Re:Amazon “suicide kits” have led to teen deaths
by haswell on Sunday October 09, 2022 @01:19PM
(#33143231)
I was referring to the product pairings, not the search result autocompletion (which does seem like a pretty glaring problem).
Cutting off a search early is easy. Deciding that “even though these are commonly purchased together, this pairing shouldn’t be shown” seems like a fundamentally different problem.
Re:Amazon “suicide kits” have led to teen deaths
by indymike on Sunday October 09, 2022 @06:43PM
(#33146036)
Dropping results is easy to do, too. Figuring out what shouldn't be paired? That's a curation problem.
Re:Amazon “suicide kits” have led to teen deaths
by haswell on Sunday October 09, 2022 @08:50PM
(#33146752)
Agreed, and this is why I compared it to the social media curation problem.
The thing I was replying to was the sentiment that this should have been solved decades ago.
Curation/moderation is never solved, only managed.
No. Unexpected outcomes are pretty normal for search engines (and really, any complex computer system) since... forever, and really, the solution, regardless of input is is really easy: on input if (keyword in prohibited_phrases) disable recommendations, and on output if (keyword in recommendation_results) disable recommendations to user. I do believe this was solved sometime prior to E. F. Codd's 1970 "Relational Completeness of Data Base Sublanguages".