Title: What Can Heterogeneity Add to the Scientometric Map? Steps towards algorithmic historiography
ArXiv ID: 1002.0532
Date: 2010-02-03
Authors: ** - Loet Leydesdorff (주요 저자, 과학계량·네트워크 분석 분야의 선구자) - 본 논문은 Michel Callon의 연구 업적을 기리는 Festschrift에 실린 기고문 형태로 발표되었습니다. **
📝 Abstract
The Actor Network represents heterogeneous entities as actants (Callon et al., 1983; 1986). Although computer programs for the visualization of social networks increasingly allow us to represent heterogeneity in a network using different shapes and colors for the visualization, hitherto this possibility has scarcely been exploited (Mogoutov et al., 2008). In this contribution to the Festschrift, I study the question of what heterogeneity can add specifically to the visualization of a network. How does an integrated network improve on the one-dimensional ones (such as co-word and co-author maps)? The oeuvre of Michel Callon is used as the case materials, that is, his 65 papers which can be retrieved from the (Social) Science Citation Index since 1975.
💡 Deep Analysis
📄 Full Content
The Actor Network represents heterogeneous entities as actants (Callon et al., 1983;1986). Although computer programs for the visualization of social networks increasingly allow us to represent heterogeneity in a network using different shapes and colors for the visualization, hitherto this possibility has scarcely been exploited (Mogoutov et al., 2008).
In this contribution to the Festschrift, I study the question of what heterogeneity can add specifically to the visualization of a network. How does an integrated network improve on the one-dimensional ones (such as co-word and co-author maps)? The oeuvre of Michel Callon is used as the case materials, that is, his 65 papers which can be retrieved from the (Social) Science Citation Index since 1975. 2
My methods are standard and straightforward. Author names, the names of the respective journals, the titles, the references, etc., can all be attributed to documents as units of analysis. I construct a matrix with the 65 retrieved documents as the cases, 48 unique coauthors of Callon as variables, and the 27 words which occurred more than twice in the titles of these documents as another set of variables.
The papers appeared in 26 journals during the period 1975 -2009. 3 These journal names are added as a third set of variables. The number of variables therefore is (48 + 27 + 26 =) 101. The matrix is normalized in terms of co-occurrences among the variables using the cosine for the similarity (Ahlgren et al., 2003;Leydesdorff, 2008). The figures are drawn using the spring-based algorithm of Kamada and Kawai (1989) as available in Pajek for the visualization. 4 The size of the nodes is in proportion to the logarithm of the frequency of occurrence in the data.
Figure 1 provides a full representation including the three relevant sets of variables. Let us first compare this extremely rich representation with the co-author map of these same papers in Figure 2 and the co-word map in Figure 3. (Leydesdorff, 1989). The structure in the data is therefore less pronounced than with citation or co-authorship relations. One may need additional (statistical) analysis to distinguish the groupings clearly. In Figure 3, the three main factors are circled for the sake of clarification. Words in these three components correspond to three of Callon’s main research interests. However, the three factors explain only 31.1% of the variance contained in the datamatrix.
Both the co-author and co-word maps thus are relatively uninformed when compared with the integrated map in Figure 1, with the journals also added. One needs additional information-for example, from factor analysis-to understand the structure of the semantic map. The co-author map is easier to understand in terms of institutional affiliations, but this perspective is not informative without local knowledge about the cognitive agendas which motivated these authors to collaborate.
The static representations cannot teach us anything about the evolution of the research trajectory of the author. Figure 4 provides the breakdown of Figure 1 in six periods of five years, that is, 1975-2005. Recently, these figures can also be animated using, for example, the dynamic version of Visone (at http://www.leydesdorff.net/callon/animation;5
cf. Leydesdorff & Schank, 2008). 1975-1980 1980-1985 1985-1990 1990-1995 1995-2000 2000-2005
In summary, Michel Callon was right when he hypothesized that one has to combine the information contained in the various maps in order to obtain a meaningful and rich representation. Author names contribute to the semiosis in actor networks. Social and cognitive structures are interwoven into textual domains. Unlike social network analysis, with its main focus on agents, scientometrics is interested not only in the social structures but also in understanding the semantic map (Callon et al., 1993). Conversely, the cognitive constructs (e.g., clusters of words) can inform the appreciation of social relations. Adding the journals further enriches this map as any other relevant category might do (e.g., institutional affiliations). Further interpretation may increasingly lead to the development of algorithmic historiography (Garfield et al., 2003) as a field which
Callon and his colleagues (1983 and 1986) have envisaged.