Share |
print

« Back to NoveList Blog Home

Good Recommendations: Are People Better Than Algorithms?

Post by Amie Reno
Posted August 12, 2014 in Readers' Advisory News

NoveList SpotLight Image

In our continuing series on good recommendations, we've explained how genres, appeals, and subjects feed into our recommendations at NoveList. And Shauna Griffin shed light on our process for creating hand-written recommendations. But NoveList has another type of recommendation too, and in this post, I hope to demystify our recommendation algorithm.

Now some people might hear the word 'algorithm' and immediately think that there is no way a computer could possibly understand how to recommend books. While algorithms do make use of computing power to create recommendations, let me assure you that there are a lot of real people (librarians and engineers) at NoveList working together behind that algorithm to make sure the results are high quality.

At its core, our algorithm combines 2 genres, 2 appeal factors, and up to 3 subjects to find similar titles. That is a highly simplified version, and we have a team of librarians (with over 60 years of combined experience working in libraries!) who gather regularly to refine and tweak the reasoning.

We use our expertise to ensure that adult, YA, and juvenile fiction and nonfiction materials all get the best treatment the algorithm can give them. This past year we have been focusing intensively on improving a few key areas:

Using our hand-written recommendations to improve the algorithm

Basically, we're taking advantage of the fact that we already have some awesome handwritten recommendations for series by turning them into title recommendations. You may have noticed these recommendations in NoveList starting last fall. They always read something like this: "NoveList recommends 'Under the never sky trilogy' for fans of 'Divergent trilogy'. Check out the first book in the series."

Weeding out bad combinations

We know librarians love weeding! But actually this is a chance for us to say 'Genre A should never be paired with Genre B'. There are lots of reasons why this might be the case, but mostly they just don't make good recommendations. Here are some examples of combinations you won't see as a result:

  • Funny  /  Serious
  • Disturbing  /  Heartwarming
  • Charming  /  Nightmarish 
  • Energetic  /  Detached  
  • Cozy mystery stories  /  Creepy
  • Cozy mystery stories  /  Hardboiled fiction

Adding in listen-alikes

This year we added audiobook records in NoveList Plus, so we had to make sure they were in the algorithm too.  Audiobooks recommendations match on genre, appeal, and subject just like the print titles do, and they also match on other audiobooks with a similar feel to the narration.

Improving recommendations for authors who write both adult and juvenile works

Some authors, such as James Patterson, write very differently for adult audiences than for children and teens, and we needed a way to create recommendations that stuck to the audience levels. Enter a new system for matching recommendations to the target audience. This was actually less of a change to the algorithm itself, and more of a technology solution (that's where those software engineers really make a difference!). 

So… the answer to question in the title of this post -- are people better than algorithms? At NoveList, we don't choose one or the other. People are the key to everything we do -- whether it's handwritten recommendations or an algorithm. We work continuously on making better recommendations for you and your readers.

So, keep an eye out for our improved recommendations, and let us know how we're doing! We have some other ideas up our sleeves that we'll be working on in the coming months. 


Amie Reno is a Senior Cataloger at NoveList.