Counting people, then weighting them

After taking a break, I wanted to revisit the ADRs from the build and show a screenshot for each one.

The first is episode-weighted people. At first counting seemed straightforward: we counted the connections between two shows. But once the ingest pulled full per-episode credits, counting stopped working. Most cast entries were for a single episode, so two shows looked alike because they had cycled through the same one-episode guests. So I weighted each person by how much of a show they actually made.

ER’s top ten similar shows, ranked by counting shared people and then by weighting them

The cost is that weighting needs data the catalog does not always have. A person’s share is their episode count divided by the show’s total, so a show with no per-episode credits scores zero against everything. Not low. Zero. It vanishes from the graph instead of ranking badly in it, and covering that gap is why the fallback ladder exists.

For a more in-depth explanation, read this ADR.

← The 80-Day Project