As the computer and the Internet became increasingly important tools for social interaction and information exchange among people, the Journal of Computer-Mediated Communication published a special issue whose theme was “Studying the Net” in 1997 (see http://www.ascusc.org/jcmc/vol3/issue1/). In that issue, Garton, Haythornthwaite and Wellman (1997) and Jackson (1997) suggested that the methods of social network analysis could be applicable to understand the interplay between computer-mediated communication (CMC) processes. In particular, Jackson (1997) argued that hyperlink-based social network analysis could be a strong approach for studying the representation and interpretation of the Web's communication structure.
The theoretical framework of hyperlink network analysis is based on the use and application of traditional social network analysis, which studies the relations that exist among people, organizations, and nation-states (Wasserman & Faust, 1994; Wellman & Berkowitz, 1989). Social relations are generally arranged based on exchanges among social actors. The contents of exchanges can be visible or intangible, and include manufacturing goods, knowledge, political power, citation, social support, media content, or information. The exchange takes on the patterns or regularities that could not be found if social members are analyzed individually. In other words, exchange relationships among members of a social system can be represented as networks-sets of ties describing their interconnections. In the field of communication, social network analysis examines the relationships among a social system's components (generally the individual) based on the stable patterns of use of the communication system (consisting of channel/media, message, and symbol) (Monge & Contractor, 2000; Richards & Barnett, 1993; Rogers & Kincaid, 1981).
With the introduction of computer technologies into society, several researchers have examined CMC networks among computer conference users (Danowski & Edison-Swift, 1985; Rice, 1982, 1994; Rice & Barnett, 1986). Following this approach, Paccagnella (1998) recently used the social network approach to examine the structural communication patterns among people using Italian cyberpunk computer conference systems such as a bulletin boards. Marc Smith (1999c) and others (Kang & Choi, 1999; Paolillo, 2001) have analyzed the network of text message flows among Internet users. In relation to a scholarly setting, some studies (Koku, Nazer, & Wellman, 2001; Haythornwaite & Wellman, 1998; Matzat, 2001; Walsh & Maloney, 2002) have employed network analysis to examine the pattern of communication relations and media use among researchers. A number of factors were examined: Working and social relations such as scientific productivity, the frequency of collaborative communication, the information exchange relationships, and the types of communication technologies used. Also, a series of studies conducted by Haythornthwaite (2000, see http://alexia.lis.uiuc.edu/~haythorn/) and the Human-Computer Interaction group of Cornell University (Gay et al., 2001) have investigated online communication networks among students from the perspective of social networks. In contrast to these studies, some research (Hampton & Wellman, 2000; Matei & Ball-Rokeach, 2001) extended the role of CMC networks to offline life. They examined online social ties among people in relation to social interactions in their offline world.
As claimed earlier, the Web forms an important CMC network that may contain social networks (Wellman, 2001). As social members start to use hyperlinks to create and maintain their personal or organizational ties online as well as offline, a social network connected by hyperlinks becomes a part of CMC networks. In other words, a hyperlink network can be described as a specific type of CMC network, in which Web site authors are interconnected by hyperlinks (Park, in press).
A hyperlink is a technological capability that enables, in principle, one specific Web site to connect seamlessly with another. The shared (bilateral or unilateral) hyperlinks among Web sites allow documents and pictures to be referred to through the Web. The information or contents may be ‘transmitted’ through the simple click of a mouse (Pirolli & Card, 1999). A hyperlink between two Web sites functionally brings two sites closer together. While any individual and organization have complete freedom in choosing the selection of hyperlinks on their Web sites, hyperlink structures are likely to be designed, sustained, or modified by Web site creators to reflect their communicative choices and agendas (Jackson, 1997; Park, 2002). That which binds together the nodes of the Web, Web sites, can be social networks as well as technological components (Kling, 2000). From this perspective, we can potentially discern fingerprints of social relations through the analysis of configurations of hyperlink interconnections among Web sites that represent a social system's components such as people, private companies, public organizations, cities, or nation-states.
