Abstract
This study is made for investigating the performance of Artificial Immune Recognition Systems on genre and author detection by using method referenced as [1] based on representation of a document in a different scheme. Most of the studies done nowadays depend on Bag of Words model which takes the roots or the stems of the words as features. This situation both increases the number of features and the classification time. In this study, the method we named as YTU is used which applies a weighting algorithm on word stems and decreases the number of features to the number of classes resulting in lower classification time and better performance. Artificial Immune Recognition algorithms AIRS1, AIRS2, AIRS2Parallel which we tested by this method increased the performance in genre and author detection. In the experimental results section of the paper the comparison of the classification performance of mostly used classifiers on author and genre detection Naive Bayes (NB), Support Vector Machine (SVM), Random Forest (RF), KNearest Neighbourhood (K-NN) and the Artificial Immune Systems algorithms are presented. Especially in genre detection AIRS2Parallel classifier gives the highest performance of 99,6% with Random Forest and K-Nearest Neighbourhood. This shows that Artificial Immune Recognition algorithms can be used in genre detection.