Modern conversational information access systems continuously extract user information from conversations to construct user profiles, which then drive personalized retrieval. However, as these profiles grow, they inevitably accumulate noise, potentially undermining the very personalization they are meant to support.
The paper, titled "RALI@iKAT 2026: Retrieval-aware User Profile Denoising via Personalized Conversational Dense Retrieval," proposes a method to address this issue. Based on the title, the approach integrates retrieval awareness into the denoising process, using personalized conversational dense retrieval to refine profiles.
The abstract is brief and does not include experimental results or implementation details, so the reported effectiveness of the method is not yet available. Still, the work highlights a practical challenge in conversational search and offers a direction for keeping user profiles accurate over time.