Journal of Intelligent & Fuzzy Systems 25 (2013) 673–683 DOI:10.3233/IFS-120674 IOS Press 673 A neuro-fuzzy immune inspired classifier for task-oriented texts Maryam Tayefeh Mahmoudi a,b , Fattaneh Taghiyareh a,∗ and Babak N. Araabi c a School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran b Knowledge Management & e-Organizations Group, IT Research Faculty, Research Institute for ICT, Tehran, Iran c Control & Intelligent Processing Centre of Excellence, School of ECE, College of Engineering, University of Tehran, Tehran, Iran Abstract. Increasing growth of task-oriented texts specifically in organizations, have become a catastrophe nowadays. To overcome this problem, potential classification methods are improved. This paper outlines the capability of neuro-fuzzy approach and artificial immune recognition systems to enhance task-oriented texts classification. Task-oriented texts stand for various kinds of texts which are organized to help the users with their different tasks such as: research, development, learning, justification, innovation and analysis. In this respect, seven major attributes with three nominal values of low, medium and high are considered to classify text into six task classes. To illustrate the capabilities of proposed approaches, Takagi-Sugeno as a neuro-fuzzy approach using lolimot learning algorithm, is compared with multilayer perceptron (MLP), and Radial Basis Function (RBF). In the meantime, various versions of Artificial Immune Recognition Systems (AIRS) including AIRS1, AIRS2, Parallel AIRS and Modified AIRS with Fuzzy K-Nearest neighbor (Fuzzy-KNN) are also evaluated in comparison with the above mentioned algorithms to classify the same text. The experimental results of classification on a dataset of 540 data reveals that, due to the distributed characteristics of a text, and the complexity of tasks respectively, evolutionary and neuro-fuzzy methods are expected to be particularly workable and successfully applicable to task-oriented text classification specifically for the purpose of decision support. Keywords: Artificial immune recognition system, neuro-fuzzy approach, task-oriented text, text classification. 1. Introduction Automatic text classification has become one of the prime research issues during recent years. Many algorithms are invented to investigate that. Some are focused on common text mining algorithms such as [25]: k-nearest neighbor (KNN), Na¨ ıve Bayes, Deci- sion Tree, etc. some are improved by combining the features or classification results such as: boosting [24], owa operator and decision template [7, 29], Majority voting [25], etc. The others are evolved from evolution- ary algorithms such as: Neural networks [32], Genetic ∗ Corresponding author. Fattaneh Taghiyareh, School of Electri- cal and Computer Engineering, College of Eng., University of Tehran, Tehran, Iran. Tel.: +982182084181; Fax: +982188352148; E-mail: ftaghiyar@ut.ac.ir. Programming [26], particle swarm optimization [33], artificial immune recognition systems [5], etc. The increase of complex tasks and functionalities from one side and distributed ness of text and its features from the other side, leads the classification algorithms into hybrid and evolutionary approaches, which may have the potential to handle large dimensioned feature space and in the meantime the inherent complexity and uncertainty within the prospected functionalities of a text. Due to the significance of comprehensive texts in making efficient decisions in different virtual and real areas such as organizations, learning and research insti- tutes, the playing task of texts can be appropriate criteria to classification. Moreover, text’s tasks may exhibit itself in an aggregation of a variety of considerations in 1064-1246/13/$27.50 © 2013 – IOS Press and the authors. All rights reserved