This idea has been merged into another idea. To comment or vote on this idea, please visit WDS-I-18 Passages and relevancy training.
Relevancy training for whole documents is one thing, however allowing for relevancy training on passages would in my view greatly increase the power of WDS. We find a lot of use cases where the correct document is returned, but we need WDS to return the most relevant passage from that document, but the accuracy for passage retrieval (defined as, if the returned passage is what would be expected from a manual human search of the doc) is less satisfactory. This feature would greatly assist with handling long tail questions for chat bots, and scaling bots with the WA + WDS design pattern. I would be interested to know how feasible this feature is.
Why is it useful?
|Who would benefit from this IDEA?||As Developer i want to be able to improve / train the relevancy of passages returned in an NLQ search so that I can increase the accuracy of results to user queries|
How should it work?