In the context of computer-mediated collaborative learning environments, the effectiveness of learner interactions remains a central concern. This thesis explores the factors influencing collaboration and the sense of social presence in collaborative serious games, with a particular focus on the impact of virtual environment characteristics on interaction dynamics.

Through two experimental studies, we examined how perspective (first-person vs. third-person) and communication modalities (text, audio, video) shape the collaborative experience. The first study revealed that the first-person perspective enhances the sense of psycho-behavioral accessibility and encourages information sharing among players. The second study demonstrated that the audio modality fosters a higher sense of social presence and better co-regulation compared to text and video modalities.

To deepen our understanding of collaborative dynamics, we developed an analytical approach combining interaction trace analysis and content analysis of exchanges. The successive application of frequency analysis, co-occurrence network analysis, and Markov chain modeling allowed us to identify specific patterns in the structuring and sequencing of collaborative activities. This approach was extended using machine learning techniques to automatically classify utterances and enable systematic comparisons between communication modalities.

Our findings highlight the importance of carefully considering virtual environment characteristics when designing collaborative learning experiences. They emphasize the central role of social presence as a mediator between environmental properties and collaborative behaviors. These results open up new perspectives for the development of learning environments better suited to effectively support learner collaboration.