The NoteLLM framework proposes a unified approach to address I2I note recommendation tasks by utilizing LLMs. It introduces Note Compression Prompt, Generative-Contrastive Learning (GCL), and Collaborative Supervised Fine-Tuning (CSFT) to improve note embeddings and enhance recommendation systems. Extensive experiments demonstrate the effectiveness of NoteLLM in improving user engagement and recommendation accuracy.
Existing online methods typically input whole note content into BERT-based models for similarity assessment, overlooking the potential of hashtags/categories. The introduction of LLMs in I2I recommendations shows promise in enhancing system performance. The framework combines GCL and CSFT to generate hashtags/categories effectively while improving note embeddings for better recommendations.
Key contributions include addressing I2I recommendation tasks with LLMs, proposing a multi-task framework for learning I2I recommendations and hashtag/category generation, and validating the effectiveness through real scenarios on Xiaohongshu platform.
إلى لغة أخرى
من محتوى المصدر
arxiv.org
الرؤى الأساسية المستخلصة من
by Chao Zhang,S... في arxiv.org 03-05-2024
https://arxiv.org/pdf/2403.01744.pdfاستفسارات أعمق