Though I believe this report is about public library users, I suspect this applies also to academic library users to some extent.
This also reminded me of a panel I moderated at IATUL 2012 last year, the panel was designed to pick the brains of non-librarians and one of the panelists asked a room full of academic librarians why libraries did not have Amazon style recommendations.
The responses from the librarians in the room were interesting ranging from privacy issues due to the Patriot Act and lack of technology.
I think, by now most libraries or at least academic libraries have finishing implementing web scale discovery systems which is doing what Lorcan Dempsey of OCLC would call "aggregating supply". The next logical step is to "aggregate demand", using recommender systems.
While such recommender systems are not common yet to library or library related systems in both the public and academic libraries, they are not totally unknown, here are some I am aware of.
I start with recommender systems for books (mostly) and then move on to article level recommenders. I focus mostly on recommenders that make use of circulation data, or usage data of other users "People who borrow/read X also...", though matching based on other item characteristics might be included as well. Personalised recommenders that either ask for your explicit action to say rate books and make recommendations based on that or those that pull data from your past actions (e.g past borrowing, past action in reading articles, adding to lists) and match against other users are also included.
Ones that simply recommend based on themes or best selling lists are of less interest to me as they are not as personalised, so I left out apps like Gimme and the YALSA Teen Book Finder App (thanks @CarliSpina), and also non-library related recommenders - BookRx, WhatshouldIreadnext (I am less strict for the article recommenders)
As stated in the already mentioned blog post
"What is BookPsychic? Launched in August, BookPsychic is an easy and fun personal recommender system for library patrons—like Netflix or Amazon, but all about what’s in and what’s popular at your library. As you rate books and DVDs, BookPsychic learns more and more about your tastes, and comes up with recommendation lists. And everything shown or recommended is available at your library. Simple “bookstore” genres, like “Recent fiction” and “History,” help you zero in on the books you want."
You pick a library that is enrolled in Book Psychic, rate the book in the library system presented to you by Genre and you will see recommendations. Pretty simple and effective.
A nice touch is you can import ratings from other systems such as LibraryThing itself and Goodreads.
One thought does strike me, with so many "social reading" systems like Goodreads, which Lorcan Dempsey of OCLC would no doubt consider at the web scale level, a alternative strategy would involve libraries supporting these systems rather than building their own recommenders, and either providing those systems with holding data or at the very least provide a easy way to check/link if book is available.
This parallels the support of Google Scholar etc with OpenURL resolvers, except in the case of books this is even much similar with ISBN searches in Opacs.
2. Huddersfield Book Recommender system
Strictly speaking there are many types of recommenders and many ways to classify them, these range from those that are basically showing "similar books to X" on the item record of X and those that really track who you are or at least recommend based on your explicit ratings or loan records and adapt to your individual preferences (e.g Book Psychic above). (There's perhaps a third type, where users explicitly give recommendations by manually adding "similar titles", or allowing others to follow what they loan out or rate.)
I suppose the "similar to x" type recommendations relating to each title are in a way in most opacs and discovery systems, since many allow a one click to items with same subject, author etc.
But "similar to x books" based on borrowing activities of people who borrowed X, is indeed an advancement.
That's what the Dave Pattern has done to create a homebrew recommender system for the University of Huddersfield Catalogue.
Below shows a recommendation for the book in the catalogue. Example is suggested by Dave himself.
Sometimes the recommendations give odd results below is one for Men are from Mars, women are from Venus , this could be due to the lack of circulation data.
David has written a few blog posts on this topic including showing the impact of this service
3. National Library Board (Singapore) Recommender - Sharealike system
The National Library Board (NLB) in Singapore here also has a book recommender system though admittedly I don't know much about how it works (please note I work at the National University of Singapore Libraries which is independent of the National Library Board).
There are some details in this recorded talk , where it is mentioned that NLB is unique in having lots of circulation data to mine through due to high borrowing levels in Singapore. So definitely some of the recommender system is based on circulation records, though I get the sense some is based on similarity in item characteristics?
In any case as a user I do see the following popup when I check my loan account.
I blocked out the titles I borrowed, but you can probably guess what I borrowed in the past.
Is this recommendations based on mining circulation records of other users who have borrowed similar items as the ones you have done, perhaps akin to the Huddersfield one or is it recommending based on other characteristics of items I have borrowed? Not sure here.
There are kiosks around in the public libraries, that allow you to slot books in, and it will recommend similar books - aka "Title Recommendation System". Same comments as above is this based on circulation data?
That said, the National Library Board also has Read on site that allows you to check directly book recommendations and those are definitely based on circulation data.
Here's the entry for The Hobbit, where you can clearly see "Out patrons also borrowed the following"
I am unsure how "Quick Picks" is determined, but I would guess this is based on similarity in subject, author etc?
