Re-reading my post from last month, Thoughts On Automated Recommendation Services for Libraries, I realize that I’m somewhat wrong about the role of locality in automated recommendation systems. Amazon and Netflix can (and I think they do) integrate user location data in their recommendation algorithm. They can skew results based on strong trends that appear among proximate user groups.
I’m also somewhat wrong about how recommendations work in BiblioCommons—ratings and reviews for individual titles are aggregated from multiple libraries, as well as user-created reading lists. Reviews from local library staff are prioritized over others, but I don’t know if local library user-created reading lists are prioritized. Regardless, these are individually curated pieces of content, these sections aren’t automated in the same way that Netflix is. This content is related to an individual item in the catalog and doesn’t generate lists of recommendations as one typically expects from Readers’ Advisory services.
The “Similar Titles” type of content in the sidebar in BiblioCommons is what I think of when I look for reading recommendations, and this content is also the most directly analogous to Netflix / Amazon. However, this content doesn’t get generated by BiblioCommons at all—these titles come from third party services like NoveList.
So locality can be a meaningful factor in automated recommendation systems—such systems are smart enough to recognize that local trends are important, even if they aren’t yet intelligent enough to know what local trends mean.
But libraries still don’t have enough data to make such algorithms work. We still rely on curation and the personal touch to create real value for our patrons.
John the Librarian 