Precision tuning does not affect prefixing

Hello dear community,

While doing search through AppSearch, we notice some aggressive matching patterns that end up highlighting irrelevant words on the search results. The relevant parts of the results are still correctly highlighted and the scoring is alright, but the irrelevant words being highlighted introduces doubt regarding the precision of the search engine from users.

Here's an example in the image (query tester on Kibana, we get the same results through the search API): Basically, the "de" is being prefix-matched with a lot of other words. This case is particularly problematic because "de" is a stop-word in french and is being matched with a whole lot of words that by times may contradict the query itself and introduce a lot of confusion (for instance, the query includes "september" and some results highlight "december")

So it's basically an issue of prefixing. We attempted to address this through the precision tuning feature and it doesn't seem to have an effect on that (the example in the image is actually with a precision of 11). It appears as if the precision parameter isn't having any effect on the results.

Is precision tuning unrelated to highlighting? is it language-dependant? what are other ways to reduce this fuzziness?

As noted in the documentation, synonyms aren't being used.

All the stack is on the most recent release (7.16.3)

Thanks for the detailed post, Meave!

This indeed looks like a bug.

Although the plans might change, we added it to our near-term roadmap to look into.

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.