PEARL introduces a novel conversational recommendation dataset synthesized with persona- and knowledge-augmented LLM simulators. The dataset addresses limitations in existing datasets by providing more specific user preferences, expertise in the target domain, and relevant recommendations. The dataset construction process involves grouping real-world reviews to extract detailed persona and item knowledge. Experimental results show that models trained on PEARL outperform those trained on human-annotated datasets in recommendation tasks. Human evaluation also indicates that PEARL is preferred over existing datasets for its quality and utility.
Til et annet språk
fra kildeinnhold
arxiv.org
Viktige innsikter hentet fra
by Minjin Kim,M... klokken arxiv.org 03-08-2024
https://arxiv.org/pdf/2403.04460.pdfDypere Spørsmål