Gregory's Server@amin How does the algorithm handle multiple words? We tried wurlitzer synth design as a test, and the first search result that wasn't on page one of the synth search or the design search was all the way down at number thirteen. That seems low to us - combining search terms to find something at the intersection is a pretty standard approach.
|
Top-level
 5 comments
@amin *nods* That would pretty well explain it - the oddity of the single result that popped up from the depths to the first page made us wonder if something else was happening. Thanks for the response! @amin our leading concern reading the methodology is that white reactionaries with money to burn on disseminating their propaganda will be overrepresented in search results, which isn't great - there are a lot of alt-right news sites with no ads no idea how to address that, though |
@Packbat
Currently it does combine tokens into keywords; two maximum at the moment but I will probably expand on that. The current thing I'm dealing with is that it's not great at the fallback if it doesn't find that specific combination, which I'm guessing is the situation here.
To demonstrate, there are no results for "wurlitzer": https://clew.se/search?q=wurlitzer
And the "suggestions" path which is filled in using keywords from the database has nothing for "synth design", meaning that specific combination doesn't appear in the index at this time: https://clew.se/suggestions?q=synth+design
It's a pretty small index, only 30k words. Enough for the testing stage I'm at but it'll definitely need to be scaled up significantly before this is fully ready to go.
@Packbat
Currently it does combine tokens into keywords; two maximum at the moment but I will probably expand on that. The current thing I'm dealing with is that it's not great at the fallback if it doesn't find that specific combination, which I'm guessing is the situation here.
To demonstrate, there are no results for "wurlitzer": https://clew.se/search?q=wurlitzer