Personalized software tutoring and affective human-computer interaction in social networking-based language and open learning
KeywordsWeb-based tutoring ; Intelligent tutoring systems ; Tutoring of foreign languages ; e-Learning ; Computer-Supported Collaborative Learning (CSCL) ; Intelligent Computer-Assisted Language Learning (ICALL)
This Ph.D. Dissertation presents a novel approach of web-based tutoring, offering personalization to students’ needs. The implemented intelligent tutoring system, called POLYGLOT, incorporates social media characteristics in the user interface of the learning environment. These features include posting on a wall, tagging a classmate, instant and asynchronous text messaging, reaction buttons (liking and disliking) on questions and declaring the affective state. Also, POLYGLOT offers an authoring tool to the instructors in order to change the learning content and observe students’ performance. Given that POLYGLOT’s learning content concerns the tutoring of foreign languages, namely English and French grammatical concepts, it uses the Stephen Krashen's Theory of Second Language Acquisition, consisting of five hypotheses: the Acquisition-Learning hypothesis, the Monitor hypothesis, the Input hypothesis, the Natural Order hypothesis and the Affective Filter hypothesis. As such, POLYGLOT’s tutoring coincides fully with the aforementioned theory in terms of the way of instruction, means of collaboration, time constraints in learning, holding students’ records, logical gradation of learning concepts and response on negative affective state (frustration) in the form of motivational messages. To the direction of individualized instruction, POLYGLOT’s student model automatically detects the learning style of students. The students’ learning styles are based on the Felder and Silverman model and POLYGLOT classifies students as active or reflective, and sequential or global. Active learners prefer to communicate with their peers and to learn by working with a classmate so that they can discuss about the taught material. In contrast, reflective learners prefer to work alone. Sequential learners prefer to learn progressively and incrementally, having a linear tutoring progress. On the other side, global learners prefer to navigate through the learning material from chapter to chapter randomly. The automatic detection of students’ learning style is conducted by a supervised machine learning algorithm, namely the k-nearest neighbors algorithm, which takes as input several students’ features, such as their age, gender, educational level, computer knowledge level, number of languages spoken and their grade on preliminary test. Furthermore, the presented student model incorporates an error detection and diagnosis mechanism which combined two algorithmic techniques into a hybrid approach in order to infer the reason of students’ misconceptions. The first technique is the approximate string matching which finds approximate substring matching a pattern and diagnoses misconceptions such as accidental slips, pronoun mistakes, spelling mistakes and verb tense mistakes. The second technique is the string meaning similarity which diagnoses misconceptions owing to language transfer interference. Moreover, POLYGLOT employs a model for collaboration between students. This model recommends win-win collaboration between students. The recommendation for collaboration concerns two situations. In the first situation, the recommendation for collaboration concerns two students having complementary knowledge, namely student 1 has a high knowledge level on concept A but poor knowledge level on concept B and student 2 has a high knowledge level on concept B but poor knowledge level on concept A. In the second situation, student 1 conducts misconception A but not B while student 2 conducts misconception B but not A. This rationale can enhance students in the learning process and ameliorate the degrees of knowledge acquisition and knowledge restitution. In POLYGLOT, students can declare their affective state among “happy”, “frustrated” and “neutral”. However, their interaction with the tutoring system, i.e. experiencing difficulty in a test or receiving a bad grade, can be a blockage of their goal and as such the reason of feeling a negative emotion, such as frustration. POLYGLOT can detect students’ frustration by using the linear regression model. The relationships are modeled using linear predictor functions whose unknown model parameters are estimated from the data. Finally, the POLYGLOT’s response on frustration is the delivery of motivational messages based on the attribution theory, involving a three-stage process underlying that behavior must be observed/perceived, must be determined to be intentional and is attributed to internal or external causes. With the use of motivational messages, the students are assisted in the educational process and are not willing to quit learning. All the aforementioned approaches are fully implemented and POLYGLOT is evaluated. The system was used by students of a private school of foreign languages in Athens in order to learn the grammatical concepts in both foreign languages. For the evaluation of all the modules of POLYGLOT, the Kirkpatrick's Four-Level Evaluation Model was used. The results of the evaluation were very encouraging. They demonstrated that the system effectively adapts the learning process to the students’ learning style while assisting them by diagnosing their misconceptions, recommending win-win collaborations, detecting their frustration and responding to this negative emotion.