5
$\begingroup$

For my Computational Intelligence class, I'm working on classifying short text. One of the papers that I've found makes a lot of use of granular computing, but I'm struggling to find a decent explanation of what exactly it is.

From what I can gather from the paper, it sounds to me like granular computing is very similar to fuzzy sets. So, what exactly is the difference. I'm asking about rough sets as well, because I'm curious about them and how they relate to fuzzy sets. If at all.

Edit: Here is the paper I'm referencing.

$\endgroup$
4
$\begingroup$

"Granularity" refers to the resolution of the variables under analysis. If you are analyzing height of people, you could use course-grained variables that have only a few possible values -- e.g. "above-average, average, below-average" -- or a fine-grained variable, with many or an infinite number of values -- e.g. integer values or real number values.

A measure is "fuzzy" if the distinction between alternative values is not crisp. In the course-grained variable for height, a "crisp" measure would mean that any given individual could only be assigned one value -- e.g. a tall-ish person is either "above-average", or "average". In contrast, a "fuzzy" measure allows for degrees of membership for each value, with "membership" taking values from 0 to 1.0. Thus, a tall-ish person could be a value of "0.5 above-average", "0.5 average", "0.0 below-average".

Finally, a measure is "rough" when two values are given: upper and lower bounds as an estimate of the "crisp" measure. In our example of a tall-ish person, the rough measure would be {UPPER = above-average, LOWER = average}.

Why use granular, fuzzy, or rough measures at all, you might ask? Why not measure everything in nice, precise real numbers? Because many real-world phenomena don't have a good, reliable intrinsic measure and measurement procedure that results in a real number. If you ask married couples to rate the quality of their marriage on a scale from 1 to 10, or 1.00 to 10.00, they might give you a number (or range of numbers), but how reliable are those reports? Using a course-grained measure (e.g. "happy", "neutral/mixed", "unhappy"), or fuzzy measure, or rough measure can be more reliable and more credible in your analysis. Generally, it's much better to use rough/crude measures well than to use precise/fine-grained measures poorly.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.