http://arxiv.org/abs/1702.03324
The Proposal Auto-Categorizer and Manager (PACMan) tool was written to respond to concerns on subjective flaws and potential biases in some aspects of the proposal review process for time allocation for the {\it Hubble Space Telescope} (HST), and to partially alleviate some of the anticipated additional workload from the {\it James Webb Space Telescope} (JWST) proposal review. PACMan is essentially a mixed-method Naive Bayesian spam filtering routine, with multiple pools representing scientific categories, that utilizes the Robinson method for combining token (or word) probabilities. PACMan was trained to make similar programmatic decisions in science category sorting, panelist selection, and proposal-to-panelists assignments to those made by individuals and committees in the Science Policies Group (SPG) at Space Telescope Science Institute. Based on training from the previous cycle’s proposals, PACMan made the same science category assignments for proposals in Cycle 24 as did the SPG, an average of 87\% of the time. Tests for similar science categorizations, based on training using proposals from additional cycles, show that this accuracy can be further improved, to the $>95\%$ level. This tool will be used to augment or replace key functions in the TAC review processes in future HST and JWST cycles.
L. Strolger, S. Porter, J. Lagerstrom, et. al.
Tue, 14 Feb 17
22/71
Comments: 9 pages, 7 figures
You must be logged in to post a comment.