All too often I get irrelevant garbage from search results in traditional search engines, main causes are:
- Keyword-based search being grossly insufficient or getting "distracted" by almost-matching or nonmatching terms in sites. For example, in a Google search for "2 fl oz sanitizer bottle" (https://www.google.com/search?q=2+fl+oz+sanitizer+bottle&ie=utf-8&oe=utf-8&client=firefox-b-1), only a third of the first-page search results are actually 2 fl oz bottles.
- Spam and low-quality sites specifically designed to manipulate search engines
- Search engines getting distracted by insufficient partial match. Google is particularly bad in this aspect, where it will often find no results for a search, then cut out the most important terms and claim it found the best partial match.
I'd like a search engine centered around user-feedback that learns along with the user. It could require some form of account login, then every search will provide feedback options alongside every link. Most important are negative feedback: if the user isn't happy with search results, the site should give the user massive latitude to explain why the search results are trash and how they could be improved.
Example feedback options could include:
- "This search result is low-quality spam"
- "You mistakenly equated or associated a particular search term with something irrelevant".
- "A particular search term means something different per context."
- "The search was inaccurate (produced something close yet irrelevant)."
- (Teach the search engine that certain terms should be associated or dis-associated).
All feedback would be per-account to prevent spambots from spamming large numbers of false association/disassociation feedbacks and manipulating search for other people.
This program could be a website or a separate program that scans through search results from popular search engines, filters them, and stores learning metadata on disk.