Hyperlink network analysis uses a set of analytical techniques and tools (e.g., density, centrality, cluster analysis, block modelling, and multi-dimensional scaling) derived from social network analysis. The difference between hyperlink and traditional network analysis is the use of hyperlink data that can be obtained only from Web sites. The basic hyperlink network data set is an n x n matrix S (also called a 1-mode network matrix), where n equals the number of nodes in the analysis. In hyperlink network analysis, the nodes are Web sites that represent social actors such as people, groups, organizations, cities, or nation-states. Each cell, sij, indicates the absence or presence or the frequency of the hyperlinks among nodes i and j. For example, sij could be a zero or a one depending on whether there is hyperlink between node i and j. Also, the strength of the relationship can be expressed if each cell represents how many hyperlinks exist between two nodes. S is symmetrical (sij= sij) when one is not concerned with directionality of the hyperlinks. In those instances when the source and receiver of the information are differentiated, S is asymmetrical (sij≠ sij). Alternatively, a 2-mode hyperlink network matrix, an m x n, may be made. In this matrix, the rows usually represent the types of hyperlinking Web sites and the columns the contents of hyperlinks. A 2-mode matrix can be converted to a 1-mode matrix depending on the research questions of interest. In the new converted matrix, a value in the ijth cell is the number of hyperlinks for which both Web site i and j contain the same content.
One of the important outcomes of hyperlink network analysis is to identify a central node, in this case, a central Web site, generally defined as the site that provides the most and/or shortest connections to other members within the group (Scott, 1991; Wasserman & Faust, 1994). The central Web site usually plays the role of hub, broker, and authoritative or prestigious site. There exist various centrality measures. Bonacich's eigenvector centrality is often used as a global indicator in hyperlink network analysis. It is appropriate in those instances where the hyperlink network is symmetrically interconnected and the frequencies of the hyperlink connections among Web sites are not binary and are relatively dense (Bonacich & Lloyd, 2001). However, this measure provides an inadequate description of a directional (or asymmetrical) network. As a result, the directional hyperlinks can be analysed using Freeman's degree centrality. It measures the number of a Web site's direct hyperlink connections with others in the group (Freeman, 1979). Freeman's measure consists of incoming and outgoing degree centrality. In hyperlink network analysis, indegree centrality is calculated based on the number of hyperlinks a Web site receives from the other sites, while outdegree centralityis determined with the number of hyperlinks originating from a site. Besides these values, there are Freeman's closeness and betweenness centrality measures (Freeman, 1979). Closeness centrality is used to determine which Web site has the shortest path to all others in the group (Freeman, 1979). Betweenness centrality refers to the frequency with which a Web site falls between pairs of other sites in the group and represents the potential for control of communication, as a broker or a gatekeeper (Freeman, 1979). Finally, Richards' Negopy centrality is the mean number of hyperlink connections required to reach each of the other Web sites in a group, such that the lower the value the more central the site (Richards, 1995). The majority of group Web sites, such as the websites of members of a university department, are connected to the central site so that Internet users can navigate with fewer links when going through it.
There are other procedures frequently used in hyperlink network analysis. With centrality indicators, centralization and density are useful to examine overall characteristics of the network. Centralization means the extent to which a hyperlink network is organized around its central Web sites (Scott, 1991; Wasserman & Faust, 1994). Density indicates the overall level of network integration (Freeman, 1979; Wasserman & Faust, 1994). While centralization is the proportion of other Web sites' connections with a central site, density reflects how sites are connected to one another in an entire network. Next, a cluster analysis identifies those groupings of Web sites that best represent their hyperlinked relations, producing central and periphery groups in terms of density within each cluster (Aldenderfer & Blashfield, 1984). While cluster analysis sorts Web sites into discrete clusters based on the existence of (in)direct connections among themselves, block modelling discovers sites with similar positions together (Burt, 1992). Web sites in the same block disclose similar patterns of hyperlinkage connections to others rather than that they are linked to each other. Multi-dimensional scaling methods such as a correspondence analysis (Barnett, 1993; Torgerson, 1958) are able to reveal the positions that nodes occupy in space. A matrix of hyperlink connectivity is converted to two- or three-dimensional coordinates and a graphic representation, that is, a map, is drawn. A quadratic assignment procedure (QAP) examines the association between two networks or different attributes of the same network (Krackhardt & Porter, 1986). Since it is able to test whether two data matrices are similar to each other, it is often used to assess whether a hyperlink network among Web sites indicates other relations among the creators of sites.