Answers are grounded in this paper and cite their sources.
Domain lexicons are essential for identifying specialized concepts in text, but seed-based construction can miss emerging terminology. This study proposes Recursive Lexical Combination (RLC), a corpus-driven method that recursively expands contiguous Chinese lexical units while preserving word order and textual boundaries and terminates low-support paths using corpus support. Using national-level data asset policy texts, the method was configured on a validation set and evaluated on an independent held-out test set against seed-based lexicon, C-value, and direct n-gram baselines. RLC reduced candidate types from 35,436 to 15,341, achieving 56.71% candidate compression while retaining 530 gold-standard multi-unit document-term types with 80.55% candidate-pool recall. It also covered 474 of 602 true terms not found by the seed lexicon. Compression increased with term length and was accompanied by lower coverage of longer terms. After expert semantic validation, micro-averaged F1 increased from 0.2166 for automatic RLC to 0.4811. RLC therefore provides an efficient and traceable workflow for Chinese domain lexicon construction by separating search-space control from semantic validation and making the trade-off between candidate compression and term coverage explicit.
No takes yet. Share an insight, caveat, or question.
Lyu et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: