Open michelole opened 5 years ago
But Rocchio is always interactive, isn't it? I know that there is also pseudo relevance feedback which would be fully automated. What are your thoughts regarding this?
Rocchio is interactive, but there's no need to more than two rounds: we can play the user and judge the top-5 docs (as we already do when creating the internal GS).
PRF is also interesting, but might be a problem for topics with very few relevant documents.
Try the classical Rocchio algorithm for relevance feedback.
Consider https://github.com/biocaddie/elasticsearch-queryexpansion-plugin
Follow-up of https://github.com/bst-mug/trec-2017-precision-medicine/issues/87.