Brief explanation: two methods were used, Termine and Extminer. Users rated the relevance of matching items in their digests on a scale of 1-10. The digests were organised according to categories of feeds (CFPs, new items in IRs, etc). The statistic that stands out is that of 21.67% of the journal articles selected by Approach A were rated gold dust (i.e. had a rating of 8-10 by the testers.  Other things to note include very low ratings for some categories (such as component announcements, and funding opportunity news) where the metadata content is dissimilar in nature to the metadata used to create Personal Interest Profiles (which was metadata from articles taken from useage of ticTOCs).

Scholarly journals - new free service ma 1,871 
About 373 
New journal TOC RSS feeds 293 
Journal TOC RSS feeds from Revues.org   274 
I want to be completely honest with you  227 
New improved ticTOCs Journal Tables of C 196 
Bloggers help make ticTOCs launch a succ 187 
Journal TOC RSS feeds from Haworth Press 157 
Journal TOC RSS feeds for ICE/Thomas Tel 144

2. To post an image, you've got to have a URL - i.e. I can't see a way to upload an image.  OK - you can use twitpic or flickr or whatever, but it would be easier sometimes just to upload pics.

Of course - I may be missing the obvious.

The above is a diagram Simon Hodson, the Gold Dust Manager, created showing the items retrieved for one particular user in the first iteration of the Gold Dust project tests with users. 

The items are ranked according to the significance rating which was given by the software used for one of the Gold Dust approaches.  Then, the items which were rated 8-10 (i.e. the highest, and therefore, the Gold Dust) by the user in question are coloured white and the remainder coloured black. 

The diagram we are left with looks, as it should, like a pint of Guinness settling out!  And Simon added a nice caption.

In the first round of testing, a few users’ results looked more like pretty flat pints, or more like bar codes.  But, a good proportion of the Gold Dust users’ results look like the black stuff, so Simon is reasonably confident that the approach used, and the filtering methods, can have a reasonable success in pulling out significant terms for the PIPs.

The second Gold Dust iteration should be complete before the end of next week.

Roddy MacLeod
ticTOCs Management Support