Precisely. If I had to hazard a guess, I think their algorithm is hyper-optimizing for linkages, and when linkages are broad but tenuous, it hyper-optimizes for the ones with only the slightest dominance.
Hypothetically: all sorts of people are interested in the history of World War II... it's a big war, it defined politics for over half a century, and it's taught every year in the history classes of at least a hundred nations. So you're looking at a topic with hundreds of "also likes" connections but no strong winner.
Perhaps there is a 0.000001% chance that if you watch a WWII video, your next video will be one on modern Nazism (including recruitment videos). I wonder if YouTube's algorithm boosts that small fluctuation in a sea of noise into a single recommendation result (or even grabs the top 5, but the signal is so noisy that this one still shows up in that top 5)?
Hypothetically: all sorts of people are interested in the history of World War II... it's a big war, it defined politics for over half a century, and it's taught every year in the history classes of at least a hundred nations. So you're looking at a topic with hundreds of "also likes" connections but no strong winner.
Perhaps there is a 0.000001% chance that if you watch a WWII video, your next video will be one on modern Nazism (including recruitment videos). I wonder if YouTube's algorithm boosts that small fluctuation in a sea of noise into a single recommendation result (or even grabs the top 5, but the signal is so noisy that this one still shows up in that top 5)?