Student-Artificial Intelligence Conversations and Sentiments: A Process Data Analysis and Reciprocal Effects

Abstract

Research on academic outcomes has extensively explored students’ emotions, motivation, and learning behavior. While further research highlights the role of individual interactions with generative artificial intelligence (AI), a gap exists in understanding the longitudinal dynamic associations between university students' interactions with AI assistants and the sentiment expressed during these conversations over time. Based on a theoretical framework, we expected reciprocal effects between students’ interactions with AI assistants and sentiment within their conversation, considering students’ gender, age, and type of AI assistant used. Over four months, students at a university had access to three AI assistants: Alix for motivational-emotional support, Robyn for reflective discussions, and Dr. Melly for exam discussions. All AI assistants are integrated into a campus management system, ensuring compliance with data protection regulations. The sample for this process data analysis comprised n = 5,262 interactions of 422 students (56% female students, MAge = 30.42, SDAge = 9.08). Time series analysis with hierarchical Bayesian continuous time dynamic modeling revealed significant reciprocal effects between the students’ question length and the sentiment during these conversations: Students’ relatively short questions at the beginning of a conversation determined a later positive sentiment. A positive sentiment at the beginning of a conversation determined relatively short questions from the students later on, in particular from female students. Another result is that the type of AI assistance moderated the reciprocal effects. Thus, the students perceived the AI assistants as helpful tools for addressing questions during the day and nighttime in higher education.