[Yulpub] Recommender services

Thu Feb 18 10:58:44 EST 2010

Along the lines of David Stern's point, during that webinar I asked about (but didn't get an answer then - it was one of those one-way, submit your questions via this chat box) whether ExLibris will undertake or sponsor studies on that potential positive feedback loop ... the more an article gets chosen, the more it gets recommended, the more it gets chosen, etc.   I would be concerned about that too, and would think adjustments to their recommendation algorithm based on that consideration might be part of any trial/negotiation process.

The other presenter, from the California State U. system, mentioned that the deployment of bX led to substantially reduced demand/cost for ILL'd articles on one, large campus, a slight increase on another, smaller one, from 2008 to 2009 ... acknowledging that these are primarily undergraduate, teaching universities.

A recording of that webinar<https://event.on24.com/eventRegistration/EventLobbyServlet?target=registration.jsp&eventid=184680> is supposed to be available within a few days.

-- Stefan

From: yulwww-bounces at mailman.yale.edu [mailto:yulwww-bounces at mailman.yale.edu] On Behalf Of Okerson, Ann
Sent: Thursday, February 18, 2010 10:18
To: yulwww at mailman.yale.edu; yulpub at mailman.yale.edu
Subject: Re: [Yulwww] Recommender services

Dear all,  I was sure that I'd been part of a message chain on this service, and I went back into my full e-mail archive to find what it was.  For those who are interested, the discussion was with the Borrow Direct Collections AULs and I reproduce it here.  It gives some sense of peer institutions to contact.  I'll also ask how Penn and Princeton doing with the trial and report back if I hear. So far, this is all there was.   Ann

****

7/13 - From Martha Brogan, U Penn

Yesterday, I heard a presentation by Ex Libri's Orie and others about their new bX recommender service.  It is described here:

http://www.exlibrisgroup.com/category/bXOverview.  Boston College discussed their experience as a Beta site.  (Princeton has also been a beta site.)  The service aggregates use-data from all SFX contributing institutions to recommend other articles to readers across a wide range of platforms and databases (e.g., Scopus, Ebsco, PubMed, etc.) This service has lots of potential to lead towards dynamic knowledge mapping so we can see inter-relationships among research areas.  Eventually, it could give us trend analysis and help us better understand new directions in research.  In the meantime, it offers usage stats (journal and article level?) that can give us valuable insights into how our collections are used.  At an annual subscription cost of $3K and relatively simple set-up, I am interested in subscribing at Penn.

Boston College reported high % satisfaction w/ the recommender results and they had their 4 most critical librarians put it thru its paces.  My understanding is that it is phase one only and they hope to develop citation tracking soon.  It will improve as more research insts such as ours join and contribute data

WITH A CONSORTIAL PURCHASE by 6-20 members, the price drops to $15k.  I thought this would be a great joint purchase--altho the savings are not huge for us, the prospect of adding and having access to our combined usage data is a powerful concept.  However, I understand that Dartmouth and Cornell are not SFX customers; Yale, Princeton and Penn are.  Don't know about Brown and Columbia?

Given the above, perhaps this should be bounced up to NERL?  Anyway, I wanted to alert you to it and get your reaction.  It builds on work that van de Sompel and Bohlan have done at Los Almos w/ the mesur project: http://www.mesur.org/MESUR.html.

7/13 - Reply from David Stern, Brown

This tool restricts next step discovery to popular material identified by previous searchers ... while it provides an additional user- generated pointer, it does not really provide the broader research navigation elements we pay dearly for -- citation tracking, controlled vocabulary, find similar hedges. It is interesting, but may ultimately be distracting and/or dangerous if we take seriously our mission to provide high quality navigation tools. Worth looking at, but the limited number of users generating the pointers us also a concern, since the largest percentage of our users actuaalt link using CrossRef (Google without a VPN) or our A-Z list which is separate from SFX.

7/13 - Reply from David Magier, Princton

Marvin and I have been in touch about this and after he gets back from ALA, we'll put together some more info. Meanwhile here are my two cents on automated recommender/pointer services like this. Philosophically, I am all for crunching usage data as much as possible and extracting every last ounce of value out of that data that we can.  But I am also quite leery of potential over-reliance on such data for steering patrons to resources in largely unmediated ways. It's one thing to say that prior usage can be a good predictor of *likelihood* of future usefulness. But the devil is in the details of how, by whom, for what, and under which conditions resources get used successfully, and I have a feeling that the artificial intelligence algorithms behind "recommender" services are simply not subtle enough to account for all those factors properly. It's not that the resulting pointers won't be useful, but, just as with Google searching, there is a danger that automated pointing will get users to "settle" for content that is "good enough" for the moment, but not really the *best* pointer (which would have resulted from a more mediated interaction). Anyway, more on Princeton's experiences with bX soon.

