[Xerte] ANN: Xpert Image Search and Attribution

Patrick Lockley Patrick.Lockley at nottingham.ac.uk
Mon Sep 6 13:49:28 BST 2010


> would be more useful with browse facilities IMO
> -----------------------------------------------
> How would you like to browse?

Ooh, at a wild guess: tag/keyword, category, subject, that sort of
thing. Give the user a bit of guidance, Pat, and give them the
'serendipity option'. 

Most of the pots we search don't have tags, or a lot of metadata bar the
title. Worse, some of the pots are massive - how many pictures on
flickr? - and flickr doesn't have a tag browse option.

> What does Xerte Image Search offer that these don't?
> ----------------------------------------------------
> I'm assuming you didn't click on the select tab

Yes, I did. I searched for "Meall nan Eun" because that's one of my
photos on Flickr and is a pretty specific term, and it did throw up
attribution info. That's useful, though other meta-search-engines do
that as well. 

I've not seen another search engine that embeds attribution. I'd be
interested in seeing one that does if you could send a link.

> Behind the typically spartan interface what goodies lurk?
> ---------------------------------------------------------
> What would you like it to do?

Answering a Q with a Q is no answer. I'm just asking what's behind the
basic interface. 

I was offering a chance to be more accurate with my answer, but here is
the full version.

When you search the following occurs.

The query is posted via Ajax into three separate, custom APIs.

The apis are modular in nature (one for images, one for sound, one for
video).

Each API per media type contains scope for further modularity - the
image API contains two sub-API queries.

The image API also contains a unique Wikipedia API (written by me).

The video and sound apis have been based around an xpert search. Some
new custom tables have been built in xpert to facilitate faster queries
for specific content and two very basic APIs have been written to allow
these to be searched.

All this data is then returned, and parsed asynchronously (using
javascript) back into the website. The results from these APIs are then
sorted into an internal data structure to allow for all of the results
to be traversed quickly without the need to conduct another search on
that keyword.

The pages that handle the attribution are based round PHP5 and GD
libraries, with some extensions to allow for the creation of powerpoint
files from basic templates.

Hope this helps.




More information about the Xerte mailing list