Get Rid of Bad Analytics with Semantic Analysis


Nicole Laskowski of Search CIO reports, "Traditional data analytics has three rungs: descriptive, predictive and prescriptive analytics. But Scott Mongeau doesn't believe the ladder goes far enough -- especially as businesses reach to make better decisions based on big data analytics and data science. A business analytics consultant for Deloitte Nederland, Mongeau believes both the rigorous testing of models (what he calls diagnostics) and a better way to describe the meaning and context of projects to computers (what he calls semantic analysis) need to be built into the three rungs if data analytics is going to help solve big problems. He argued that without the rigorous testing of analytical models and semantic analysis -- and a social network in which to share the theories behind analytical models -- big data analytics will be susceptible to bias and, in the end, fail to help people make better decisions."


Laskowski continues, "That's because data science -- and science in general -- is hard, said Mongeau in a TEDxtalk for Rotterdam School of Management. Humans are predisposed to make bad decisions with data -- even when they aren't doing the computing. 'As the economic Nobel Prize winning, experimental psychologist Daniel Kahneman puts it: We were built to win, not necessarily to be right,' he said. So feeding data into a model for analysis doesn't erase the potential for bias and misinterpretation. That's because models themselves are biased -- they are all just a representation of reality, he said. And people can engage in 'over fitting' or placing too much significance on the results from a sample; or they mistake correlation for causation; or they have a false sense of confidence in 'the big data approach' -- namely that having so much data and so many variables cancels other factors out."


Read more here.


Image: Courtesy Flickr/ Search Engine People Blog