Testing Reading Tactics for Automated Reading Assistance: Is it Useful to Apply Old Knowledge?
Zhe Yu, Tim Menzies · arXiv (Cornell University) · 2017
Given the growing number of new publications appearing everyday, literature reviews are important for software engineering researches to stay up-to-date with their field. A state-of-the-art text mining method (FASTREAD) supports SE researchers in selecting literature of their interests (this approach uses an active learner to suggest what small number of papers, out of many thousands of candidates, might be worth reading). While FASTREAD allows researchers to skim 90% fewer papers in fresh new literature reviews, we find two common scenarios where knowledge from old literature reviews can further enhance new one: i.e. a) researchers updating their old literature reviews; b) researchers initializing new literature reviews on topics similar to those of an old review. This paper tested two novel UPDATE and REUSE extensions to FASTREAD. While UPDATE was proved to be quite straight-forward (just importing data from old review to bootstrap FASTREAD), REUSE was much more complex and required additional architecture. Overall, we find that UPDATE, or our augmented REUSE tactic, allows researchers to skim an additional 20% to 50% fewer papers by applying old knowledge in the corresponding scenarios. Hence, we strongly recommend sharing and applying old knowledge for new literature reviews.