AI TrackerFrontier desk · AEST
Back to news
normalaitech

Author Representation Strategies for Zero-Shot Authorship Attribution: A Comparative Study of LLM-Based and Embedding-Based Approaches

arXiv cs.CL · 3 Oct 2026, 02:16 AEST

A study investigates author representation strategies for zero-shot authorship attribution, focusing on various approaches including LLM-generated descriptions and style embeddings. The findings highlight the importance of author representation, with a proposed two-stage framework achieving the best performance.

Key facts

Read aloud

A study investigates author representation strategies for zero-shot authorship attribution, focusing on various approaches including L L M-generated descriptions and style embeddings. The findings highlight the importance of author representation, with a proposed two-stage framework achieving the best performance. Authorship Attribution AA requires capturing fine-grained stylistic characteristics, making it particularly challenging in zero-shot ZS settings where no task-specific supervision is available. We investigate the effect of author representations on ZS AA by evaluating a label-only prompting baseline together with three author representation strategies: representative writing samples, L L M-generated descriptions, and style embeddings LISA. The results show that label-only ZS AA is ineffective, while incorporating author-specific representations consistently improves attribution performance. Among the evaluated approaches, the proposed two-stage LISA framework achieves the strongest overall performance, whereas L L M-generated style descriptions provide a substantially more compact representation of author style at the cost of some attribution performance.

Sources in this cluster