ResearchBenchmarksRelation Extraction on PG-Fiction full (test)Relation Extraction on PG-Fiction full (test)Follow76Positive-Class Micro-F1Best SLM30.65642.42854.265.972Jun 21, 2026Evaluation resultsMethodMethodLinksPositive-Class Micro-F1Best SLMPrompting=schema-enume...Prompting=schema-enumerated, Selection=strongest tuned SLM per benchmark2026.0676RoB-baseBackbone=RoBERTa, Para...Backbone=RoBERTa, Parameters=125M, Inference mode=fine-tuned per benchmark2026.0668.6ClaudeInference mode=zero-shotInference mode=zero-shot2026.0633.4GPT-5.4Inference mode=zero-shotInference mode=zero-shot2026.0632.4