Reproduce “HierLLM: Hierarchical Large Language Model for Question Recommendation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuxuan Liu, Haipeng Liu, Ting Long
Jilin University researchers developed HierLLM, a hierarchical framework leveraging fine-tuned Llama2-7B for personalized question recommendation. The system consistently surpassed baselines in learning effect on public datasets, demonstrating effectiveness in cold start scenarios and with extensive question banks.
Key insights
- The hierarchical design effectively addresses the large question set problem by reducing the decision space in stages, improving recommendation efficiency and precision.
- Integrating and fine-tuning LLMs (specifically Llama2-7B) provides robust reasoning and background knowledge, mitigating the cold start problem by enabling effective recommendations even with limited history.
- An auxiliary proficiency prediction loss, combined with policy gradients, helps the model learn effective student learning state representations from historical interactions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.