When I enter the three character search term POS, which in a linguistic community would be understood as “part of speech” Discourse takes a three letter prefix and returns all sorts of inappropriate topics. If I search here for POS i get topics and replies with works post, posts, position, possible, etc. or BOX the results include boxes, If I search amongst the LLVM community for PUBLICATIONS I get a raft of posts with example C/C++ source code that contain the word public. This prefix or lemma level search is highly frustrating and prevents the discovery of relevant material..
AI-generated summary
A user, glimfeather, reported issues with Discourse’s search feature, specifically when using acronyms as prefixes. The search results include irrelevant topics due to the prefix or lemma level search. glimfeather finds this feature frustrating and considers it a bug, as it prevents the discovery of relevant material.
Moin suggests that the feature category may not be the best fit for this issue. Moin also provides a workaround by using quotes around the search term, such as "pos ", to get more accurate results.
However, glimfeather questions why this feature is not user-accessible, suggesting that the “advanced” search page should have an option to enable or disable stemming, making non-stemming the default. This would allow users to choose how they want to search and avoid having to remember specific syntax.
If this is a feature — as its move to that category suggests — then it is a bad one and becomes useless/
Why does Contribute > Feature not fit?
Having a look at the category description
It is not only about existing features but also about enhancements.
Maybe. However, as someone who worked on text retrieval systems for decades, this “feature” should be consider a bug. As it currently stands the search feature is practically useless for anything but the most trivial of searches.
You can get rid of all the post and postition results by searching for " pos ".
https://meta.discourse.org/search?q=%22%20pos%20%22
2 Likes
Okay so it works but why isn’t it a user accessible feature? At least on the “advanced” search page where users could enable or disable stemming — making non-stemming the default would be less confusing — at their choice rather than having to remember arcane syntax. And " publication " produces different results from “publication” — includes publications — but maybe we are supposed to be satisfied that stemming to public doesn’t happen then.
1 Like