Data mining is a technique that tries to sort through massive amounts of data to extract useful information. It is often used by business intelligence, financial analysis and marketing organizations. 数据挖掘研究院
But now some are saying that data mining is not being used properly and that often people are wasting their time crunching numbers trying to find relationships between things that just don’t exist. Among the critics, the most notable is Wharton School Marketing Professor Peter S. Fader. 数据挖掘研究院
Fader is highly regarded. For example, in mid-July Peter Fader was awarded the 2007 Robert B. Clarke Outstanding Educator Award to recognize an academic’s overall achievement in direct/interactive marketing. So when he says something negative about today’s way of operating, people listen.
In the area of marketing and retail sales, a data mining technique called the Market Basket Analysis is frequently used. For example, from sales data kept at a clothing store, data mining could identify customers that prefer silk shirts to cotton ones. Think Amazon — they’ve taken data mining to the next level with their product recommendations. 数据挖掘研究院
In an interview with CIO Insight magazine, Fader commented, “Data mining is good at saying if something will happen or not, but it’s not particularly good at saying when things will happen… when you want to make forecasts about what particular customers are likely to do in the future, not just which brands they’re likely to buy next, you need different sets of tools.”
Fader says that marketers often collect 600+ parameters of data about each customer, but for some questions they will never be able to collect enough data. Many actions occur purely by chance and there is no way to know the reason or motivating cause for why something occurs.
Fader argues that a better approach is to try to incorporate more uses of probability models into the analysis, much more like the way that actuaries work at insurance companies. He says that data mining applied as a way to achieve personal one-to-one marketing will never work unless the company can establish very deep relationships with each customer. And for most B2C type sales, no company has the resources to cultivate such close relationships.
By applying probability instead of data mining, marketers can create much better models to capture information about timing (how long until something happens), counting (how many arrivals), and choice (given an opportunity to do something, how many people will decide to do it).
Fader comments that to answer some questions, modeling by Excel can produce better results than sophisticated data mining models can. 数据挖掘实验室
