Korhan Günel

Dealing with Learning Concepts via Support Vector Machines

Extracting learning concepts is one of the major problems of artificial intelligence on education. Essentially, the determination of learning concepts within an educational content has some differences as compared with keyword or technical term extraction process. However, the problem can still taught as a classification problem, notwithstanding. In this paper, we examine how to handle the extraction of learning concepts using support vector machines as a supervised learning algorithm, and we evaluate the performance of the proposed approach using f-measure.

Relation extraction among learning concepts in intelligent tutoring systems

This paper addresses the question of how to extract the relevance among the learning concepts in an intelligent tutoring system using the mathematical modeling of the search engines. To test the proposed approach, two learning domains have been selected from mathematics. For each domain, five distinct chapters have been quoted from the books written by various authors. After extracting candidate concepts, some feature of them have been determined. After feature extraction, the relationships among the concepts have been detected using context vector models, and finally, the concept maps have been automatically constructed as maximum spanning tree.

Syllable based Turkish speech recognition using Dynamic Time Warping and Multilayer Perceptron

We have implemented syllable based isolated word Turkish speech recognition systems with Dynamic Time Warping (DTW) and Multilayer Perceptron (MLP) in this study. Lpc, parcor, cepstrum and mfcc features are used for these applications on the dictionary which includes 200 words. After recording the word utterances, the onsets of syllables are determined and the syllable feature database is constructed. Using this database, the most similar syllables are decided by DTW and MLP. The recognized syllables are concatenated in order. If the constructed word is in the dictionary, the recognized word is found. According to the features, the best results are obtained by DTW with mfcc features. The recognition accuracy rates are 95.1% and 92.6% by DTW and MLP recpectively.

An optimization algorithm for determining a point heat source position in a 2D domain using a hybrid metaheuristic

Abstract In this paper, we study an inverse problem of determining the unknown heat source position of a point-wise heat source in a two-dimensional steady-state heat conduction problem governed by a linear elliptic equation with the Dirichlet boundary conditions. The problem is solved by the hybridization of particle swarm optimization and the gravitational search algorithm with a newly defined mutation operator. Some empirical studies are also performed to gauge the accuracy of the proposed approach.

A Feature Selection Application Using Particle Swarm Optimization for Learning Concept Detection

Recent developments of computational intelligence on educational technology yield concept map mining as a new research area. Concept map mining covers the extraction of learning concepts, specifying relations among them, and generating a concept map from educational contents. In this study, we focused on determining the features that characterize a learning concept extracted from an educational text as raw data. The first three features are detected by using a hybrid system of Multi Layer Perceptron (MLP) and Particle Swarm Optimization (PSO), and the performance of the applied method is gauged in the viewpoint of a typical classification problem.

Geometrical Modification of Learning Vector Quantization Method for Solving Classification Problems

In this paper, a geometrical scheme is presented to show how to overcome an encountered problem arising from the use of generalized delta learning rule within competitive learning model. It is introduced a theoretical methodology for describing the quantization of data via rotating prototype vectors on hyper-spheres. The proposed learning algorithm is tested and verified on different multidimensional datasets including a binary class dataset and two multiclass datasets from the UCI repository, and a multiclass dataset constructed by us. The proposed method is compared with some baseline learning vector quantization variants in literature for all domains. Large number of experiments verify the performance of our proposed algorithm with acceptable accuracy and macro f1 scores.