Contributions to affective learning through the use of data analysis, visualizations and recommender sytems

Student modeling is an important issue in telematics learning environments, e.g. learning resources can be adapted based on the students' information. An emergent area of student modeling is the inclusion of affective information. The improvement of emotion detectors based on the students' events in different telematics learning environments is an open issue. Moreover, there is a need of proposing and evaluating new visualizations involving affective information, and proposing generic solutions for the recommendation of learning materials based on the affective information. This PhD proposes two different models for the detection of emotions in two different telematics learning environments. The first model uses a Hidden Markov Model to infer the emotions in a programming learning environment in which students should use different tools to learn how to program. The second model uses a set of rules to infer the emotions in a Massive Open Online Course platform in which students should solve exercises and watch videos. An evaluation of the first model for the detection of emotions was performed using a controlled experiment, comparing the results of the model with the students' answers regarding their emotions in different instants of times. The results showed that the model was not able to detect accurately the students' answers regarding their emotions. Other models of the literature applied in other learning environments were tested and they were not able to predict accurately the students' answers regarding their emotions. Therefore, the detection of emotions based on students' events in these types of environments might not be feasible, or the reference data of students' answers to a survey with different questions about emotions should be redefined. Moreover, this PhD proposes a set of affective-related visualizations for learning environments. Some of these visualizations only involve affective information, while others combine this affective information with other related to the students' activities with the learning platforms. Some of these visualizations were evaluated with real students and results showed a good usability, usefulness and effectiveness. Finally, this work proposes a generic framework for enabling the recommendation of learning resources based on affective information. The solution includes an Application Programming Interface for the definition of the different possible events. A specific implementation of this recommender has been developed as a plugin of the ROLE SDK platform.Programa Oficial de Doctorado en Ingeniería TelemáticaPresidente: Carlos Enrique Palau Salvador.- Secretario: Eva María Méndez Rodríguez, Eva Maria.- Vocal: Ruth Cobos Pére