پیش بینی فن آوری های در حال ظهور با کمک داده کاوی مبتنی بر متن: یک رویکرد خرد

Text data mining should be useful for anticipating new technologies and new uses for existing technologies, insofar as one can attempt to connect complementary pieces of information across two different domains, or subsets, of the scientific literature. The present study attempted to predict genetic engineering technologies that may impact on viral warfare in the future. The analysis was carried out using a combination of conventional Medline searches and the package of advanced informatics techniques known collectively as Arrowsmith. The findings strongly indicate that genetic packaging technologies such as DEAE-dextran, cationic liposomes and cyclodextrins are plausible candidates to enhance infections caused by viruses delivered via an aerosol route — despite the fact that no studies have yet been reported that have examined this issue directly, and certainly not in the contexts of viral disease or viral warfare. The critical factor was the overall strategy of approaching the problem: first, to define two specific fields explicitly (in this case, genetic engineering and viral warfare) that are hypothesized to contain complementary information; second, to identify common factors that bridge the two disciplines (i.e. research on viruses); and third, to progressively shape the query once initial findings are obtained. Thus, in contrast to some current perceptions, the process of text data mining is neither automatic nor is it restricted to those who have access to macro analyses using customized computer systems.

Technological innovation often proceeds by applying advances made in one field to a separate arena. Once the innovation is implemented, the transfer of knowledge may appear obvious or even inevitable, but without the benefit of hindsight it is surprisingly difficult to identify specific technologies that are ripe for transfer. One must simultaneously identify a need in one domain and a tool in another, possibly quite disparate domain that potentially satisfies that need — and such a task requires more than expert knowledge. A large body of research knowledge is published in the form of papers and technical reports that are accessible via bibliographic databases, leading several workers to advocate the development of techniques for knowledge discovery in databases (Fayyad and Uthurusamy, 1999), and in particular, strategies for text data mining (Swanson and Smalheiser, 1997, Hearst, 1999 and Kostoff, 1999), in order to ‘discover’ useful knowledge that is implicit within the published record. Anticipating new technologies and new uses for existing technologies should be ideal applications for text data mining, insofar as one can attempt to connect complementary pieces of information across two different domains, or subsets, of the scientific literature, that may not have been noticed by workers beforehand. Text data mining strategies can be divided into two types, macro and micro. Macro analyses perform data-crunching operations over a large, often global set of papers encompassing one or more fields, in order to identify large-scale trends or to classify and organize the literature. Several examples of macro analyses have been published by ourselves and others (Swanson and Smalheiser, 1997 and Kostoff, 1999). In contrast, micro analyses pose a sharply focused question, in which one searches for complementary information that links two small, pre-specified fields of inquiry. We have previously shown the value of this micro approach in helping to formulate and assess hypotheses arising in biomedical research (Smalheiser and Swanson, 1998a and Smalheiser and Swanson, 1998b), and in the present paper, it is demonstrated how the micro approach can be employed for helping make policy decisions regarding technical innovation. Genetic engineering technologies have the capability to alter the make-up of biological organisms and thus have the potential to impact on the way that nations may conduct, and hopefully may defend against, the threat of biological warfare (BW). To anticipate possible threats that may be developing, one needs to learn what relevant genetic research is being done around the world — not only research that is explicitly intended for military applications, but also research being conducted in medical, biotechnological, public health, agricultural or zoological contexts that might be potentially applied to BW applications in the future. This is a task for military intelligence, but it can be difficult to distinguish research intended for BW from that directed toward, for example, vaccine development or gene therapy, and intelligence officers need to prioritize which kinds of genetic research are most in need of being tracked. The task is made even more difficult by the multiplicity of BW scenarios that must be considered — for example, whereas battlefield deployment of BW agents would necessarily induce acute, fulminant disease that incapacitates troops, a terrorist threat might well involve dissemination of agents that induce chronic rather than acute symptoms. Taking the complementary approach of using informatics to predict emerging genetic engineering technologies that may impact on BW, specifically viral warfare, the question is “Given the state of published research right now, what BW applications are possible?” whether or not there is evidence that anyone is actually exploring those avenues. The analysis was carried out using a combination of conventional Medline searches and the package of advanced informatics techniques known collectively as Arrowsmith (Swanson and Smalheiser, 1997), which seek to find meaningful relationships between two largely disparate literatures or fields of inquiry — in this case, genetic engineering vs. viral warfare. Furthermore, the focus was on research findings that were so strong and consistent that they were reflected directly in the titles of papers, although the abstracts and text of key papers were also assessed when relevant.

The analysis presented here indicates strongly that genetic packaging technologies such as DEAE-dextran, cationic liposomes and cyclodextrins are plausible candidates to enhance viral infections via an aerosol route, acting by one or more steps — despite the fact that no studies have yet been reported that have examined this issue directly, and certainly not in the contexts of viral disease or viral warfare. Thus, our analysis was not simply an exercise in summarizing the known state of the art, but rather gleaned previously unexamined ‘nuggets’ of potential technology transfer from a mountain of raw data scattered in the current literature (Hearst, 1999) — without the need for large-scale computing, and without relying on expert knowledge of the fields in question. Although Arrowsmith software was employed to help define and juxtapose the two literatures in question, these advanced programs were not essential, and indeed we found in this case that the same outcome was obtained using conventional Medline searching techniques alone. The critical factor was the overall strategy of approaching the problem: first, to define two specific fields explicitly (in this case, genetic engineering and BW) that are hypothesized to contain complementary information; second, to identify common factors that bridge the two disciplines (i.e. research on viruses); and third, to progressively shape the query once initial findings are obtained. Thus, in contrast to some current perceptions, the process of text data mining is neither automatic nor is it restricted to those who have access to customized computer systems. Nevertheless, given the broad scope of the field of genetic engineering, it is likely that the present example is but one among many other examples of potential technology transfer from the field of genetic engineering to the field of BW. To gain a more comprehensive and systematic overview of this issue it will be necessary to undertake a macro analysis of the two literatures, e.g. using Arrowsmith to identify a series of items that link the two literatures as a whole, followed by micro analyses to examine these items individually (Swanson and Smalheiser, 1997). Thus, the full power of text data mining is probably best captured by coupling macro analyses with micro analyses.