KBC: The Knowledge Build Complexity

Abstract

A knowledge-based complexity measure is introduced for assessing the complexity of a given entity. At its core, the measure relies on reducing the entity to a sum of human knowledge inputs. A knowledge baseline is first established, below which the knowledge levels are considered negligible. A complexity tree is then derived, with complexity nodes defined by an `is a component'-type relationship, and each branch terminating when it arrives at the knowledge baseline. Different trees may be partially compared by summing the complexity along all branches, or, more fully, by considering the tree structure and individual complexities at each node. The paper concludes with an example in practical use, a routine which receives an entity name and performs a basic recursive knowledge search with a specialized web crawler on the Wikipedia corpus. To help determine relevance for recursion, the NLP BERT machine learning model is used for relation extraction.