Search results
Suggestion for live site slowness reports
Submitted by
jenn.
on 2008-11-10 13:39.
After I grumbled at Jonathan yesterday during the read-timeout episode, we talked a bit about how we could keep track of performance issues without the admins feeling harrassed or the reporters fearing for their lives.
Submitted by
bnwest.
on 2008-10-15 14:28.
Discussion of changes required to CNX LaTeX importer for the CNXML 0.6 language change.
Submitted by
jccooper.
on 2008-09-10 17:23.
The new version of SVN is out, and it has (finally) built-in merge tracking. And it's better than svnmerge.
Submitted by
reedstrm.
on 2008-08-14 15:42.
Turns out it is possible to sign on to jabber from more than one client, simultaneously. (This is the root cause of the 'jabber keeps disconnecting me' problem)
Submitted by
kef.
on 2008-07-15 13:09.
Jared Spool's article, http://www.uie.com/articles/search_results_part2/, has some interesting results from researching various companies' search results pages.
- Provide a lot of info in the results to support user tasks. He gives the example of price, distance, and room amenities for hotel searches, to support various types of customers. For us, I think that means providing axes of information about the content in the initial results page, which we do fairly well. At some point, including lens information, ratings and reviews will probably be important.
- No one ever complains that the page has too many results -- 10 per page can be limiting, but I think right now performance issues force us to limit the UI.
- Most relevant results should be first. Well that is obvious. But the interesting result is that people stop searching down if irrelevant results appear (relevancy threshhold). As we get textbook searchers, we really need to reconsider how often modules are obscuring collection results that may be more relevant to the searcher.
- Eliminate wacko results. Just thought the psychology here was interesting. Programmers see wacko results as a weird corner case of an algorithm, but searchers see them like Tourette's syndrom (Spool's term) -- random results inserted into an otherwise sane response -- and it makes them not trust the other results.
- Prevent pogosticking -- The more times someone goes to look at content and comes back to the search results (which Spool calls pogosticking), the less likely they are to ultimately find what they want. (I think that this would be a good stats problem for our new book -- probability of success given Y pogos) Preventing this is related to the "Provide a good scent in the initial display).
Submitted by
bnwest.
on 2008-07-14 16:46.
We need a Word Importer test bed so that we have a set of word documents which represent the word documents that we are receiving from the wild and which we can be examined in great detail. A goal is to do a better job importing word docs from authors who import the most content into cnx.org.
Submitted by
jenn.
on 2008-07-11 11:40.
Some notes on where our patches for Trac live and how I use them
Submitted by
reedstrm.
on 2008-07-02 12:53.
Chuck and I ran a few 'quick' tests of module PDF performance, using an amazon EC2 instance. Here's what we determined.
Submitted by
ew2.
on 2008-06-26 11:18.
As we move toward having more outside development, automated unit tests could help keep our code in top shape.
