A useful level of analysis for the study of innovation may be what we call ‘‘knowledge communities’’—intellectually cohesive, organic inter-organizational forms. Formal organizations like firms are excellent at promoting cooperation, but knowledge communities are superior at fostering collaboration—the most important process in innovation. Rather than focusing on what encourages performance in formal organizations, we study what characteristics encourage aggregate superior performance in informal knowledge communities in computer science. Specifically, we explore the way knowledge communities both draw on past knowledge, as seen in citations, and use rhetoric, as found in writing, to seek a basis for differential success. We find that when using knowledge successful knowledge communities draw from a broad range of sources and are extremely flexible in changing and adapting. In marked contrast, when using rhetoric successful knowledge communities tend to use very similar vocabularies and language that does not move or adapt over time and is not unique or esoteric compared to the vocabulary of other communities. A better understanding of how inter-organizational collaborative network structures encourage innovation is important to understanding what drives innovation and how to promote it. Keywords Knowledge communities Innovation Dynamic clustering
Technical and scientific innovation is accomplished through a joint effort of thousands of researchers working for different kinds of organizations, including firms, universities, hospitals, research think tanks, and government labs (Fleming and Sorenson 2001; McCain 1987; Osareh 1996; Small 2003). In some cases the work of a group of researchers across firms assumes a joint coherency, and the group begins to function, in many ways, as if it were a new emergent organizational form (Crane 1972; Kuhn 1962). These scientific ‘‘knowledge communities’’ comprise an inter-organizational large-scale network in which researchers work together by building on each other’s advances. Researchers in knowledge communities tend to produce at higher rates of innovation than less cohesive researchers (Boyack and Borner 2003; Merton 1972; Murray and Stern 2005; Narin et al. 1997). In the context we study in this paper, for example—technical computer science paper publications— a disproportionate 56.61% of citations are received by papers in cohesive knowledge communities, even though only 43.67% of papers are in such communities.
Using citation and rhetorical data to look at the network typology of these innovating communities, we begin to isolate key structural factors that drive their success. We attempt to quantify some of the substantive differences between knowledge communities in both use of previous knowledge and use of rhetoric. First we examine community cohesiveness, or the extent to which communities build on each other’s knowledge and language.Nextwe examine community uniqueness, or the extent towhich a knowledge community is different from others in its use of past knowledge and language. Finally weexamine community flexibility, or the rate of change a knowledge community has shown over time in use of knowledge and language.
Modern analysis in innovation has largely focused on the study of formal organizations, usually firms. But how researchers communicate and learn between these organizations is where much of the value is generated (Crane 1972, 1989; Fleming and Sorenson 2001; Kuhn 1962; Small 2003). Knowledge communities can exist in any specialized areas of research where there is free exchange of information; examples include looking for a cure for Alzheimer’s disease, trying to improve Internet search, and looking to improve the gas efficiency of diesel engines (Culnan 1986; Guimera et al. 2005; Hargens 2000; Small 1994). Knowledge communities are by their very nature homogenous, unifying people of similar research interests and specialties to learn from each other and build on each other’s ideas each for their own—though similar—purposes (Merton 1972). Members of knowledge communities can work together as closely as the members of most firms, yet may never meet (Crane 1989). While firms are excellent at facilitating cooperation by unifying incentives, knowledge communities are superb at encouraging collaboration; this embodies the distinction between working together and working jointly. An interactive knowledge community offers a built-in and knowledgeable audience for research, a stimulating intellectual dialogue, and an accelerated technical environment (Crane 1972, 1989; Kuhn 1962). Being a member of a knowledge community is less a conscious choice than a reflection of being a part of a stream of the knowledge that involves the collaboration and cooperation of other researchers in a tight, cohesive pattern of research.
We draw on two emerging areas of network research that have examined innovation in large-scale networks, though with different emphases: research on small worlds, and research on geographic technology clusters.
In his work on ‘‘Small Worlds’’ (Watts 1999), Duncan Watts observed that the aggregate structure of connections between people and other networks was not random. These ideas have begun to allow network theorists to effectively grapple with largescale networks of all sorts, including, for example, the small world of the casts of Broadway musicals (Guimera et al. 2005; Uzzi 2005). This research shows the benefits behind cohesive communities for research engaged in collaborative innovation. Michael Porter and others have argued for a different logic behind such communities of expertise. They study geographic clusters of competence, showing how a bubble of geographically close, dense connections, high expert knowledge, and unified interests has led to sustained advantages in innovation and core competencies (Porter 1998; Porter et al. 2004). This strain of thought assumes that communities are geographical, like Silicon Valley in the 1990s—and that it is the social reinforcement, and to some sense external dislocation, of people into the ‘‘bubble’’ that allows for such effective innovation in business.
Both the small world and Porter’s lines of research integrate well with the arguments developed within the sociology of knowledge since the 1960s. Merton (1972), Kuhn (1962), and Price (1963) looked at the realm of science and research and argued that paradigms—lenses for viewing the world, frameworks of meaning—created tight clusters of researchers who built on each other’s ideas. The stronger the paradigm, the more intellectually coordinated the cluster could be, since it had more clearly defined questions, methodologies, and language (Pfeffer 1993; Yoels 1974). On the other hand, maladjusted and restrictively strong paradigms can lead to myopia, since higher commitment to a framework more strongly excludes other frameworks (Meyer and Zucker 1989; Pfeffer 1993).
Research within the scientific academy and within firms, traditionally studied separately, has begun to be more integrated as university researchers seek more patents, academic and corporate researchers fruitfully collaborate, and firms such as Google emerge from the ivory tower (Henderson et al. 1998; Murray 2005; Murray and Stern 2005; Narin et al. 1997; Shane 2002). Nevertheless, fundamental and proprietary research settings rest on substantively distinct incentive structures (Gittelman and Kogut 2003; Murray and Stern 2005). Fundamental science encourages the broad distribution and sharing of knowledge, since the success of university researchers is bound up with the attention of and usefulness to their fellow researchers. The importance of research within firms, while it builds on fundamental science (and sometimes contributes to it), rests on the ability to extract value from proprietary, and thus private, knowledge.