Optimising Parameters for ASKNet: A Large Scale Semantic Knowledge Network Creation System
Brian E. Harrington, Simon Kempner · 2011
Abstract—ASKNet is a system for automatically constructing semantic knowledge networks from natural language text. ASKNet uses existing natural language processing tools to extract entities and relations from text, and then through a combination of lexical information and a novel use of spread-ing activation, combines that information into a large scale semantic knowledge network. The ASKNet system is large, and quite complex. Historically, users of the system have had to rely on a combination of intuition and empirical evaluation of small sample networks in order to obtain reasonable settings for the various system parameters. In this paper, we develop a testing harness and gold standard that allow us to use simple machine learning methods to find optimal settings for all of the system’s parameters. This system also aids future development of internal system algorithms, and can be adapted easily to novel domains.