10/14 - Martha Brogan, Penn

You may recall that I wrote you all during the summer after ALA about Penn's interest in bX recommender. We plan to start a trial of this service later this month.  If you are interested and also conduct a trial, let's explore the possibility of joining as a consortia.  It seems as though we would have a very powerful data set that could help inform us about important patterns in research.  You do not have to be an SFX-subscriber to use this service.

Info at: http://www.exlibrisgroup.com/category/bXOverview

10/19

From: Syed Ahmed <Syed.Ahmed at exlibrisgroup.com<mailto:Syed.Ahmed at exlibrisgroup.com>>
Date: Mon, Oct 19, 2009 at 10:26 AM
Subject: RE: bX Recommender Service
To: Adam Chandler <alc28 at cornell.edu<mailto:alc28 at cornell.edu>>

Hi Adam,  Marian is no longer with Ex Libris and I'm taking over her responsibilities. At the moment, bX is in beta mode and only works with SFX. We are working on making it available for non-SFX resolvers but we don't have a timeframe when this will be available.  Please let me know if you any other questions.   Syed

From: yulwww-bounces at mailman.yale.edu [mailto:yulwww-bounces at mailman.yale.edu] On Behalf Of Bauer, Kathleen
Sent: Thursday, February 18, 2010 9:41 AM
To: Kramer, Stefan; yulwww at mailman.yale.edu; yulpub at mailman.yale.edu
Subject: Re: [Yulwww] Recommender services

Hello,
The bX Recommender Service does sound very promising. Nisa Bakkalbasi had also sent to me word of a presentation by Nettie Legace of Ex Libris which she saw at the recent Electronic Resources and Libraries conference:

"Services recommending books - BibTip, LibraryThing, University of Huddersfield borrowing recommendations, and articles - bX from Ex Libris, PubMed, Synthese (CISTI) now exist in the academic context. JISC in the UK is sponsoring a major project, MOSAIC: "Making Our Shared Activity Information Count." This session will provide an overview of these recommendation systems, describe their different approaches to data mining, and discuss their role in improving information retrieval and user experience in a now nearly fully online scholarly information world."

When the Usability and Assessment Advisory Group looked at functionality needed for the next generation catalog, one of the things that we thought should be available would be ways to harness data about what patrons are using. Traditionally, scholars have used citation information to know what articles and books are influential in a field. Ejournals and services such as SFX record information about what journals and articles are being accessed: the data recorded are a potentially rich source of information that could also be used to identify key articles and journals, in a way that would be faster than citation information. This wouldn't replace citation information, but it would add to it. It's important to note that this information is also not linked to individual patrons-data are pooled, so that overall, not individual use, are being examined.

Scott Matheson and I have been looking at ways we might improve the library's interfaces by mining use data.   We've been concentrating on mining search log files from the web site. Based on the what we are hearing, we will add bX to the list of services to consider for making better use of data to improve interfaces and services.  If people hear of other promising services, please pass them along.

Thanks,
Katie

From: yulpub-bounces at mailman.yale.edu [mailto:yulpub-bounces at mailman.yale.edu] On Behalf Of Kramer, Stefan
Sent: Wednesday, February 17, 2010 4:07 PM
To: yulwww at mailman.yale.edu; yulpub at mailman.yale.edu
Subject: [Yulpub] Recommender services
Importance: Low

I attended a Library Journal-sponsored webinar about the bX Recommender Service<http://www.exlibrisgroup.com/category/bXOverview> today; it's supposedly based on the work of Bollen & Van de Sompel<http://public.lanl.gov/herbertv/papers/jcdl06_accepted_version.pdf>.  It sounds quite intriguing ... has something like this been considered at YUL?  Other "recommendations based on other readers' choices" activities mentioned in the webinar, aside from the ever-cited Amazon.com example, included MOSAIC<http://sero.co.uk/jisc-mosaic.html> and BibTip<http://www.bibtip.org/>.

-- Stefan

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.yale.edu/pipermail/yulpub/attachments/20100218/8826416d/attachment-0001.html 