nodes for female students, and rectangle-shaped nodes for male students. This makes easy to discover the fact that networks in many co-ed classes have gender-based social groups. (We will describe rationales for other visual elements in the illustration of networks, when we find more time.)
We use the Graphviz package developed at AT&T Research Labs for graphic layout of networks. The Bayesian model is built upon the JavaBayes package developed by Fabio Gagliardi Cozman at CMU.
For more illustrations of social networks in online communities, click here.
--- Tyng-Ruey Chuang, Biam Chee Low, and Chyi-in Wu.
Last updated by Tyng-Ruey Chuang on June 27, 2000.