LibraryThing vs Bookish in Four Reviews

Review One

Recommendation Evaluation: Bookish and LibraryThing in seven online catalogs

February 2, 2015

I chose to look at the same set of titles in each of the seven catalogs, for the sake of consistency and comparison: In the Woods(2007) and The Secret Place (2014) by Tana French, the first and most recent books, respectively, of this author of Irish psychological mysteries; The Devil in the White City (2003) by Erik Larson, an older but still popular nonfiction title about a serial killer loose in Chicago during the 1893 World’s Fair; My Sister’s Keeper (2004) by Jodi Picoult, a novel narrated by a teen girl whose sister has cancer; and the short story collection Delicate Edible Birds (2009) by Lauren Groff.

Note: Initially, I looked at these titles in all seven catalogs in Firefox, but only Bookish recommendations showed up there, so I switched to Chrome, where both Bookish and LibraryThing (LT) recommendations appeared.

For Tana French’s books, the results from both recommendations engines were good: that is, they included recent titles in the same mystery/suspense vein, featuring female characters or detectives, possibly including a touch of the supernatural. Bookish did not include any other Tana French titles in its recommendations; LT did include up to three French titles, in addition to some others. The only questionable inclusion was Station Eleven, a new (2014) popular literary sci-fi novel.

Though Bookish only provided 2-4 recommendations for Devil in the White City, they were remarkably spot-on in terms of subject matter (history/murder/true crime). LT’s recommendations, on the other hand, included other Larson titles and other popular nonfiction authors, such as Jon Krakaeur and Simon Winchester. In this case, Bookish’s recommendations would better serve a reader looking for similar subject matter, while LT’s would better serve readers looking more generally for compelling narrative nonfiction.

My Sister’s Keeper returned the most disappointing results from Bookish: all five titles were other novels by the same author. LT did a little better, including novels by other authors, like Kim Edwards’ The Memory Keeper’s Daughter, along with 3-4 Picoult titles. Picoult is prolific, but it’s safe to assume that readers who liked My Sister’s Keeper have already considered reading other books by the same author.

With Delicate Edible Birds, I was interested in whether the recommendations would be for other short story collections. Bookish did indeed recommend recent, popular/well-reviewed story collections (e.g. Vampires in the Lemon Grove by Karen Russell, Bark by Lorrie Moore). Again, LT took a slightly different tack: its recommendations included both of Groff’s previous books (novels, not story collections), two other story collections, and one other novel. In the Edmonton catalog, I also tried Bark by Lorrie Moore, which was one of Bookish’s recommendations for Delicate Edible Birds. Bookish’s four recommendations were all story collections - including Delicate Edible Birds and three of the same titles it had recommended for that collection, which makes me wonder if the recommendations aren’t that customized at all. LT provided five excellent recommendations for Bark: story collections from Elizabeth McCracken, Lydia Davis, George Saunders, Tom Perrotta, and B.J. Novak.

Both catalogs may be weaker on older titles: for example, I tried John Berendt’s Midnight in the Garden of Good and Evil (which came up as a recommendation for Devil in the White City once), and Bookish suggested just one title, by the same author; LT suggested two novels, Snow Falling on Cedars and Saving CeeCee Honeycutt. CeeCee is set in Savannah, but otherwise, these novels have no apparent relation to Midnight in the Garden…

Overall, the two recommendation engines both did well providing relevant, recent titles; none of the recommended books from either Bookish or LT were published earlier than the mid-1990s, and most were much more recent, within the past few years. Bookish sometimes returns fewer results, but they are more accurate in terms of subject matter; however, results should always include more than one author. LibraryThing’s results reflect a crowd-sourcing/folksonomy technique: the recommended books may be less similar in subject matter, but they have other appeal factors that may suit readers just as well, introducing them to something new at the same time. Bookish was more likely than LT to include titles in other formats (audiobook, e-book, downloadable audiobook, etc.), but the format is less important than the content.

Depending on what is desired, either of these recommendation engines would be a good choice: I’d choose Bookish for the most accurate recommendations (closest in terms of subject matter), and LibraryThing for providing sound recommendations with an added dash of serendipity.

*******

Review Two

While playing around in a variety of Bibliocommons catalogs I found myself with more questions than answers when it came to the recommendations provided by LibraryThing and Bookish.  Curiosity is probably the common denominator in librarians so I shouldn't be surprised but it was hard to stop thinking about the algorithms being used as opposed to just evaluating the results.  

Factors considered:

Quality

I quickly learned that the quality of Bookish recommendations was either spot on or so confusing I couldn’t figure out what the connection between the two titles was. The best example was suggesting War & Peace by Leo Tolstoy for Little House on the Prairie by Laura Ingalls Wilder.  The age range of the material did not seem to be a factor except for one book I evaluated, Drama by Raina Telgemeier. Recommendations were all age appropriate for that title alone.

