Attributes based on natural language descriptions (Kansei words) are common in affective design questionnaire data. Such words are usually inherently vague and exhibit characteristics such as explicitly borderline cases and blurred boundaries between those cases to which the word does and those to which it does not apply. In this paper we propose an integrated treatment of vagueness and uncertainty which combines three value logic and probability by defining a probability distribution over valuations in Kleene’s logic. Such an approach naturally results in lower and upper uncertainty measures on the sentences of the language, quantifying the uncertainty that a given sentence is true or that it is not false respectively. Within this framework we propose a representational model for opinions in the form of a graph of conjunctive clauses ordered by precision and weighted according to their respective lower and upper uncertainty measures. Furthermore, by extending the idea of scoring functions to a three valued setting we propose an approach for ranking different designs which takes into account both the level of belief in an opinion and also its relative strength. The potential of this approach is illustrated using a case study involving questionnaire data about Kutani traditional Japanese craft designs.
|Conference||6th International Symposium on Integrated Uncertainty in Knowledge Modelling and Decision Making (UIKM 2019)|
|Period||15/03/2018 → 17/03/2018|
|Series||Lecture Notes in Computer Science|