Reproduce “Prompt-based Personality Profiling: Reinforcement Learning for Relevance Filtering” 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
Jan Hofmann, Cornelia Sindermann, Roman Klinger
Researchers introduce RL-Profiler, a reinforcement learning framework that filters relevant textual instances from lengthy user profiles to enhance prompt-based personality profiling by large language models. This approach reduces LLM processing time by over 76% on real-world data while achieving comparable macro F1 scores and demonstrates a 29 percentage point increase in macro F1 on artificially balanced and enriched datasets.
Key insights
- Reinforcement learning can effectively induce a relevance filter for LLM input by optimizing for downstream task performance without requiring direct human relevance annotations.
- Pre-training the RL agent with pseudo-labels derived from Normalized Pointwise Mutual Information (NPMI) significantly enhances training stability and performance.
- Reducing the input context for LLMs through intelligent filtering leads to substantial gains in computational efficiency and inference speed, making author profiling more practical.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.