A Knowledge Discovery from Full-Text Document Collections Using Clustering and Interpretable Genetic-Fuzzy Systems

Abstract

The paper presents a concept of a hybrid system consisting of two our original techniques from the computational intelligence area and its application to knowledge discovery from full-text document collection. Our first technique - self-organizing neural network with one dimensional neighborhood and dynamically evolving topological structure - aims at automatically determining the number of groups in the document collection and at grouping the documents in terms of their similarity. In turn, the main goal of our second approach - multi-objective evolutionary designing technique of fuzzy rule-based classifiers with optimized accuracy-interpretability trade-off - is to extract the most important keywords from documents and to generate classification rules which can be helpful in understanding and isolating the subjects of documents collected in the founded groups. The proposed concept may also be useful to develop systems operating in a wide area of human language understanding problems.