gabrielmagno commented on PR #1245: URL: https://github.com/apache/solr/pull/1245#issuecomment-1370930774
Just to bring a "data-science KNN user" view on this. I think there is actually _two_ related but different ways one could use the Dense Vectors and KNN features. The first one is the most obvious and straightforward: you use it as KNN per si, i.e. you bring the top K documents most similar to the target. The second way to use the KNN is to use the _similarity score_ that is calculated with the vectors. In this case we use the vectors and their similarity for _ranking_ rather than for _retrieval_. And there are some use cases that we could even combine multiple similarity scores to create an aggregated score, or even combine the similarity score with the actual "lexical score" (BM25). For this second use case we often use a very high K to guarantee that we calculate the similarity score for _all_ the relevant documents. So, considering this second use case of using the KNN for calculating the similarity score, having the ability to filter the documents based on the value of the similarity is very useful. And as previously mentioned, sometimes this is not even directly a single similarity score, it could be a combination of multiple scores, and we could apply the Post Filter on top of the aggregated score. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@solr.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@solr.apache.org For additional commands, e-mail: issues-h...@solr.apache.org
