1. What is “ObjectAnalytics”?

Real world data typically is “complex” – not just hard to make sense of content-wise, also technically, meaningful data is usually complex structured. If your company has gathered valuable data and spent effort to organize those data well, then typically it is not a flat table, rather hundreds of tables from different sources, additional tables with meta data and catalogs…

….and this complexity soon clashed into what today’s analytical technologies can digest.

The term “ObjectAnalytics” describes a novel paradigm which allows to analyze complex objects statistically.

An example object is “The Patient”. It has master data and lots of dependent data such as time series of events along the lifetime of the patient. You can easily imagine the complexity of an electronic health record. You will find diagnoses, prescriptions, treatments, information about hospital cases, each case in itself a complex structure… and all this complexity is far beyond what a statistician can easily analyze, and also beyond what business intelligence technologies can process in one comprehensive view.

Millions of patient records can certainly be managed well in a relational database … transactionally! But once you want to draw meaningful conclusions on a larger number of patients, you need to have all data for each single patient together in one place – instead of one patient scattered across hundreds of tables. That’s why developers prefer to work with “objects”; in terms of its member fields, sub-objects and arrays of sub-objects each single object instance holds just everything about, e.g., an individual patient collected in one place in an easy to navigate data structure.

An Xplain Data server may hold millions of such object instances – i.e. millions of concrete patients with their health history. The Xplain Data server exposes analytical capabilities on such a collection of objects. You can define analytical operations on those objects, for example longitudinally analyze an event stream or different streams relative to each other (e.g., analyze prescriptions relative to diagnoses). There are diverse methods available to do “analytics on objects”, e.g., you may define a “relative time axis” or “rank aggregations” to analyze the flow of events.

Once defined, you can execute that analytics on millions of object instances (individual patients) in a second to collect the required statistics. You are no longer iterating rows in a table (that is what classical algorithms do – which painfully forces you to map your complex world onto a flat table). You are iterating whole objects, millions of them with billions of attached events, thereby executing the required analytically operations on each object instance.

There are diverse technical interfaces available, primarily a Web-based interface. Frontends to the Xplain Data backend are usually Web-based frontends. One such client is the Xplain ObjectExplorer – our Web-based frontend which you can use to explore Object Views. We often use the acronym XOE for it. The XOE implements an interactive usability concept which lets you explore complex objects statistically. You can define new artifacts on an object, initiate queries and view results (just by click and drag-and-drop in the browser). As an analyst, the Xplain ObjectExplorer will be your primary tool working with ObjectAnalytics.

The XOE communicates with the backend via the Web-based interface. Developers may use this interface to develop their own frontends, e.g. with domain-specific visualizations and presentations of results. An Xplain Data ObjectAnalytics Database is the best choice of technology in case your analytical demand requires access to a comprehensive object “as a whole”.

Viewing things as a whole is in particular important once you want to understand cause and effect. According to a definition of D. A. Kenny (“Correlation and Causality”, 1979), direct causal factors cannot be “explained away” by other factors. To detect potentially causal factors, we therefore have to search for as many other factors (“cofounding factors”) as we can, or – in other words – we need to view the object in focus of analysis with all its related data. Basically, ObjectAnalytics has been invented as a key to Causal Discovery.