Encoding words into Cloud models from interval-valued data via fuzzy statistics and membership function fitting

Abstract

When constructing the model of a word by collecting interval-valued data from a group of individuals, both interpersonal and intrapersonal uncertainties coexist. Similar to the interval type-2 fuzzy set (IT2 FS) used in the enhanced interval approach (EIA), the Cloud model characterized by only three parameters can manage both uncertainties. Thus, based on the Cloud model, this paper proposes a new representation model for a word from interval-valued data. In our proposed method, firstly, the collected data intervals are preprocessed to remove the bad ones. Secondly, the fuzzy statistical method is used to compute the histogram of the surviving intervals. Then, the generated histogram is fitted by a Gaussian curve function. Finally, the fitted results are mapped into the parameters of a Cloud model to obtain the parametric model for a word. Compared with eight or nine parameters needed by an IT2 FS, only three parameters are needed to represent a Cloud model. Therefore, we develop a much more parsimonious parametric model for a word based on the Cloud model. Generally a simpler representation model with less parameters usually means less computations and memory requirements in applications. Moreover, the comparison experiments with the recent EIA show that, our proposed method can not only obtain much thinner footprints of uncertainty (FOUs) but also capture sufficient uncertainties of words.