Knowledge Representation in the Big Data Age

Enormous amounts of data are now at our fingertips. It is produced with breathtaking speed, while at the same time it differs widely in formats, perspectives, usefulness, reliability. One of the foremost research challenges in Computer Science is how to make non-trivial use of this wealth of data, and it seems obvious that the development of impact methods and tools will require cooperative efforts which span many sub-disciplines of Computer Science, while working hand-in-hand with application areas. Facing this challenge, KR needs to redefine itself as an application-oriented discipline. The following are three corresponding challenges which seem to be of central importance. Closing the gap between deduction and induction. Current KR methods work at their best if data and knowledge bases are hand-crafted or hand-curated and adhere to a single modeling perspective. But this approach does not scale up to Big Data requirements, neither in terms of volume nor in terms of noise present in the data, which may come from errors or from different modeling perspectives. At the same time, inductive methods like those based on statistical machine learning or data mining are good at filtering out noise, generally perform better under higher data volume, and are designed to detect higher-level features in data—however they lack deductive reasoning abilities and largely remain black boxes. A successful dealing with Big Data necessitates that we have the best of both worlds available, which means that we have to establish methods which bridge the gap between logic