Skip to content

Standard Skill: Query Log Mining

Query Log Mining is a Standard Skill: it finds out from a Data Source’s Query Log what people actually ask of it, where the agent’s queries fail, and which answers were rated down, so that new Context Entries come from real use. The agent runs it only when you ask it to mine, analyse or review the Query Log, never on its own. To pick it yourself, type / and choose query-log-mining.

The Right to accept Context Change Requests on the Data Source, the same Right that lets you read its Query Log in the Data Studio. Without it, the agent tells you that someone who holds it can run the Skill.

  1. It reads the Query Log of the period you name, or of the last 90 days, at most 1,000 entries, and says so if the period holds more. The agent never learns who asked: the Query Log it reads names nobody.
  2. It groups what it read into frequent questions (however they were worded, with how often), errors (by table or column and kind of error) and answers rated down (with the likely cause, such as a wrong join or a missing filter). Every number in the report is counted, not guessed.
  3. It compares them with the Semantic Layer and calls something a gap only when no entry covers it, or when the entry that covers it wasn’t followed.
  4. It reports the three lists and three to seven gaps, each with the kind of entry that would close it.

It proposes nothing itself. It offers to close the gaps you pick with Improve the Semantic Layer, which has you confirm every Verified Query’s result first.