[Web4lib] Re: Google News Timeline

[Peter Noerr]

Except, of course, your second example is a false hit, and thus doesn't fit Katherine's (or your) criterion of all being relevant. (And you are postulating an interestingly sophisticated search engine to retrieve the second record from the query!)

We could argue about the silliness of the different relevance algorithms, and virtually all of them break down when the calculated relevance of a record is 100% or 0% - as happens an awful lot of the time.

My question was really; If relevance ranked results are silly, what is a not silly display ranking? And why is it not silly?

In this very specific case (all articles from a journal issue) there is a possible canonical order in the page numbers as well as sequences for the title and author rankings. Do any of these make any more sense than a relevance order? And is the order only "non-silly" when seen in the context of the user's use of them?

Peter

> -----Original Message-----
> From: Thomas Dowling [mailto:tdowling at ohiolink.edu]
> Sent: Friday, April 24, 2009 11:56
> To: Peter Noerr
> Cc: web4lib at webjunction.org
> Subject: Re: [Web4lib] Re: Google News Timeline
> 
> On 04/24/2009 02:33 PM, Peter Noerr wrote:
> > Katherine, I'm intrigued by your last sentence. ("That's just silly and
> increasingly sophisticated users know it.")
> >
> > Why is relevance ranking silly in this case or any other? (I'm not
> arguing for it, I would like to hear your reasons)
> >
> 
> If you search for "articles from volume 5, number 3, September 2007", which
> hits are most relevant?  To humans, the answer is none: they're all the
> same.
> To a relevance algorithm, the answer is likely to be either "the shortest
> one
> since 'volume 5 number 3' makes up a higher percentage of the record"; or
> "the
> one that happens to have 'a volume of 3 to 5 gallons' in the abstract,
> since
> there are more matches on search terms".
> 
> The disconnect can certainly seem silly.
> 
> 
> --
> Thomas Dowling
> tdowling at ohiolink.edu