Watson Analytics is an IBM attempt to make Watson cognitive technology to work with Big Data while being available for the masses. As it stands now, it’s a collection of different tools that help users to understand their data and predict what may happen in the future. The focus is on business applications, but of course it depends on the actual data. There is some natural language processing functionality, which users can use to ask (via typing) some insights about the data, but majority of the features use traditional web user interface.
Some of those tools are well designed and implemented, others needs more work.
One of the most interesting features is ability to predict or detect dependencies within the data source.
what drives Sold Price?.
The thing that seems to missing is ability to understand non-linear data, something that spans beyond a single data source. One of the selling points of Watson technology that won Jeopardy, is that it can get information from multiple sources and find a way to connect not obviosly related facts.
In order to make it happen, human should define a context of how different datasets connect to each other and Watson can be a data scientist assistant who can do big data crunching, while humans can be in the control and understand how it connects to reality.
Ok, we need some kind of framework that can help us with this task.
Fuzzy-logic Cognitive Map
One of the approaches could be Fuzzy-logic Cognitive Mapping, which allows to define mapping of how different variables (or concepts) influence each other.
Fuzzy-logic Cognitive mapping operates on the level of concepts though, but what I'm talking about is applying idea to real datasets (down to numbers) and with help of humans and Watson try to create a model that can explain dependency between different variables/datasets.
So in a nutshell it could be a high level modeler can help with seeing the bigger picture or finding what is missing. It really makes sense because in many cases the situation is not linear, as such it cannot be expressed with single dataset.
The actual relations, presented as positive (+), negative (-) or netural can be nonlinear, expressed with some kind of algorithm, formula, fuzzy logic etc.., Watson can be trained to detect what is the best thing that represent the relationship. And with this system it’s possible to connect it to the real numbers. The big question is, how to decouple influence of multiple factors and predict what will be the influence of the single connection? The answer to this question is not simple, but maybe cognitive systems like Watson can learn to solve this problem. It does not need to understand what it actually means, only apply some learned techniques to known facts (expressed as symbols or numbers).
It may sound like a big task to know where to start. That’s how experts in the domain can help by providing their understanding at first, by defining the mental model. The actual result will be different models with assigned predictive strength (as Watson Analytics does it) which can be verified by the experts.
Another visual tool can be useful to understand traditional
Analytics data is the flow diagram. It's like page visits flow, but applied to any type of data that represents the series of events.
Known application is a hospital patience flow (IBM product), but this is just the first thing that comes to mind. It could be a customers flow, sales pipeline, company incidents, technological events etc...
Based on specific situation, the canvas can facilitate thinking process and aggregate data in a single place/dashboard.
Watson Analytics provides a feature to
Assemble data mined facts in a configurable dashboard.
But it makes sense also to go other way, from the template (or empty canvas) which will helps users to start thinking about the questions that needs to be answered. It's just an example, of course canvases must be adapted to be used with data insights, not just some abstract ideas.