LibraryThing was much more geared to what I would expect to see on a Barnes & Noble receipt. Many of the recommendations were other books by the same author (averaged about 3 out of 5) with the other recommendations being popular titles in the same genre that came out roughly the same time as the original book. For example, Annihilation by Jeff VanderMeer had recommendations in Science Fiction that are popular now. The Martian, Ancillary Justice, Station Eleven. Popular? Yes. A good match based on appeal characteristics? No.

Quantity

Bookish definitely provided less hits than LibraryThing. There were many searches that gave me no hits from Bookish when I’d have five from LibraryThing.  It was also not consistent between catalogs. In some, I’d find a strong match from Bookish and poor matches in other catalogs.

Ordering

Ordering did not vary in most of my searches. I did find it interesting that LibraryThing would recommend two titles in a series that were not a part of the original search but would rank the second or third title in the series above the first. This seems counter-intuitive to a readers’ actual process of wanting to start a series from the beginning. In the instances where I was familiar with the series, they were series that it was important to read them in order.

Interestingness

As a librarian, I quickly got comfortable with what LibraryThing was going to give me with each search. The results were consistent. With Bookish, I was fascinated each time because the results were so wildly divergent. As a readers’ advisor, sometimes the result was perfection. Other times, I was flabbergasted.

-----

Bookish is good at surprises.  It never pulled titles in the same series or by the same author. It seemed to want to match more on subject headings than anything else but would hit gold randomly.  They are really bad at being consistent and provide too many suggestions that are nearly impossible to understand  (How to Win Friends & Influence People by Dale Carnegie for Stiff by Mary Roach). I found it very hard to trust their recommendations and if I was trying to find a book to read based on titles I’d already read I would quickly dismiss them as unreliable.  I hate saying that for it appears they try to delve deeper.

LibraryThing is good at giving readers titles by the same author and in the same series. They hit the popular titles that came out at the same time as the first title looked up which historically, gives an interesting snapshot of that year in reading. But, other than that-the recommendations are really not very connected to appeal factors. Both appeared to do a bit better with nonfiction recommendations than fiction just because they hit on similar topics.

If I were given a choice, I would pick LibraryThing. It is going to hit the basic, fluff level of the reader instead of diving deeper into reading styles and tastes but it is consistent and logical. It reminds me of Wikipedia a bit. It is a good start for basic information but if you want more you are going to have to put more effort into it as a reader, i.e. databases such as Novelist or a trained librarian.

Titles researched:

·      Annihilation by Jeff VanderMeer

·      Fifty Shades of Grey by E.L. James

·      Stiff: The Curious Lives of Human Cadavers by Mary Roach

·      To Kill a Mockingbird by Harper Lee

·      Big Little Lies by Lian Moriarty

·      Locke & Key by Joe Hill

·      Drama by Reina Telgemeier

·      Freakonomics by Steven Levitt

·      Little House on the Prairie by Laura Ingalls Wilder

·      A Child Called It by Dave Pelzer

*******

Review Three

Recommendation Evaluation

I searched for two non-fiction books and two fiction books in the seven catalogs.

Boys in the Boat was my first search.  I felt that the results were much richer from Library Thing.  It seemed to understand that this was a work of narrative non-fiction and it didn’t feel it necessary to only recommend non-fiction.  Bookish did not recommend any more than four items.  In the case of Daniel Boone there were only 2 recommended books.  Also Library Thing’s titles varied from catalog to catalog.  Bookish showed the same four titles across all catalogs.

The Life-Changing Art of Tidying Up: The Japanese Art of Decluttering and Organizing was the second non-fiction search.  Bookish did not recommend anything across all 7 catalogs.  Library Thing was less successful with this search.   The recommended titles were current non-fiction best sellers. None of these had anything to do with organization or decluttering, with the one exception being Edmonton which did suggest the Messie Manual.

My first fiction title was All the Light We Cannot See.  The results from Bookish slanted toward novels about World War II and seemed heavy on the spy/espionage aspect which is not in keeping with plot of All the Light. Bookish maxed out the recommendations at four again.  Library Thing slanted their results more toward literary fiction and I felt that Library Thing won this round.  Their lists were richer and more varied.

Gone Girl was the second fiction title. The surprise here was that Library Thing had the most predictable results while Bookish had some more creative recommendations. Bookish did seem to recognize that  Patricia Highsmith and Ron Rash both were pros  at Bad Girls who are Unreliable Narrators which is what readers seem to be searching for when asking for a read-a-like.  But Bookish was light again on the quantity with Chapel Hill and Saskatchewan only offering one title.

