Researchers have formalized the task of assigning incoming papers to a user-specific hierarchy without per-user training as personalized hierarchical paper routing (PHPR). They propose PaperRouter-Agent, a training-free LLM agent that grounds routing decisions in folder members rather than folder names alone.
- The agent narrows the candidate hierarchy and retrieves folder-specific evidence.
- It verifies fit by inspecting member papers and incorporates similarity-gated feedback from past user rejections.
- On real personal libraries, it raises Recall@1 from 0.39 to 0.61 and Recall@3 from 0.57 to 0.83.
- The largest gains occur on organizational folders defined by metadata like venue or year, where single-shot methods collapse (Recall@1 0.09 to 0.50).
- On the LaMP-2 benchmark, accuracy improves from 44.5% to 51.5%, yielding a +9.0 macro-F1 over a single-shot baseline.
This approach provides low-cost practical use for organizing personal libraries by accurately handling evolving and private folder structures.