-
-
Notifications
You must be signed in to change notification settings - Fork 58
Description
This may be related to #988, but it specifically concerns the snippets extracted. These are supposed to show the most relevant text, with the search term highlighted. However, in many cases, we get quite random text selection and no highlighted search terms. For example, in Full English Wikipedia searching for Kasos Massacre, although the top result is correct, the snippet shown seems completely random:
It would have been much better just to show the lede of the article in this case:
The Kasos massacre was the massacre of Greek civilians during the Greek War of Independence by Ottoman forces after the Greek Christian population rebelled against the Ottoman Empire.
I realize this is low-level, but there may be something wrong with the setup or retrieval parameters given that the vectors to the relevant piece of full text appear so very wrong in many cases. It would almost be better just to extract the lede instead and show it if none of the search terms are actually in the snippet, though that would be a kludgy workaround.