Ich habe alle Videos in einem YouTube-Kanal nach Videotitel und Ähnlichkeit der Miniaturansichten gruppiert.
Ich habe Einbettungen für den Titel und das Miniaturbild jedes Videos erstellt, es dann mit UMAP auf 2D reduziert und geplottet. Ähnliche Videos landen nahe beieinander, sodass unterschiedliche Arten von Inhalten Cluster bilden.
Link ist ytmap
Von Economy_Future_6752
7 Kommentare
WTF does similarity mean?
did PewDiePie and Linus Sebastian do no nut november?
Without explanation on what we re seeing, nobody can understand this. What is „similarity“? What are the dots representing? What is the density conveying?
this is interesting. Small suggestion: the glass morphism on the search bar makes it hard to read the placeholder text
YouTube video titles are pure clickbait, so it’s most likely meaningless. You could analyze the contents and summarize for semantic similarly.
Nice and kind of unexpected.
I wonder how each of those videos „would move“ (see the Time Curves paper – essentially they drop the timestamp, take all pixels and project each frame using MDS… UMAP should work too. Lines are connected in temporal order)
What are those? Petri dishes?
With WHAT?? 😅