A Unified Approach for Taxonomy-Based Technology Forecasting

Motivation

In order to clarify the intended use of our system, it must be stressed that we are not using “forecasting” in the sense of weather forecasting, where future outcomes are predicted with a reasonably high degree of certainty. It is also important to note that certain tasks remain better suited to human experts. For example, where a particular technology of interest has already been chosen, we believe that a traditional literature review would prove superior to an automated approach. Instead, the proposed framework targets the preliminary stages of technology management, where breadth rather than depth is emphasized. The main focus of our system is on analyzing broad trends occurring in a very large number of documents or other textual sources. By scanning and digesting large amounts of information, promising but less obvious developments can be detected and subsequently brought to the attention of a human expert. This way we capitalize on the strength of computational approaches before making more efficient use of valuable expert time in the critical latter stages of the decision making process.

Knowledge that facilitates forecasting the likely growth and consequences of emergent technologies is essential for well-informed technology management, which is currently relying largely on expert opinion. However, expert decisions can be influenced by personal perspectives or biases. Moreover, acquiring and analyzing such knowledge is hampered by the vast amount of data available in publications. Consequently, sifting through the—often electronically—available R&D literature is time consuming, yet non-exhaustive and subjective. In order to cope with this problem, automated forecasting techniques have been developed in recent years (see Background section). A remaining challenge is related to the knowledge organization of the acquired data. For example, in order to elucidate the advances of technologies, we want to answer questions like: “How many scientific articles have been published in peer-reviewed journals on the topic of solar energy recently?” Intelligent search techniques capable of grouping semantically similar concepts are therefore needed, such that the term ”parabolic trough” is subsumed under solar energy related technologies and hence articles about it should be included in the analysis. This underlying challenge of managing and structuring the vast amount of available knowledge from web resources is similar in web-based Technology Forecasting and general Semantic Web applications. However the former has yet to fully benefit from the advances of the latter. In particular, the state-of-the-art Technology Forecasting tools hardly make use of ontologies or taxonomies, the standard form of knowledge representation for the Semantic Web (Shadbolt, 2006).

The major novel aspect in the presented work is a modular and automated approach, which streamlines data acquisition, keyword extraction and taxonomy creation as the basis for trend detection. The framework provides evidence for growing technologies to decision makers in a logically structured manner. We therefore believe that it carries an enormous business potential within the realm of technology management.