Sciences, technologies et visualisations

Kahane Bernard, Mogoutov Andrei, Cointet Jean-Philippe, Villard Lionel, Larédo Philippe, “A dynamic query to delineate emergent science and technology: the case of nano science and technology”, In Villard Lionel, Revollo Michel, Laredo Philippe, Content and technical structure of the Nano S&T Dynamics Infrastructure, RISIS, 2015, pp. 47-70 http://risis.eu/wp-content/uploads/2015/03/Report-Task1-Nano.pdf Building a larger and relevant database out of an initial seed without relying, because of potential bias, on experts is a common challenge for those who wish to study or track a scientific or technological field. Publications and patents are not the only, but definitely an important component of knowledge generation and dissemination and one of the potential sources for innovation. Scientists communicate their findings through publications. Similarly, patents are legal documents to claim ownership of an invention but they also build a public paper trail of technology advancement. Thus publications and patents are an important, relevant and useful tool to follow and represent results of scientific and technological endeavours (Huang, 2010). Data mining is the extraction of relevant and useful information from large volume of data. Publication and Patent data systematically collected in worldwide databases such as the WoS and Patstat are used to track science and technology dynamic. Data mining faces an important challenge in a context of emergence when new technologies experience explosive growth, evolve rapidly and often cross and subvert existing scientific and technology fields. Emerging science and technology (biotechnology in the 1980s, nanotechnology today, other science and technology fields tomorrow), which often carry strong implications and potentialities for science, business and society, add to the challenge. Their content and dynamic are difficult to track at a time when they are struggling to define who they are, what they include and exclude and how they organize themselves internally. Such is the case for nanotechnology, where the quest for a relevant reliable and replicable way to extract relevant publications and patents, is an on-going process involving several teams worldwide (Glanzel 2003, Noyons 2003, Mogoutov and Kahane, 2007, Porter et al., 2008, Kostoff 2007, Leydesdorff and Zhou, 2007). Nanotechnology is a rapidly evolving emerging and dynamic field. Analysts argue that it is likely to be a “general purpose technology” (Youtie 2008, Laredo et al. 2010) with a potential impact across an entire range of industries and great implications on human health, the environment, sustainability and national security. The perceived potential value of nanotechnologies has led to the increased will of governments, academic institutions, firms and other societal actors to better understand what is happening in the field, who is active and where. There is thus an important challenge to develop robust methods to track the nanotechnology field while it rapidly develops and evolves. As a matter of fact, good quality and comprehensive extraction of data is a prerequisite for meaningful understanding and analysis. Huang 2010 as well as L'huillery et al. 2010 have compared the different methodologies developed, and reported on their robustness as well as on the similarities and discrepancies of results obtained. They confirmed the robustness and interest of the evolutionary lexical methodology we have developed (Mogoutov and Kahane, 2007). At that time, three requirements were central to the approach developed. First, it should not depend upon experts. Indeed, the on-going and extensive use of expert-based approaches is costly, time-consuming, and challenging to replicate such that the same outcomes result. This is an important restriction when facing a highly dynamic field where borders are constantly evolving requiring terminology requalification at different times. Second, it should allow updates in order to replicate and compare results while the nanotechnology field (and its lexicon) develop and expand. And third, it should be able to track the relative evolution of subfields inside nanotechnologies: in 2007 we translated this into a third requirement of being “modular”. While the initial development of our methodology was performed in order to extract data from 1998 to 2006, we later engaged in producing an update that could expand the database backward and forward in order to cover years 1991-2011. In our initial methodology, the selection of relevant terms was performed with knowledge built and keywords selected on one single year (2003). A simple solution was to reproduce the selection of terms for 2011, driving us to two semantic universes of nanotechnology, respectively built in 2003 and 2011. However Bonaccorsi (2010) has demonstrated that in a dynamic field such as nanotechnology, keywords often display short life and experience a type of Darwinian selection process. Using this approach, the characterisation of the evolution of the field over 20 years would have only relied on two years for the identification of relevant keywords. There would thus be a risk that we miss the richness of the exploration that shapes the dynamics of knowledge production. Not considering transient keywords that might have emerged and then disappeared, would be a serious drawback in such a dynamic field. There are multiple reasons for this. Two are of particular importance. One is about the learning that a stream of research, even if it goes on with a life of its own, has been experimented but proved not to be useful for colleagues at the time. The other lies in the fact that streams of research which for a while turn to be a dead end, can nevertheless reappear later and become a key resource as demonstrated in many instances. Such a limitation becomes even more visible when taking the whole period under review for identifying relevant keywords. This drove us to add a fourth requirement for such an approach: What is needed is a methodology, which allows us to incorporate and discard in real time relevant terms as they appear and disappear in the nanotechnology story. We need a methodology that allows us to track keywords as characters appear and disappear along the storyline in a movie. Thus, using nanotechnology as a showcase, we here report a data search strategy made of three consecutive steps. As in all the data search strategies for nanotechnology, we start with an initial seed built through the nanostring. We then use the same principle that we applied in our previous approach, that is expanding the initial seed through a dual process where additional keywords observed during a given period are sorted according to their internal specificity (e.g. the extent to which they provide value added meaning to a publication) and then tested in the overall database for ‘external specificity’ (e.g. the ratio of articles in the seed vs. articles in the overall database of publications). This selection of keywords is first applied on the whole dataset covering the 20 years, enabling a “static extension”. The third step builds the “dynamic extension” where additional keywords are identified through a yearly analysis of internal specificity within the nanostring, and selected depending upon their ‘external specificity’. Besides being applied in a specific way for nanotechnology, we claim that such a three steps strategy has universal value to describe the dynamics of emergent and fast evolving fields, transcending pre-existing classifications.