Number of unique k-mers

[BrinthaVP]

Are there any suggestions for choosing the threshold for unique k-mers for amplicon sequencing reads?

I have a set of V3V4 reads and tried assigning taxonomies using KrakenUniq (612 identified) and Qiime+Dada2 (425 identified). Dada2 removed a lot of reads as chimeras. Though around 408 are commonly identified by both of them, there is a significant difference with respect to the the number of reads assigned. Should I apply threshold based on unique k-mers on krakenuniq output? I personally prefer Krakenuniq as it runs much faster but at the same time concerned with the number of reads due to false-positives.