Page 60 - Technology and Innovation Journal - 19-1

398 DEMPWOLF & SHNEIDERMAN

these dynamic and complex adaptive systems has Why Innovation Is Hard to Measure and How
become an important priority within science policy Event Analytics Can Help
and scientometrics (2). A streamlined definition of innovation is the
Our modeling of research and development activ- process of working on marketplace problems that
ities enriches the prevailing network approach with prompts innovators to transform ideas and scientific
event analytics by focusing on time-stamped point knowledge into new products (broadly defined to
events, such as getting a patent, or interval events, include services). The innovation process connects
such as the funding period covered by a grant or marketplace problems with research events; how-
contract. ever, each product follows a unique path involving
We see STI processes as comprising sequences of different types of activities, including research, pub-
point and interval events that together result in the lication, invention, prototyping, proof-of-concept,
translation of knowledge and research into new prod- and several commercialization events culminating
ucts and services in the marketplace. (The phrases in a new product launch. The trajectory a product
“products” or “products in the marketplace” are con- takes may involve multiple events within any stage
strued broadly throughout this paper to include all and may involve revisiting a prior stage if remedial
types of innovation and all types of “marketplaces,” work is required. Thus, the first difficulty in measur-
including public domain.) Point events are associated ing innovation is the unique and variable nature of
with a single date and time (e.g., the date of a patent
application), while interval events are associated with the innovation trajectory or sequence of events for
start and end dates and times. Research projects or each product.
research grants with start and end dates are examples A second difficulty is that early-stage research
of interval events. These events generally fall into one events are often undertaken for the purposes of
of several categories, including research, invention, knowledge creation and publication. In fact, the
proof, and several types of commercialization events. explicit innovation goal of a new product may not
(The order of activities here generally follows the yet exist. There is a temptation to define the distinc-
linear model of innovation. This ordering is primarily tions among science, technology, and innovation
a matter of convenience and should not be construed more rigidly, but this creates as many problems as it
as proffering any particular model or theory of STI solves. The creative moment when the product is first
processes.) Each event is associated with a document envisioned involves a specific set of conditions that
or record that describes the event, the key people and are a function of the sequence and characteristics of
organizations involved and what roles they played, events up to that point. It is as if the innovation path
when and where the event occurred, along with other suddenly appears midway through the journey.
attributes. The information from these records, espe- Mathematically, this describes a Markov chain or
cially dates, may be used to model event networks of Bayesian network model, in which each event in the
people, organizations, places, and documents. sequence is influenced by the cumulative effect of
Events that contribute to the development of everything that has happened up to that point. Nei-
specific products and services may be associated ther the final destination nor the intermediate events
with each other, creating product and service event can be known with certainty. They may, however, be
sequences or trajectories. The trajectories may be estimated based on certain probability distributions.
connected through the networks of people, organi- Modeling and analyzing innovation event trajecto-
zations, places, and documents involved and through ries for successful products a posteriori establishes
their contributions to specific product and service the basis for estimating those baseline probability
event sequences. Conceptually, this dual modeling distributions. This, in turn, allows the formulation
structure (innovation networks and innovation event and testing of more sophisticated hypotheses. It may
trajectories) provides a linkage between STI as com- also allow the development of predictive models or
plex adaptive systems and STI as complex processes. facilitate machine learning and the development of
related big data applications. Finally, the goal would
be prescriptive modeling that would enable policy