µ2: tweakable library recommender

The focus this µsprint was to make a new tweakable recommender for danish library data. A prototype of this is now running on https://bib-recommend.veduz.com/, – with ½ million library materials.

The motivation for this is to make a recommender, where each danish library can customise it with their preference for how the recommendations should be. The next step would be to make it compatible with the API DDB-CMS uses for recommendations, and to make a webapp where librarians can customise the recommender service.

Takeaways

  • The recommendation algorithm, is fast enough to run live with tweakable data
  • leveldb mutates while only reading data
  • docker deployment/publish details+script (publish.sh)
  • ½ million recommendations is too few to be production-usable, needs fallback and/or subject-based recommender