Abstract:

In the globalizing world, the importance of knowing a foreign language is increasing. One of the important methods to reduce language teaching difficulties is personalized learning approaches, which have become easier to manage with the developments in the world of technology. Thanks to personalized learning, educational methods and materials can be presented based on the interests and needs of each individual, even in the same classroom environment. It is known that presenting the learning contents in accordance with the interests of the learners will increase the efficiency of learning. Thus, in this study, the individuals’ areas of interest were estimated through machine learning methods in order to use it as an infrastructure for personalized foreign language learning. A questionnaire was designed and applied to 164 people from different sectors. In the questionnaire, there were 11 questions consisting of options that people could choose as much as they want, and a section where they could choose their area of interest. Individuals could choose areas of interest as much as they want from technology, health, business life, different cultures, sports and fine arts, at least one of which is compulsory. The collected data were transformed into mathematical form and analyzed with k-nearest neighbor, random forest and artificial neural network methods. The Bayesian optimization method, which produces better results in a shorter time than the traditional grid search method, was used for parameter optimization of methods used. Since a user could select more than one area of interest, all machine learning models were built with a multi-label prediction approach. In this context, separate estimations were made for 6 areas of interest for each person, whether they were interested or not, and the success rate was calculated considering this situation. When the analysis results were examined, the best success rate was obtained with the random forest algorithm with 78.12%. It was foreseen that this result would be sufficient for the system to be designed and better results would be obtained with the increasing number of data.

Keywords: