Bugzilla – Full Text Bug Listing
| Summary: | [discussion] automatically infer interest of newly created tasks | ||
|---|---|---|---|
| Product: | z_Archived | Reporter: | Eugene Kuleshov <ekuleshov> |
| Component: | Mylyn | Assignee: | Mik Kersten <mik.kersten> |
| Status: | RESOLVED WONTFIX | QA Contact: | |
| Severity: | enhancement | ||
| Priority: | P4 | CC: | eclipse.dserodio, janvik, mlists, murphy |
| Version: | unspecified | Keywords: | helpwanted |
| Target Milestone: | --- | ||
| Hardware: | PC | ||
| OS: | All | ||
| Whiteboard: | |||
|
Description
Eugene Kuleshov
Maybe this even could include a bayesian spellchecker... ;-) Sorry, I just couldn't resist it... :-D The trouble with the spellchecker is who wil train the trainer? So, it is just a matter of good dictionaly, which Eclipse platform is not willing to provide. However there is a good chance to train the categorisation filter based on bayesian algorithm. So, I am actually very serious here. I'm needing spel cheking too... it's our bug 103093 blocked by Platform bug 51444, and they're blocked by getting a dictionary they can bundle under the EPL. I took "bayesian filter" out of the description because there are multiple approaches to making this work. For example, for a repository that was rich in structure (i.e. regular use of bug dependencies), the interest propagation mechanism could be used to infer interest. But bugs.eclipse.org definitely lacks this kind of structure, so the bayesian filter approach could work well. A secondary issue is how to encode the interest. One way is the context model, but we don't yet have a mechanism for sharing that across developers. Another simple approach could be using the CC field (e.g. so that it shows up in a query of "all reports I'm CC'd on"). Anvik and Murphy have demonstrated a very cool classifier that recommends assignments, and it could be really useful if that approach could be applied to automatically determine CC's: http://www.cs.ubc.ca/~janvik/papers/icse2006/whofix.pdf Out of scope. |