Soft computing as a basis for perceptual information processing: Application of text categorization

Intelligent information processing tasks such as summarization and categorization are performed based on criteria driven by users' perceptions. However, conventional information processing systems determine such criteria in a static manner and their results are often different from what users expect, forcing users to adapt to these systems (instead of the other way around). The reason for this kind of behavior lies on the rigid scheme and underlying computational paradigm on which such systems are based.; Soft computing, an emerging computational paradigm consisting of technologies such as fuzzy sets, probabilistic algorithms, neural networks, and genetic algorithms, arises as antithesis to reasoning hard computing and aims to capture and emulate the nature of human beings including tolerance for imprecision and granulation of information. One of the main features of soft computing is its adaptive behavior so that systems adapt to users' perceptions. Soft Computing has been successfully applied in continuous domains such as image recognition, signal processing, and market forecasting. Fuzzy sets play a core role in soft computing by providing an appropriate representation of uncertainty, granulation, and bias of perception.; In this thesis a simple intelligent information processing method based on users' perception, named Perceptual Information Processing (PIP), is proposed. This generic method has an adaptive and feedback behavior consisting of the following three tasks: conceptualization to form initial perceptions, recognition to perform a required intelligent task, and refinement to adopt users' perception by comparing results of the system and the user. Fuzzy sets are used for representation of perception, and a soft computing method called Mass Assignment Theory (MAT) is used as the computational core. MAT preserves the consistency between data sets and perceptions in a natural way based on what a human being recognizes. For illustration purposes, an application to text categorization based on word frequencies is studied.; This study aims to provide an implementation of Computing with Words and its use towards a computational approach to perceptions as proposed by L. A. Zadeh.