A big problem was that for years RDF was The Way to do semantic data, but it completely sucks shit at even basic stuff like encoding an order list of items. It’s an incredibly frustrating format to work with.
Aside from overpromise/underdeliver, I think an easy answer is advertising and lock-in to ad-driven sites became the norm. There’s too much money coming in to switch to something that might cut into it. Described well in this article. Ironically, it has an auto-playing, video ad pop up as I tried to skim it to make sure it was correct article.
Yeah, the rise of Facebook / Twitter / LinkedIn / Yelp / Redfin walled-gardens of information definitely created barriers to a shared semantic web. It was also very labor-intensive to create and maintain metadata of any value. Ontology is hard! At the time it was proposed, the semantic web was a cart without a horse. The advocates wanted content producers to annotate everything with rich metadata, but where were the killer semantic apps?
Of course, in the mean time Google was creating innovative new ways to collect and search data without relying on hand-crafted semantics. Who wants to write RDF when machine learning algorithms can infer “good enough” semantics from the content itself?
I have tried to discover RDP, the semantic web and triple stores many times in the past.
The popularity of the semantic web and RDP preceded modern JS. In fact, I believe modern JS was a reaction to it. Less complex, more “getting things done”. I don’t have a chronology though.