Wrong algorithm to compute similarity between words
I'm surprised that some request yield videos with absolutely no relationship with request. For instance, using the acronym "dns" returns videos like https://videos.domainepublic.net/videos/watch/34e3401a-e9d5-4e1a-b968-a50e81b66059 where "dns" never appears.
I suspect that the algorithm uses similarity, not identity, and therefore "des" is regarded as similar to "dns" which is clearly a bug (they only have 66.6 % of the same letters)
Another example, "bgp" returns https://video.ploud.fr/videos/watch/b716b943-84b0-4ed4-a923-f094f9474092 which never mentions "bgp" but "bop".