COMBINING ADVANCED VISUALIZATION AND AUTOMATIZED REASONING FOR WEBOMETRICS: A TEST STUDY

Abstract This paper presents a first attempt for performing a preci se and automatic identification of the linking behaviour in a scientific domain through the analysis of the communication of the related academic institutions on the web. The proposed approach is based on the paradigm of multiple viewpoint data analysis (MVDA) than can be fruitfully exploited to highlight relationships between data , like websites, carrying several kinds of description. It uses the MultiSOM clustering and mapping method. The domain that has been chosen for this study is the domain of Computer Science in Germany. The analysis is conduced on a set of 438 websites of this domain using all together, thematic, geographic and linking information. It highlights interesting results concerning both global and local linking behaviour. keywords multiple viewpoint data analysis, clustering, mapping, webometrics 1. Introduction The use of links between academic websites to create an informal mode of scholarly communication is a promising new field of investigation. Hence, thanks to qualitative analyses of links (Wilkinson et al. 2003), the metrics based upon link counts can be seen to be measuring an agglomeration of connections related to scholarly activities in a wide variety of ways. However, thanks to others analyses (Chu 2005), the evaluative link studies should not only consider link counts but also reasons for linking to ensure the validity of such research. Moreover, the linking behaviour has been highlighted as dependant of the studied domain. As an example, (Thelwall et al. 2003) found that the mathematics and computer sciences domains are more interlinked than other scientific domains. Several recent works have gone one step further, by demonstrating that a precise analysis of linking behaviour depends on the capability of taking into account several complementary factors, like discipline and geographic factors (Thelwall 2002). Nevert heless, performing such a kind of analysis at a large scale prohibits the use of a manual mode, as it has been proposed up to now.