I was a bit curious about how they integrated this with normal catalogue and it seems it's handled in their new Primo system dubbed NLB SearchPlus
On each item record, there is a link to a "Recommendation" tab that actually links out.
In the catalogue accessible on the mobile web site, you can see a "other related items" but I don't think that's based on "our patrons also borrowed.."
4. Ex Libris Bx Recommender + ScholarRank
Academic library services like their public library counterparts generally don't have recommender systems. Offerings by Ex Libris systems seem to be an exception.
In the video below they talk about ScholarRank and how ranking is done. The first few points are pretty standard but around 3:05 they talk about how ranking is based on user characteristics.
Example given is how a search for "Mercury" would give different relevancy ranking if done by a chemistry student as opposed to a student majoring in the music. The system also takes into account whether someone searching is a Phd student or a fresh undergraduate.
Very interesting, though I suppose this can only happen if the user has already authenticated?
Still the above merely changes the relevancy ranking, but what about outright recommendations for articles?
This is where the bx recommender comes into play.
Essentially, this leverages usage data from users of SFX, perhaps one of the most popular OpenURL resolvers in the world. By association of articles accessed by researchers via SFX in the same session , they are able to create recommendations essentially "researchers who searched/accessed this article also....."
The interesting question is where to embed these recommendations, the video mentions Primo and some other ILS can put such recommendations next to results. Below shows an example from Central Michigan University Library using Primo Central.
If Recommendations are available there will be a recommendations tab you can click on.
But more commonly I see it appearing in link resolver screens typically for libraries using SFX by the same company. But it can be done on other link resolver systems. Here's an example by QUT using 360link.
5. Google Scholar Citations
One of the things that initially escaped my notice was that after creating a Google Scholar citation profile, Google Scholar will start recommending items.
The main limitation of this is that it is recommending based off articles or works you added to your Google Scholar citation profile.
Below it recommends a article on wikipedia because I have an article published relating to that done when I was in library school.
The issue with basing recommendations just off published works you have done is, often by the time you have published something you pretty much scoped out the area and are actually least in need of the recommendations because you did the literature review already!
It's true this features helps you keep uptodate about developments in the field after that, but it hardly helps at the start and for new researchers just starting off or even established researchers seeking to expand to a new area this feature is a non-starter.
I wondered if one could work around this problem by adding works you find interesting to a profile to see what recommendations popups but keep the profile private, but this doesn't work since recommendations work only for public accounts.
6. Citeulike + Mendeley
Instead of trying to abuse Google scholar citations, one can probably use Citeulike (which was recently acquired by Springer) to generate recommendations.
Unlike Google where you can add only what you published, this is based on items you add to your library of items you are interested in.
How does it work? The settings have some information
The other system that is very similar is Mendeley, which I have written on before.
There are 2 types of recommendations. The first is available via the web version and shows "related research" link for each item shown. Note, when you click on the links, some will have no recommendations.
Before are some details.
7. Read by QxMD
Currently the Krafty Librarian blog is tracking and reviewing 4 Medical related iPad app service namely
Flipboard for Medical Journals (or all academic journals in the case of Browzine) and are a important development due to the high tech usage in the medical world.
However, for the purposes of this blog post, I am more interested in apps that are positioned as a Zite-like app for academic articles. Not only must the app, aggregate articles and display them in a easy to read newspaper/newszine like format, they must also learn from what the users select to read, their thumbs up and down and use machine learning techniques to customize the articles to display.
I am not sure, which of these apps are closest to Zite, but as noted in the review on the Krafty Librarian blog, QxMD makers of Read seems very proud of their "machine learning" etc technques '
"“Rather than simply relying on our users to tell us which journals they want to read, we use a combination of machine learning, semantic analysis, crowd-sourcing and proprietary algorithms to figure out which articles our users should likely be reviewing.”
comments at http://www.imedicalapps.com/2013/01/flipboard-medical-journals-read-qxmd/
There are plenty of other pilots and proof of concepts for recommender systems in the library world of course. This includes
- SPLURGE: Scholars Portal Library Usage-Based Recommendation Generation Engine - hat-tip @tedlawless
- RISE - Recommendations Improve the Search Experience - JISC - hat-tip @richardn2009
In years gone by, I also blogged about attempts to make your own recommender using a bayesian filter of RSS feeds. , though with more and more apis available, there are more options now including a interesting idea here to use this idea combined with information drawn from your Mendeley library using the Mendeley API .
I have no idea how good in general these recommendations are, in any case when I asked on Twitter for any library related recommender systems my network might be aware of, one of the wry replies was "yes, the librarian".
Indeed, just as the meme that states Librarians are the original search engine, Librarians are also the original recommender systems.
There was also another piece of irony that passed me by until now, to write this post on recommender systems, I had to ask recommendations from people on the Twitter network. So should I include Twitter in here?