I am looking for software and/or papers on automatic data insight generation. I think it is more nuanced than AutoML --- or I misunderstood it as an umbrella term.

By data insights, I loosely mean any result that is possibly interesting; examples include category outliers; correlation; anomalies, overall trends, and seasonality in time series; associations; predictability; and so on. I am interested in how these are mined automatically:

  • Given the combinatorial explosion in the number of possible insights to be generated, how do we choose where to concentrate our efforts on?
  • How do we evaluate the "interestingness" (or "confidence") of the data insights we have found?
  • The only software I know that generates such insights automatically is Microsoft Power BI[0]. Narrative Science is another, but it seems much more advanced --- generating natural language reports using ontologies --- than what I am interested in.

    P.S. Also asked on StackExchange: https://datascience.stackexchange.com/questions/85417/papers-on-automatic-data-insight-generation

    [0]: https://docs.microsoft.com/en-us/power-bi/consumer/end-user-insight-types

    [1]: https://narrativescience.com/lexio/how-it-works

    Similar questions and discussions