In this year’s datalift summit, we hosted a workshop on building a dashboard using FastAPI and streamlit with which you can implement your own recommendation engine for interesting newsletters. The result looked something like this:
We used Kern refinery to build labeled data for our dashboard, as we wanted to predict the topics of each story of a newsletter. To do so, we labeled ~100 records by hand, and then applied some labeling functions and active learning models as heuristics. The labeling functions mostly consisted of so-called distant supervisors, which are functions that look up terms and patterns of a given lookup list.
You can find the repository on GitHub, where we provide the code and concepts of the workshop. It was a lot of fun!