Maybe this question is too general but..
I have a dataset of 30k+ documents which is loosely based on the gov2 dataset (I think it may be a subset).
I am struggling to improve the precision@10 metric, which is how this task will be evaluated.
I have tried a custom analyzer that does some English possessive stemming and some other light stemming with kstem.
I have also combined a must with a match_phrase query..
Any thoughts on how I can change my ES config to increase precision on this information retrieval task?