Re: Stereotypes and fuzzy logic in user modeling

Wed, 26 Jun 1996 12:46:02 +0200 (Hans Zimmermann) wrote:


>can anyone give me some comments on the following problem concerning the use
>of stereotypes together with fuzzy logic in user modeling?

>In fuzzy logic the defuzzification of a membership function leads to a crisp
>value for a property of an object (for example a new value for a valve as
>part of a technical process).

>I would like to adopt this method for the "defuzzification" of stereotypes in
>user modeling. In this context a stereotype is a collection of typical
>properties of a user. If we use fuzzy logic in the trigger rules that
>control the ascription of stereotypes to specific users, we get membership
>functions for each ascribed stereotype. So far this is equal to regular fuzzy
>logic rules.
>If we try to determine the properties of a user Ben, who is supposed to be a
>"UNIX-expert", we have to perform a kind of defuzzification. We can do this
>in two ways: either we state, that Ben has all properties contained in the
>stereotype to a restricted extent or we give up some of these properties
>and claim that he has the rest of the properties to the full extent.
>A combined point of view is possible, as well.
>In my opinion the second alternative mixes up imprecision and uncertainty as
>the membership of a property to a stereotype is always uncertain. So the
>first alternative would be eventually the clearer one. But if we consider
>our own intuitive way to form conclusions out of a statement like "Ben is a
>70% UNIX-Expert" we probably come to a result, that uses both alternatives
>mentioned. So, what is the better solution? Does anybody have a meaning about
>the theoretical aspects and the solution of the problem?


>Hans Zimmermann.

>Sender: Hans Zimmermann
>Address (Office): Burlafingerstrasse 4, 89233 Neu-Ulm, Germany
>Phone (Office): +49-7308/919096
>Fax (Office): +49-7308/919097


It soulds like a problem for a fuzzy SQL database in which "Ben" would
be compared to a "best match solution" along with other candidates,
and then using fuzzy inferencing, he would be ranked along with the
other candidates as to how well they meet the search criteria. Unless
he exactly matched the properties of the individual fuzzy sets in the
search engine however, his rating would always be between 0 and 1.

I hope this helps in answering your question