I have a query that works fairly well for scoring matches in human name and address fields. However, some documents get an undesirably high score. For example, when I search for the string "CHR",
DONALD MCCLURE at the address #41 IROQUOIS/CHICKASAW
gets a higher score than CHRIS LE at the address 2021 FARMINGTON LAKES DR APT
I believe the issue is TF/IDF --> Elasticsearch may be assigning the less relevant document a higher score because the ngram CH is uncommon in the address field.
I am using both ngram and edge_ngram analyzers. I include reproducible code and more details in this stack overflow post... https://stackoverflow.com/questions/62032693/weird-relevance-ranking-in-elasticsearch
Cheers!