Comment on SearNGX should be a federated search engine

• fmstrat@lemmy.nowsci.com ⁨5⁩ ⁨months⁩ ago

  This is close to what I was thinking, but rather than crawling independently, leverage the API results from queries to build a list of sites (and then perhaps crawl). Potentialy a tag index of sorts. I’m not solid on any idea as I haven’t investigated SearNGX enough to see how it works under the hood, but yes, on the same plane of thought.

  source

  • Max_P@lemmy.max-p.me ⁨5⁩ ⁨months⁩ ago

    I ran a YaCy instance for a while like a decade ago. It does federate index requests, and when you search it propagates the search request across a bunch of nodes. When my node came online it almost immediately started crawling stuff and it did get a bunch of search queries. But the network was still pretty small back then and the search results were… not great. That’s the price of independence from Google’s and Microsoft’s giant server farms, it’s hard to compete with that size.

    But at the rate Google and Bing are enshittifying, I think it’s worth revisiting.

    Using ActivityPub for this would be immensely wasteful. It’s just not feasable that all instances would have the whole index because it’s so large. Back when I tried it, the network still had several TBs worth of indexed pages. This is firmly in the realm of distributed P2P systems. One could have an ActivityPub plugin however to receive updates from social media near instantly and index those immediately with less overhead. But you still want to index wikipedia, forums, blogs, whatever the crawlers can find.

    source
  • hendrik@palaver.p3x.de ⁨5⁩ ⁨months⁩ ago

    Sure. SearX is a meta-search engine. It does (only) queries to other search engines to get results. YaCy on the other hand is itself a search engine. It has the data available and doesn't do queries to other engines. In theory you could combine the two concepts. Have a software that does both. But that requires some clever thinking. The returned (Google) ranking only applies to the exact search term. And it's questionable if you can store it and do anything useful with it except for when some other user searches for the exact same thing. And also the returned teaser texts are very short and tailored to the search query. So maybe also useless. It'd be hard.

    One thing you could do is crawl the results that users actually click on. And I think YaCy already does that. AFAIK they had an browser add-on or a proxy or something to intercept visited pages (and hence search results).

    source