Across all 4 searches I thought that Bookish was light on the quantity, especially for non-fiction. That there were no titles recommended as a read-alike for Life-Changing Art of Tidying Up was surprising to me.  When titles were suggested the highest number Bookish recommended was four. Library Thing’s highest number was five. While I felt that while Bookish did a better job with Gone Girl read-alikes, on the whole Library Thing gave more robust and interesting results.

I feel that Library Thing is the better product.  The lists were longer, more interesting and more varied.  The recommendations were fairly solid and while not as precise as an in-person Reader’s Advisory interview, they weren’t all bad.  Bookish felt uninspired to me in the titles that it was presenting.

******

Review Four

LibraryThing vs. Bookish

This review covers a comparison of the recommendations produced by LibraryThing and Bookish book recommendation engines on seven different public library online catalogs.  The searches used to generate the recommendations included a variety of genres, including adult fiction and non-fiction, children’s titles, recent releases, popular titles, and classic literature.

Overall Findings

As a general rule, neither recommendation engine failed by any standard expected of a recommendation engine.  Each produced recommendations useful to both librarians and/or library users.  The consistency of the products for producing quality recommendations varied by book type.  Overall, LibraryThing produced more recommendations than Bookish for all genre types.  Both programs often struggled to match the content and tone of the searched book with other similar books; opting instead to recommend books of the same easily recognized genre.  This issue was more obvious with non-fiction works where the content of a book is critical to finding a second book covering a similar topic.

There were examples of a recommendation engine falling wildly off the mark in matching a searched book with other similar books.  The two products however had a tendency to balance each other out.  Even when one missed the mark the other product would usually produce a useful recommendation or two.

Types and Usefulness of Recommendations

Searching for a variety of literature titles produced a hit and miss list of recommendations.  LibraryThing overall produced more recommendations than Bookish.  However the former was much more likely to add to their list with books by the same author.  While this can be valuable information, it is not hardly surprising or unique.  It is also quite easy to find books by the same author in other fashions and is a likely search by both librarians and patrons making these recommendations superfluous.

Bookish tended to produce the more surprising and unexpected recommendations of the two products.  On more than one occasion, this author was forced to add a new book to her own “to-read” pile due to an unexpected connection.  This recommendation engine relied less on producing results for the same author, resulting in possible missed books in an author’s series, but obviously working harder to match book content.  On the other hand, when Bookish missed, it did so badly.

When it came to finding children’s books each product obviously had a different formula they followed to produce results.  LibraryThing stuck to the trusted formula of showing books written by the same author, which did result in more overall recommendations.  Bookish did a better job matching a book’s reading level than with producing recommendations based on content and topic.  Though Bookish too on occasion, used the fallback of same author results.

Non-fiction titles surprisingly proved the most difficult of the recommendations.  The usefulness of the recommendations would likely variety by person to person.  Did the patron like the book 1968 by Mark Kurlansky because they enjoy reading history based on a particular year, or did they want to know more about the Vietnam War?  When searching for similar titles to Mary Roach’s Stiff: The Curious Lives of Human Cadavers do you want more humorous, science books or something else?  With non-fiction there are many possible ways to frame matches for recommendations and both products tried a variety of methods resulting in a wide range of books.  Bookish was a bit better at closely matching content, but overall produced fewer recommendations.  On the other hand, LibraryThing produced more results, but it did not guarantee success.  Though LibraryThing was obviously using the formula of producing results based general book topic, and not extremely specific content.

Final Recommendations

Which of the two recommendation engines a library should choose to utilize on their website would be based on their own criteria and needs.  If library wishes to have the product that displays the most results than the obvious choice is LibraryThing.  Overall, LibraryThing’s recommendations were more consistent, though they were rarely surprising.  The product did rely heavily on showing books by the same author or, for non-fiction, producing results using a very broad matching category.

If a library wanted a recommendation engine to produce more unique results which did a better job of matching book content, and not just the general genre, than Bookish would be the recommendation engine of choice.  Bookish produced a wide range of results and on occasion had no problem recommending fiction titles which matched the subject matter of a non-fiction book.  As a trade off there was more than one occasion when Bookish was trying a bit too hard and missed the mark.  Libraries using Bookish however would have to accept a product that produced fewer results overall and was therefore less visible to their patrons.

While it would be nice to say one product was vastly better than the other this was not actually the case.  Very rarely did the two recommendation engines produce the same results; however this seemed to be for the benefit of the user as they had a better chance of finding a match for what they wanted.  Together LibraryThing and Bookish were more likely to find the right answer than if they worked alone.