HANA Ontological

Opportunity:

With something as powerful, both in terms of technology stack and business reach, as HANA the intelligence should not be optional configuration…it should be implicit.

So, what I am trying to suggest is: Lets write algorithm to identify what can be predicted and then..well…predict it!

Philosophy:

Ontology is a famous term in the world of Semantic Web Technology.

Ontology, in analytic philosophy, concerns the determination whether some categories of being are fundamental and asks in what sense the items in those categories can be said to “be.”

Building on same philosophy, for a modeled view in HANA, calculation or analytical, it is possible to make sense of the dimension and measures beyond the scope of the final output structure.This ofcourse can be done manually by analyzing underlying objects, but in this context we are discussing the approach which is explicit, formal and conceptualized.

Why do we need this? There could be many applications which can be made on top of this, for instance we can keep looking into system if there is any hidden pattern based on values of derived dimensions for measures in context, derived dimension here would be the ones which are not directly exposed in the view, or may not make 100% business sense in design time.

Suggested Approach

Lets see the steps in which a machine/program can do this:

1. Semantics: Getting info of the dimensions/measures and their type. The HANA objects which can help get these are:
   1. “_SYS_BI”.”BIMC_MEASURES”
   2. “_SYS_BI”.”BIMC_DIMENSIONS”
   3. “SYS”.”VIEW_COLUMNS”
2. Building Knowledge Source: Beyond the scope of measures and dimensions exposed we can derive relationships between hidden dimensions, their values and the current measure in context. The HANA object which can help get these are:
   1. “SYS”.”OBJECT_DEPENDENCIES”

The output of 2.1 can then be ran through the 1.x and 2.1 steps recursively to get a concrete knowledge source. We need to do this until following conditions are met:

1. The underlying dependency has no measure
2. The underlying dependency does not have measure same as measure in context

After the recursion is done, the derived knowledge source can then have derived-measure, ones which are recursion stop conditions, as values of derived dimensions and the actual measure in context as the final value. Such list over a common dimension across dependencies, say time, can then be the entire knowledge source representation of business data and their relationships.

This derived knowledge source is now ready for next step of algorithm in pipeline…the predictive analytics!