You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When doing QA on databases, the right answer is often everything connected via a certain property. E.g. when asking "Who are the children of Marge Simpson?", we want to return a list of all entities connected via a particular property, even if individual evidence might prefer some to the others.
In general, thiss hould get a lot better when Honza Pichl's improved property selection is finished. But we still should have a general procedure for this. Proposal:
In phase2 (out of pipeline phases 0, 1, 2 - the final one), aggregate probability for each property to carry answer (aggregate features of answers with this property -> classifier)
-> answer feature, as well as a "group-by" operation that'll return "superanswers" with the answers chunked by source property
Prerequisite: List-aware gold standard, fully move to JSON-based datasets
The text was updated successfully, but these errors were encountered:
The relevant line of the error is the last one. The dbpedia provided by the ailao guys wasn't available at the time. Sometimes it's just a temporary failure and the db will be back in seconds but sometimes the db is really down since as far as I know it (not member of the team, no connection to them) they change their architecture quite often, but they keep the repo updated with the newest endpoints.
The system is prepared to handle this, so as you can see, it will run the request again in a couple seconds. If the system can't solve the problen itself in a couple of seconds, you should check if you can reach their dbpedia instance ping <db url>. You also can set up a dbpedia instance and set it up in the config (there is docs for that)
Also please open a new issue if you have any problems, to make issue management easier. 🙂
When doing QA on databases, the right answer is often everything connected via a certain property. E.g. when asking "Who are the children of Marge Simpson?", we want to return a list of all entities connected via a particular property, even if individual evidence might prefer some to the others.
In general, thiss hould get a lot better when Honza Pichl's improved property selection is finished. But we still should have a general procedure for this. Proposal:
Prerequisite: List-aware gold standard, fully move to JSON-based datasets
The text was updated successfully, but these errors were encountered: