Sunday, July 12, 2009

#ala2009 #ExLibrisALA Liveblogging the bX Panel

My notes on "bX: Users Who Looked at This Article Also Looked At...", Sunday, July 12, 10:30-12:00PM at ALA Annual in Chicago aren't exactly liveblogging - the Hilton doesn't have free wifi. But here they are, for anyone who was interested in the session and couldn't make it.

Oren Beit-Arie, Ex Libris
Robert Gerrity, Boston College
Nettie Lagace, Ex Libris

OBA: bX is new, launched just a few weeks ago. Web as web of users, sharing opinions and navigation aids like tags. Their selections and preferences help other users in e-commerce - why not in academic space? Two ways this happens: explicit user contributions, like reviews, ratings, and tagging, as in a social OPAC; and in implicit contributions, in captured user data.

Examples of implicit contributions: UKSG Usage Factors project, Project MESUR funded by Mellon. This started 2 years ago at Los Alamos to look at a range of measures of scholarly impact.

Evaluation of scholarship is based on print paradigms - authorship and citation. (Is that necessarily print based?) The alternative is usage based. (But this isn't peer review; it allows non-experts to have a say.)

Recommender systems - Wikipedia definition - information filtering system to present information likely of interest to a user

This is needed for scholarly research because of information overload - users need to find relevant information. Particularly for researchers in areas new to them, or in interdisciplinary research (good point).

Other recommenders: BibTip, in BC OPAC; LibraryThing

bX focuses on the core unit of articles, across the distributed academic publishing universe, and structural analysis of use patterns, not just popularity. Derives from research by Bollen and Van de Sompel, partnered with extensive global list of universities.

bX is built on OpenURL. Harvests usage logs from link resolvers (SFX) to build very large aggregations to mine. There are about 3000 link resolvers in the world, about 1800 of which are SFX. All major information providers participate in OpenURL.

Example: user puts keywords in EBSCO database, clicks SFX button to choose article. In next window, in addition to full-text links and OPAC link, gets links to additional articles. Would get identical results if they started from Scopus. (Is that really an advantage? Don't different users prefer different databases? I think of undergrads using EBSCO and faculty using Scopus. Wouldn't originating database be a source of information on user preference?)

Results can be returned as XML, RSS, etc. and be embedded in other portals, etc.

Article relationships are created by observing users' choices within a given session. This is aggregated across all users across many SFX layers.

More types of services to follow: trend analysis for collection development, comparison of citation patterns and usage patterns, Map of Knowledge from Mesur.

(Once we get into claims about evaluation, the idea about "usage" is slippery - just because someone clicks through to an article, or even downloads it, doesn't mean that in the end they find it useful. Not a good way to evaluate scholarship, even if it's a good way to evaluate a collection.)

Conclusion - moving from search to discovery

Netflix, Amazon, iTunes, MapMyRun - recommender services everywhere so why not in the library?

Usage data is an untapped resource. 283,000 requests to SFX so far this year at BC, over 200,000 clickthroughs. Allows information about usage by time and day, top journals, etc, but there's more. Does require libraries to share data, but at aggregate level that protects user privacy.

OPAC recommend system BibTip see DLIB May 08 - drawbacks are that they're specific to the BC OPAC and take a long time to build.

BC is bX beta tester, Nov 08 to April 09, with subject specialists testing predetermined resources and independent research for both validity of recommendations and user interface. Also tested different versions of bX algorithm.

Tester ratings of quality of recommendations: only 10% not good. Quick and easy to implement through SFX web admin module, used default settings. Requires configuring OAI server to harvest data from users.

BC: About 4% of users who see recommendations have clicked through since May, but no marketing yet. Definition of success is if some reasonable percentage of users find it useful and the rest find it unobtrusive (I agree).

Not many recommendations in humanities yet, since mostly books. Would like to see a feedback mechanism for users to evaluate recommendations.

Demos from Google Scholar, PubMed, EBSCO

Google Scholar - search, results, click through to library SFX links - next page shows full text links, catalog links, then bX recs with SFX links
PubMed (What I want to know is how similar the bX recommendations are to PubMed similar articles. Which gets more clickthroughs from users - bX recommendations or PubMed similar articles?)
EBSCO - NL points out that the keywords in recommended articles are different from the original article - bX helps identify synonyms

bX setup goes through SFX admin. Register, enter license key, activate target and test, and publish data. $3,000 for a single library. $15,000 for 6-20 library consortium. 30 trials available. Subscribers are strongly encouraged (but not required?) to contribute their users' data. Contributers have the option to get recommendations based just on the users from their institution, although no one has done that yet.

One questioner asked about skipping the SFX window and linking directly to the full text, which users often prefer. NL and OBA answered that bX can be configured to include a link to recommendations in the banner frame so that you can still link directly to the full text.

No comments:

Creative Commons License
Libri & Libertas: Books & Freedom in a Web 2.0 World by Laura H. Wimberley is licensed under a Creative Commons Attribution - Noncommercial - Share Alike 3.0 United States License.