The Innovation Genome (2022-2026)

The Innovation Genome (2022-2026). Shuai Wang (PhD candidate) – supervisory team: Monika Kackovic (FEB), Stevan Rudinac (FEB), Nachoem Wijnberg (FEB), Marceel Worring (FNWI).

(2021) Data Science Centre Interdisciplinary Program merit grant of €335,000 [Granted].

The focus of this project is gaining a fine-grained understanding of the role of influence relations in determining the likelihood of producing (successful) innovations. Not only understanding the features of successful innovations but also the complex flows of information, ideas, and practices surrounding such innovations is paramount to developing proactive approaches for intelligent solutions. We embrace this grand challenge by taking a two-tiered approach to investigating innovations vis-à-vis patterns of influence among individuals and within and across categories to (1) study the effects on competitive dynamics and socio-economic performance over time and (2) identify universal ‘secret ingredients’ successful innovations share. Our goal is to analyze and map the so-called innovation genome in the cultural industries (visual arts) and science by extracting, modeling, and interactively visualizing complex traits (e.g., inventions) and patterns of influence. Innovations can be understood to be inventions that become influential and successful. In a general way, inventions can be said to occur by means of re-combining existing knowledge and resources, while innovations occur during experimentation with inventions, ideas, practices, etc. For example, in the cultural industries, there is a conjecture that the stylistic innovation of (post) Impressionism was strongly influenced by artistic inventions of Japanese woodcuts, or in science, there is an underlying hypothesis that mathematical theorizing invented in physics is a core tenet of neoclassical economics.

To systematically test these and other widespread suppositions, we use categorization as our fundamental theoretical lens. We focus on analyzing individual and categorical dynamics for two distinct yet interrelated reasons. First, we want to know why some innovations become influential and/or socio-economically successful while most do not. What are the commonalities or ‘secret ingredients’ present in successful innovations compared to those that fail? Second, we systematically distinguish and compare patterns of influence, which enable us to analyze the extent to which particular individual innovators (“great man theory of history”) and/or categorical developments are necessary to explain particular innovations or whether they could easily be replaced by someone or something else. We aim to develop multimodal geometric deep learning approaches, informed by state-of-the-art business theories, to model actors, innovations, and categories of interest into a joint

semantic space. We adopt a visual analytics paradigm for studying the data, deploying our novel, interactive visualization, and interactive (human-in-the-loop) learning tools together with advanced network analysis and time series modeling approaches. The empirical domains we analyze are science and visual art, both offering large and accessible databases. For instance, publicly available databases of scholarly publications and patents, such as Scopus, Thomson Reuters, and Google Scholar, and large, domain-specific collections of multimedia data, such as WikiArt [9]. These data are suitable to the aims of this project because they not only lend themselves very well to the tools of text, image, and metadata analysis but also are domains in which innovativeness is considered the prime determinant of success, and in which success - in terms of competitive performance - can be readily measured.