|Abstract:||Utilizing scientific literature as a potential source for new knowledge is an extremely attractive idea.
Swanson has repeatedly demonstrated that complementary pairings of ideas from existing knowledge
domains can be applied in the solution of new problems. However, the nature of the information search
used to identify such pairings is fundamentally different from the searches of databases or digital
libraries. Pairings exhibiting direct interrelationships in their literatures are presumably already known to
be related, and to be retrievable via existing techniques. Relationships must be found that would
currently escape detection, such as indirect relationships via some third knowledge domain.
Supercomputers can perform portions of the indirect search process more quickly. They also make
possible, indeed, require, a structurally different approach. The literatures of two research topics are
tentatively assumed to be related. The search for avenues of relatedness may then proceed from "both
ends toward the middle". Reported herein, an attempt along these lines has proven successful in
replicating detection of the indirect relationship between Raynaud's Disease and dietary fish oil, as
originally discovered by Swanson using largely manual techniques.
The supercomputing analytical approach reduces reliance upon human expertise in the early, screening
stages. As such, it can be "scaled up" more readily, enabling at least a cursory statistical examination of
many more pairings of research literatures. Not arguing for the substitution of relatively weak statistical
clues in preference to researcher understanding in the search process, it is possible to employ the
supercomputer as a form of filter or sorter. Research literature pairs exhibiting stronger than normal
statistical ties could be identified, based on the joint occurrence of certain highly ranked tokens. Such
results could be used to suggest research directions, or to help in selecting among choices of research options.