The USC Andrew and Erna Viterbi School of Engineering USC Signal and Image Processing Institute USC Ming Hsieh Department of Electrical Engineering University of Southern California

Technical Report USC-SIPI-305

“The Fuzzy Logic Advisor: a Paradigm for Social Judgments”

by Jerry M. Mendel, Matthew Martin, Sheila T. Murphy, Lynn C. Miller and Nilesh Karnik

November 1996

The goal of this study has been to demonstrate how fuzzy logic systems may be applied to enhance our predictability of social judgments above and beyond that provided using traditional linear models. To illustrate the utility of this fuzzy-logic approach, we have focused on one specific type of social judgment, judgments of flirtation.

A fuzzy logic system is a nonlinear device that accepts numeric measurements as inputs and provides a numeric output (e.g., a flirtation level). Its distinguishing feature is that its nonlinear nature is established using concepts from fuzzy sets and fuzzy logic. In engineering applications, the fuzzy-logic system's input and output numbers must be available through direct measurement; however, this is not the case for the social judgments that we are interested in. A person does not measure flirtation directly, nor does a person measure the indicators of flirtation directly, e.g., eye contact and touching.

We have therefore distinguished between quantities that can be measured directly and those that can only be "sensed." Flirtation and its indicators can only be sensed (e.g., "you'll know it when you see it"); hence, there is a mismatch between a social judgment and its indicators and the use of a fuzzy logic system to make a social judgment. We rectify this mismatch by combining the fuzzy logic system with an encoder that converts sensed indicators of flirtation into numbers, and a decoder that converts a numerical value for flirtation into a meaningful linguistic judgment for flirtation. The interconnection of encoder, fuzzy logic system, and decoder is what we call a "Fuzzy Logic Advisor (FLA)."

We have provided qualitative and quantitative descriptions of the FLA, so that ultimately anyone can easily program the relatively small number of equations that describe it. We have also provided a methodology that represents a complete procedure for designing the FLA, one that can be applied to many social judgments, such as flirtation and sexual harassment. The 9 steps of this methodology are: (1) Identify a social judgment (e.g., flirtation, sexual harassment, etc.) that cannot be measured directly, but can be 'sensed' to different degrees; (2) Determine the 'indicators' of the social judgment; (3) Establish meaningful scales on which a person can easily convert a sensed value for each indicator, as well as the social judgment, into a number in their encoder; (4) Establish fuzzy sets for each of the indicators as well as the social judgment; (5) Establish membership functions for the fuzzy sets; (6) Establish rules between the indicators and the social judgment; (7) Transfer the rules and membership function information into a mathematical fuzzy logic advisor; (8) Establish a vocabulary for the user of the FL Advisor; and, (9) Map the vocabulary into quantitative measures of uncertainty for rule antecedents. Surveys play a central role in the design of a FLA. They are used to implement Steps 2, 4, 5, and 6 of this 9 step procedure.

We have demonstrated that FL Flirtation Advisors are at least an order of magnitude better than a standard regression model, i.e., their (consensus) mean-squared errors are at least an order of magnitude smaller than the four regression models we looked at. Hence, we have demonstrated that fuzzy logic systems may indeed be applied to enhance our predictability of social judgments above and beyond that provided using traditional linear models, which was our goal.

We have also demonstrated how the FLA can be used as a diagnostic training tool. First it can be used to diagnose problem areas; then it can be used to sensitize an individual to same-sex or opposite-sex flirtation patterns. In this way, people ought to be able to improve their social acumen regarding norms associated with fliration -- and how one's perceptions depart from normative ones.

Finally, we wish to reiterate the fact that in this report we only consider the social judgment of flirtation; but, we believe that the basic methodology that is described in the context of flirtation can also be applied to many other social judgments (e.g., sexual harassment).

To download the report in PDF format click here: USC-SIPI-305.pdf (28.0Mb)