the growth and behaviors of the future Web, as well as to engineer systems with desired properties in a way that is significantly less hit or miss.

 

from Power Laws to People Mathematically based analysis of the Web involves another potential failing. Whereas the structure and use of various Web sites (taken mathematically) may have interesting properties, these properties may not be very useful in explaining the behavior of the sites over time. Consider the following example: Wikipedia ( www.wikipedia.org), the

the linguistic content of its pages. The figure shows the same kind of Zipf-like distribution found in the original Web graph analyses. There is also some evidence¹⁶ and a lot of speculation²⁹ that similar effects can be seen in the use of tags in Web-based tagging systems. Current research is also exploring whether these results depart from such models as preferential attachment3 used to explain the scale-free features of Web graphs.

Unfortunately, whatever explains these effects, another aspect of Wikipedia’s use is not explained by these

 

Figure 3: Results of an analysis of the link structure of Wikipedia with respect to the use of link labels, not the linguistic content of pages.

+Predicate occurrence distribution

1

0.1

0.01

P(k)

0.001

0.0001

1e-05

1

10

100

1000 10000

k

100000 1e+06

1e+07

 

online wiki-based encyclopedia, includes more than two million articles in English and more than six million in all languages combined. They are hyperlinked, and it is logical to ask whether the hyperlinks have structure similar to those on the Web in general or whether, since this is a managed corpus, they have yet other properties.

Answering can be done in a number of ways; Figure 3 shows the result of one of them. In this case, DBPedia ( dbpedia.org), which is a dump of the link structure of Wikipedia using the labeled links of the resource description framework, or RDF, has been analyzed with respect to the use of the link labels; that is, we are looking at the structure of Wikipedia as opposed to

models and does not necessarily follow from these properties. Wikipedia is built on top of the Media Wiki software package ( www.mediawiki.org/wiki/Me-dia Wiki), which is freely available and used in many other Web applications besides Wikipedia. While some of them have also been successful, many have failed to generate significant use. A purely “technological” explanation cannot account for this; rather, something about the organizational structures of Wikipedia and the needs of its users accounts for its success over other systems built from the same code base. The model by which articles are created, edited, and tracked is provided by the underlying technology. The social model enabled by humans interacting

in ways allowed by that technology is more difficult to explain. The dynamics of any “social machine” are highly complex, and dozens of academic papers, from multiple disciplines, have been written about it; en.wikipedia. org/wiki/Wikipedia:Wikipedia_in_aca-demic_studies uses Wikipedia itself to maintain an up-to-date reference list.

The idea of a social machine was introduced in Weaving the Web, ⁸ which hypothesized that the architectural design of the Web would allow developers, and thus end users, to use computer technology to help provide the management function for social systems as they were realized online. The social machine includes the underlying technology (mediaWiki in the case of Wikipedia) but also the rules, policies, and organizational structures used to manage the technology. Examples abound on the Web today. Consider the coupling of the application design of blogging-support systems (such as LiveJournal and WordPress) with the social mechanisms provided by blog-rolls, permalinks, and trackbacks that have led to the so-called blogosphere. Similarly, the protocols used by social networking sites like MySpace and Facebook have much in common, but the success or failure of the sites hinges on the rules, policies, and user communities they support. Given that the success or failure of Web technologies often seems to rely on these social features, the ability to engineer successful applications requires a better understanding of the features and functions of the social aspects of the systems.^b

Today’s interactive applications are very early social machines, limited by the fact that they are largely isolated from one another. We hypothesize that (i) there are forms of social machine that will someday be significantly more effective than those we have today; (ii) that different social processes interlink in society and therefore must be interlinked on the Web; and (iii) that they are unlikely to be developed through a single deliberate effort in a single proj-

b. When we say “success” or “failure,” we are referring not to the business factors that determine whether, for example, Facebook or MySpace will attract more users but to the success or failure of the sites to provide the particular types of social interaction for which they are designed.