Corpus Construction and NER Model Comparison for Formation Name Recognition in Completion Reports

Completion reports are central to the oil and gas exploration and development process, in which formation names serve as fundamental entities for geological modeling and reserve calculation. However, general-purpose Chinese named entity recognition (NER) tools do not include formation names as an entity type, and the petroleum domain lacks Chinese corpus annotation at the sequence labeling level. Based on the World Oil Outlook 2026 (WOO 2026) published by OPEC, this study employs AntConc frequency analysis and KH Coder co-occurrence network analysis to design data-driven annotation rules, and completes independent annotation and adjudication review through Doccano. The results show that the fully-featured conditional random field (CRF) model combined with rule-based post-processing achieves an F1 score significantly superior to the rule-based baseline. This provides foundational resources and methodological references for Chinese information extraction in the petroleum domain.