Generating Recommended sentences for Conflict Prediction and Resolution based on RAG
Jin Su Kim · Korean Institute of Smart Media · 2025
Recently, as team projects have become a common way of performing work, conflicts due to emotional conflicts in communication via messengers have frequently occurred. Such conflicts can weaken teamwork and have a negative impact on learning and project performance. In this study, in order to detect and prevent conflicts in team project conversations early, we generate recommended sentences for conflict resolution using the RAG(Retrieval-Augmented Generation) technique based on emotion prediction. The conversations between team members are first predicted using an emotion model, classified into positive/negative emotions, and the conflict status is predicted by integrating the semantic similarity between the conversations converted into vectors. Finally, recommended sentences that can help resolve conflicts are generated using the predicted conflict status, predicted emotions, and conversation sentences. The generated recommended sentences can be expanded to specific areas by searching not only general sentences that help resolve conflicts, but also specialized information files stored in a local database, thereby generating efficient recommended sentences for conflict resolution.