Argument-based machine learning (ABML) enables an interaction between a machine learning algorithm and an expert in a given domain, in order to achieve successful knowledge elicitation from the domain expert. The expert provides knowledge in a quick and efficient way by explaining only automatically chosen critical" examples. ABML can also be used as an argumentation-based teaching tool. By providing more information about the quality of the given arguments, we can improve the effectiveness of the knowledge elicitation. In our thesis, we have designed and implemented two measures for estimating the quality of arguments. Evaluation of measures (2 new, 1 existent) was done through an ABML procedure, where we learned a classification model for predicting the credit score of companies. Experiment consisted of two parts: knowledge elicitation from the teacher, and knowledge elicitation from the student. The goal of the first part was to obtain a consistent data set and introduction of advanced concepts, that describe the domain. This was done with the help of a financial expert. The second part was the tutoring session, where the student learned the intricacies of the domain and achieved comprehension of the advanced concepts, by means of using the interactive tutoring loop. While carrying out the teaching trials with the students, one measure proved to be particularly successful.
|
