A CLASS QUESTION

A CLASS QUESTION

By W H Inmon

In our class on the Internet on practical text analytics we have people ask questions by email. Occasionally a question is so interesting that it deserves a detailed answer.

HI Bill - Could you please explain in lay terms the difference between machine learning & textual analytics ? How are they different in doing things?

Regards,

Akila Seetharaman

Business Analyst

There are two different interpretations of machine learning.

The first interpretation of machine learning is the one where computers are taught to make good decisions by processing data. The more data the computer processes, the more the computer learns. And the more the computer learns, the smarter the computer is, and the better decisions that it can make. At least in theory that is the way that machine learning is supposed to work. Machine learning applies to many disciplines. One of those many disciplines is text analytics. The problem with text analytics is that there are many rules the computer has to learn. Many, many rules. Machine learning for text and language is a gargantuan, almost never ending task.

Consider what you had to learn when you were a child and you were learning to speak. When you were a child learning to speak, you learned many rules. First you learned the alphabet. Then you learned words. Then you learned pronunciation. Then you learned how to spell words. Then you learned the meaning of words. Then you learned how to put words together. Then you learned how to form sentences. Then paragraphs. Then bodies of thought and expression.

In order to learn a language you had to learn a lot of rules, and those rules are stored in your brain. Every time you read or speak, those rules are there, in the background, waiting for you and waiting to govern your speech or writing. And there were many exceptions and anomalies in these rules. In the best of cases, language is quite complicated. Language is just very complex. For machine learning to even start to approximate what is required to learn to speak a language, there has to be a LOT of machine learning to take place. In truth, machine learning only addresses a small subset of what is needed in order to start to fully grapple with text and language.

Text analytics is one of only many fields of endeavor where machine learning may be used.

The second interpretation (the cynical interpretation) of machine learning is that it is a new buzzword that vendors use to make more sales. Vendors have found that they can sell buzzwords and make a lot of money from those buzzwords. Vendors don’t even need a product. All they need are Powerpoint slides and a few mysterious phrases that only a select few people understand.

Vendors invent a new buzzword that makes the customer think that the vendor knows something that they don’t know. There always has to be an element of the unknown about a new buzzword. Then vendors approach people with this unknown quantity and entice people to buy into it. In many cases, the vendor doesn’t know anything more than the customer. But the newly invented buzzword makes the customer think that the vendor does. People buy what they don’t know on the hopes that it will lead to something good. So when the latest buzzword has sold all it can sell, the vendor invents a new buzzword and repeats the process over again.

In many cases the new buzzword is merely a repackaging of something that has been around a long time.

After machine learning has run its course, there will be a new buzzword that the vendors will invent.

In the case of machine learning, when some of the vendors found they could no longer sell the buzzword “AI” they decided to invent a new buzzword. And that new buzzword was machine learning.

Those then are two different interpretations as to what is meant by machine learning and text analytics.

Bill Inmon wrote HEARING THE VOICE OF THE CUSTOMER, Technics Publications and TURNING TEXT INTO GOLD, Technics Publications and DATA ARCHITECTURE: SECOND EDITION, Elsevier Publications. In addition Bill teaches a class on PRACTICAL TEXT ANALYTICS, which is free and is taught over the Internet. For more information about the class, contact Bill at [email protected].