Drets d'ús i reutilitzacióAquesta versió del manuscrit es fa disponible amb una llicència Creative Commons del tipus Atribució No Comercial No Derivades (CC BY-NC-ND) http://creativecommons.org/licenses/by-nc-nd/4.0, que permet baixar-la i compartir-la sempre que se'n citi l'autoria, però sense modificar-la ni utilitzar-la amb finalitats comercials. ConsultesSi creieu que aquest document infringeix els drets d'autor, contacteu amb l'equip de recerca:

Monitoring students in virtual learning environments can be a time-consuming task. Professors and tutors must accompany students in an agile manner. During the COVID-19 pandemic, the use of discussion forums posed new challenges. This work proposes a conversational agent to automatically detect which pedagogical intervention is necessary to guide students in MOOCs environments. Through the attributes of the students' post messages, it is possible to classify which action will be carried out by the agent, applying specific dialogue patterns. In some more specific cases, the tutor's attention is immediately requested. The proposal was evaluated through a feasibility study to verify if semantic detection can contribute to guide the intervention process. According to the results, it is possible to support the tutor, as only 35.2% of interactions required the tutor's action.

No abstract

In the original version of the chapter, the following belated correction has been incorporated: The author name "Danielle Toti" has been changed to "Daniele Toti". The chapter has been updated with the change.

Accompanying students in virtual learning environments to identify those who need help is a difficult and time-consuming task. Identifying subjective attributes that recognize students’ feelings can help teachers and tutors in pedagogical interventions. The interaction must motivate and keep students engaged. This article proposes an architecture capable of automatically carrying out some pedagogical intervention. The architecture uses automatic textual classification models to detect multiple attributes in post messages, such as Sentiment, Post Type, Urgency, and Confusion. The main goal is to predict learning problems and act to minimize their impacts. The evaluation was performed with data from the Stanford MOOCPosts Dataset to verify if the models allow the automatic classification of subjective attributes. Our results show that the proposal outperforms other approaches in this dataset.