On the TagGetRelatedTags sproc, what is the behavior you are seeing that is not as you expect?
Is the tag search bringing back items you do not wish it to return, or is it leaving off items that should be returned. Or are you wanting to extend the tag search functionality to include metadata or textual content searching?
I can coordinate development and testing for the Tagging and DailyArticleGet requests. I am not a SQL performance or security expert though, so I will leave those requests to someone else to coordinate.
Thanks, Kevin Hodgkins 205-796-4521
Are the master, temp, and user DB's located on separate physical disks on separate physical controllers. I have a transaction intensive DB that had substantial performance issues until I added a controller and disks and moved the user databases to it. What is the physical layout of the MISESSATA2003 server?
So I have played around with the sproc some now and really just need to know exactly what you want it to do. the existing sproc would pull back all related tags. In the example I used I looked for all related tags to "Aprioirism" which returned 351 related tags. Out of these 351 related tags there were 5842 records.
If I limit the realted tags to the top 10 related tags (where both the searched tag and the related tag exist on the same item) it still returns 3729 related items.
Re TagGetRelatedTags:
The proc does what it does correctly, but the results are not very relevant. We need a better algorithm that takes into account how many tags documents have in common and prioritizes those higher. There may be other relationships that you will be able to exploit. If we can to metadata and textual content matching, that would be awesome.
I have been researching tag relevance literature over the past couple of days and there seem to only be several normally exploited methods by which tag relevance is determined. The first method is basically what the proc does - measuring co-relatedness. Another way is defining tag tuples for each combination of tags on a document and then measuring co-relatedness of the permutations of tag tuples (so basically a multiple level co-relatedness value).
I think the real value would come from the document metadata. What metadata is available though on documents (or items - whatever they need to be called) and what tables are they stored in?
I will try some things starting with document data instead of a single tag passed in through the variable.