Automated responses driven by language models are being effectively proposed to facilitate communication in crisis situations. However, one of the critical challenges faced is maintaining a consistent response style, which plays a vital role in building trust among affected individuals. The proposed study introduces a novel metric focused on style consistency and employs a two-stage fusion-based generation approach to enhance response quality and stylistic uniformity.
By initially assessing the style of candidate responses, the proposed method optimizes and integrates these responses at the instance level. This is achieved through a fusion process that significantly reduces stylistic variation, ensuring that responses are not only accurate but also reliable in tone and presentation. The results indicate that this approach provides substantial improvements over existing baseline methods in both response quality and consistency, suggesting important implications for the future of crisis communication.
👉 Pročitaj original: arXiv AI Papers
