Yesterday we did not work on anything related to TVLens. We made up for it today.
We did a lot more work on search on Day 22, so I thought it would be a good day to bolster the foundation while building a bit faster. My idea is that it may leave time in the last two or three weeks to do some high-level quality control on the codebase. Let’s jump into it.
Backlog cleanup
I had to decide whether four issues would stay in this “round” or move to a kanban board I keep for issues to consider. It is not a backlog, because items on that board auto-close after a period of time.
- Per Season on the Show Page. In the initial design of TVLens, I wanted to slice the TV show data down to the individual season. Currently the data model supports entire shows, but having the recommender take in the season’s ratings alongside the show’s ratings is a decision that needs its own analysis. For this round I do not want to spend the time on that, so we pushed it to a future issue.
- Actor Pages. Right now the only individual pages are show pages. TV show fans sometimes want to find shows with specific actors, though. The focus this “round” is the recommender, so actor pages will come soon.
Homepage assessment
During a recent test of the website, I realized that if I rated a show and then left that show’s page, I could not find related shows again. The homepage focused on Top Picks, Recently Added, and Side Quests. So I decided Side Quests would be a fun way to shuffle the recommendation section I am adding to the homepage. That way users can rate shows and still find their next show from the homepage.
Cast and crew preferences
We made this more dynamic than the previous build. In short, TVLens now weights cast or crew connections according to what a user tends to rate highly. For example, if I rate shows whose casts connect more strongly than their crews, recommendations swing toward cast rather than crew, and vice versa.
Remap Layer 1
We made significant changes to recommendation scoring, service jobs, and ranking, and we rewrote some of the tests that validate the codebase against the data in our local version of the site. We also now exclude service jobs such as Animal Wrangler, which matter less to a TV show watcher.