Academic Research: Recommending similar bugs

[Andre Klapper]

On Sun, 2014-09-14 at 23:20 -0300, H. Rocha wrote:
> The idea is to recommend to a developer who has fixed or is working on a
> bug X that there are similar bugs X1, X2, etc in the tracking system.
> Currently, I am assuming that two bugs are similar if they are assigned to
> the same component and share some similarities among their textual
> descriptions.

I highly appreciate when academia reaches out to developers. 
There seems to be a huge gap between research and getting stuff
implemented upstream.
There's been dozens of research papers (and some prototypes) on
duplicate detection in Bugzilla and I assume you're well aware of them.


Still I'm very curious:

So your basic assumption is that tickets are triaged and end up in the
correct component?  That sounds very similar to Sun et al.'s "Towards
More Accurate Retrieval of Duplicate Bug Reports" at
http://www.comp.nus.edu.sg/~specmine/suncn/papers/ase11.pdf

Did you consider using additional contextual word lists (like Alipour,
Hindle and Stroulia in "A contextual approach towards more accurate
duplicate bug report detection")?

Did you consider giving tickets created in a similar time span more
weight (Prifti, Banerjee and Cukic did that in "Detecting bug duplicate
reports through local references")?

Did you consider giving higher exposure to reports which have a high
number of comments as they likely also receive more duplicates
(Cavalcanti et al. tried that in "The bug report duplication problem: an
exploratory study")?
(All three papers don't have some public access that I'm aware of.)

In any case, that's probably all stuff to only consider once some basic
code actually HAS landed in upstream code.

andre
-- 
Andre Klapper  |  ak-47 at gmx.net
http://blogs.gnome.org/aklapper/

_______________________________________________
dev-apps-bugzilla mailing list
dev-apps-bugzilla at lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-apps-bugzilla