Role-Based Personalized Dialogue Summarization

This paper proposes a role-based personalized dialogue summarization method (RoPeD) aimed at improving the quality of summaries in multi-role dialogue scenarios. We integrate the BART model and introduce a role personalization strategy to capture multi-role interactions in dialogues, generating summaries with better coherence, structure, and semantic consistency. Experiments were conducted on the CSDS and MC Chinese datasets, and the results demonstrate that RoPeD outperforms traditional methods across multiple evaluation metrics, including ROUGE, BLEU, and BERTScore, significantly improving summary quality. In particular, RoPeD shows superior performance in handling semantic consistency and multi-role interactions, outperforming conventional methods. The experimental results validate the effectiveness of role personalization in multi-role dialogue summarization and provide an innovative solution for future applications in more complex dialogue scenarios.